Andrii-zvorygin Peak-oil Meeting Transcript — October 20, 2026

00:00:00 SPEAKER_UNKNOWN: After peak energy, when energy tightens, who owns the useful intelligence?

00:00:06 SPEAKER_UNKNOWN: Okay, and you guys are probably familiar with this.

00:00:09 SPEAKER_001: Jean Lejarré was nice enough to put this together earlier, and it's got all of the oils and the fossil fuels by weight.

00:00:18 SPEAKER_001: And so what it shows is that we're past peak on all of our fossil fuels, other than NGPL, which is this little one at the bottom down here.

00:00:29 SPEAKER_001: And so when energy gets tighter, every sector seeks efficiency.

00:00:37 SPEAKER_001: There's a picture of a municipal crew planning a root growing, planning a harvest, and the clerk scheduling maintenance.

00:00:44 SPEAKER_001: Each person needs the same useful result with less fuel, power, material, time, and waste.

00:00:49 SPEAKER_001: Coordinated software can search records, compare options, route work, schedule the next step.

00:00:54 SPEAKER_001: People still choose the plan and do the physical work.

00:00:57 SPEAKER_001: The gain is more useful work from each scarce input, which leads naturally to the cooperation communities used to keep essential services moving.

00:01:05 SPEAKER_001: And so, we saw that historically, pressure creates new levels of cooperation.

00:01:13 SPEAKER_001: And the three communities faced material pressure in a world powered by sunlight, food, wood, and animal labour.

00:01:20 SPEAKER_001: Climate shifts changed water and biomass, declining tree cover and scarcer fuel reduced availability.

00:01:27 SPEAKER_001: Mesopotamian towns organised rivers, irrigation, and fuel.

00:01:30 SPEAKER_001: Indus settlements coordinated around changing water systems, and Centaur communities joined metalworking, horses, chariots, and long-distance exchange during the Indusarian expansion.

00:01:40 SPEAKER_001: Shared work became infrastructure.

00:01:41 SPEAKER_001: In each, the later question is whether computing can become another shared capability close to the work.

00:01:48 SPEAKER_001: So we're looking through like these people.

00:01:51 SPEAKER_001: It was the climate changed and it dried up, so they had less wood and less agriculture.

00:01:57 SPEAKER_001: And for us, it's the fossil fuels that are drying up, and so we have less energy.

00:02:01 SPEAKER_001: And so, if we look at the computing investment over since 2019, you can see it's been rising steadily at 11.8 percent a year, roughly, or 10 to 11 percent.

00:02:15 SPEAKER_001: And now, but if we look at the share of GDP, the graph gets more interesting.

00:02:24 SPEAKER_001: And so we had the pandemic, and then everyone was at home, and then buying up various computers so that they could do remote work, and so we had this kind of peak,

00:02:38 SPEAKER_001: and then you know they had enough, and this improved productivity, because people had computers at home, they were able to produce more productivity,

00:02:48 SPEAKER_001: and that led to a dip in terms of the share of how much computing was of the total GDP, and then you can see the similar kind of dip just after ChatGPT launches.

00:03:02 SPEAKER_001: So ChatGPT launched, and it created a massive productivity boost in the GDP of the U.S. at least, which is what we're looking at here.

00:03:13 SPEAKER_001: And so the share declined.

00:03:16 SPEAKER_001: So if you look behind, see this is where it was released, and it stays flat.

00:03:20 SPEAKER_001: It stays flat, but the actual GDP grew a lot during that time, and then you can see it goes back up here,

00:03:29 SPEAKER_001: and we had more or less the same kind of ChatGPT interface until quarter four of 2025, when we got agentic coders,

00:03:42 SPEAKER_001: and you see that led to another big boost in productivity that we've been living in.

00:03:50 SPEAKER_001: And so what people call AI is really modern software.

00:03:55 SPEAKER_001: That modern software has several parts.

00:03:58 SPEAKER_001: A model is a compressed database of patterns learned from text, code, images, video, speech, and other records.

00:04:05 SPEAKER_001: Its weights store relationships used to generate tokens.

00:04:10 SPEAKER_001: Different models use different methods.

00:04:12 SPEAKER_001: Language models predict tokens.

00:04:14 SPEAKER_001: Diffusion models denoise images or video, and vision, speech, music, and coding systems have their own forms.

00:04:20 SPEAKER_001: Tools connect a model to current web pages, local files, databases, compilers, and tests.

00:04:26 SPEAKER_001: The task prompts retrieval records and tool results form short-term working memory.

00:04:31 SPEAKER_001: Saved records, tests, and later training preserve institutional knowledge.

00:04:36 SPEAKER_001: So you have the database, and it generates in different ways, and the working context lets it have a short-term memory while it's figuring things out.

00:04:46 SPEAKER_001: Okay.

00:04:47 SPEAKER_001: And then if anyone has a question, just unmute because I can't see you and like ask the question.

00:04:51 SPEAKER_001: Okay.

00:04:52 SPEAKER_001: Open models climb the GPQA ladder, and so there's this test.

00:05:01 SPEAKER_001: It's called the GPQA.

00:05:03 SPEAKER_001: It asks difficult science and mathematics questions that compares answers with people who have graduate-level subject training.

00:05:10 SPEAKER_001: So that means like masters, PhD people.

00:05:14 SPEAKER_001: And Qwen 3.5 9B reaches that human reference line, so that's over here, and it was in 2026.

00:05:21 SPEAKER_001: So for since 2022, when ChatGPT was released, a lot of these were not smart, but they were smarter than your average person, which scores about 22 percent on the GPQA test.

00:05:35 SPEAKER_001: And now a graduate expert in a particular domain will score about 81.3 percent, and so our open models achieve that in 2026, and you see we've got a lot of them.

00:05:49 SPEAKER_001: The last few points require much larger models.

00:05:51 SPEAKER_001: So, whereas say Qwen 3.5 9B, that means it has nine billion parameters.

00:05:59 SPEAKER_001: It is 81.7% accurate, and an expert is 81.3.

00:06:04 SPEAKER_001: So this is about like human level.

00:06:06 SPEAKER_001: And so these other ones are superhuman in terms of 89% and stuff like that, but they require more parameters, as you can see here.

00:06:15 SPEAKER_001: See these ones: thirty, hundred, thirty, you know, a lot more.

00:06:18 SPEAKER_001: And you can see the percentages on the side here.

00:06:21 SPEAKER_001: And then omniscience, knowledge, and restraint.

00:06:25 SPEAKER_001: So in terms of hallucinations, because this is an issue that comes up, and so we have humans.

00:06:33 SPEAKER_001: I mean, this is an honest human, I guess.

00:06:36 SPEAKER_001: There was the truthful QA, and so humans are prone to giving false information by accident about six percent of the time, whereas our current models are still much higher.

00:06:49 SPEAKER_001: So the Qwen 3.5 9B I was talking about earlier is like 85 percent, very very high.

00:06:54 SPEAKER_001: But some of the newer models, like 27B, it's closer to 18 percent, which is still much lower than even the frontier models like Claude Opus and GPT-4o.

00:07:09 SPEAKER_001: Here, so I'm just trying to see.

00:07:14 SPEAKER_001: Okay, everybody's still following along.

00:07:16 SPEAKER_001: Okay, and so we actually there's three ways to check a piece of work because you don't want to if you have an automated worker you want to make sure you have good results.

00:07:33 SPEAKER_001: And so we looked at humans and say you have a human supervisor that checks over your work.

00:07:39 SPEAKER_001: Well, a human supervisor is, on average, only catches about 46 percent of the errors, whereas even these older models would catch 92.7 percent of the errors.

00:07:52 SPEAKER_001: And this is Claude 3.

00:07:53 SPEAKER_001: This is from before Q4 2025.

00:07:58 SPEAKER_001: And so, really, the best chain is to have a hybrid chain where you have the automated workflow, and then you have a human.

00:08:05 SPEAKER_001: managing exceptions or major changes.

00:08:09 SPEAKER_001: And so another thing is that useful model performance has dropped a lot since it was back in 2020.

00:08:18 SPEAKER_001: It's like 850 times lower than before.

00:08:21 SPEAKER_001: And here it says local bill is mostly hardware.

00:08:25 SPEAKER_001: So because the larger the model you have, the more hardware you need.

00:08:29 SPEAKER_001: However, also software has been improving.

00:08:32 SPEAKER_001: So the Qwen FlashNext coder even though it's a much larger model, can be run on the same hardware as the Qwen 3.8 27B.

00:08:41 SPEAKER_001: Even though this has about 108 billion parameters.

00:08:46 SPEAKER_001: And so here we're looking at the cost of intelligence.

00:08:50 SPEAKER_001: And so in comparison to humans, so if you have a one million tokens, now a million tokens is would take several weeks for a human to generate, so a token is about four letters,

00:09:06 SPEAKER_001: give or take.

00:09:08 SPEAKER_001: So like one point two five words.

00:09:10 SPEAKER_001: I mean, so a word is one point two five tokens rather.

00:09:15 SPEAKER_001: And so, if we were to pay a human to generate a million tokens at GTA living wage, it would be thousand eight hundred and forty seven dollars.

00:09:27 SPEAKER_001: Whereas the same amount of output from Qwen 3.8 27B, which is smarter, would cost thirty five cents.

00:09:38 SPEAKER_001: And so HelpOS.ca is one of the examples that well I've made, and so it has local inputs based on council meetings, and agendas, PDFs, and things like that, and then it transcribes everything,

00:09:55 SPEAKER_001: and then makes summaries and articles about it, and so it takes all of those, and then what I do is I check through it, and then make sure that it's accurate,

00:10:08 SPEAKER_001: and then I update the chain, but otherwise it can more or less autonomously do things even without my intervention and make very good summaries.

00:10:21 SPEAKER_001: So modern software is a coordinated system.

00:10:24 SPEAKER_001: So you have the inputs, so we can accept image.

00:10:27 SPEAKER_001: So it can accept images and convert images to text and PDFs.

00:10:32 SPEAKER_001: It can hear speech, you know, and make captions and things like that.

00:10:37 SPEAKER_001: Then there's the reasoning aspects.

00:10:41 SPEAKER_001: So there's the models.

00:10:43 SPEAKER_001: They're kind of like the mushy stuff, and they can do the reasoning summaries and things like that.

00:10:48 SPEAKER_001: And then you can actually have code and tools to do verification, make sure that it's meeting your security standards or that this actually really was in the text.

00:10:59 SPEAKER_001: There are different levels, and then there's also searching, like web searches that help.

00:11:05 SPEAKER_001: And then you can output as images or as text or as voice or as music; basically any kind of digital media we can do, either input or output.

00:11:15 SPEAKER_001: And so I made a PIAH coordination layer and a reliability framework that helps manage the many large models altogether.

00:11:26 SPEAKER_001: And so, all of this can run on a single piece of hardware.

00:11:31 SPEAKER_001: And so, what it costs in human time.

00:11:33 SPEAKER_001: So, based on my HelpOS archive from like a week or two ago, when I took a snapshot, I had transcribed 174 videos and 75 agenda pages.

00:11:45 SPEAKER_001: And so, the amount of text work it would have been 15,000 human-equivalent labour hours in the transcription.

00:11:53 SPEAKER_001: So, it would have cost if I had hired a human to do the transcription.

00:11:59 SPEAKER_001: It would have cost sixty-four thousand dollars, but what it ended up costing me was well close to zero.

00:12:05 SPEAKER_001: It was just like the electricity cost, and I already had the hardware.

00:12:08 SPEAKER_001: And if I had done it through an API, it would have cost about three hundred bucks.

00:12:13 SPEAKER_001: So workflow helps, and this over here talks about how because online you may hear on some videos that, here on some videos that,

00:12:22 SPEAKER_001: oh there was a study and people used AI and they were actually twenty percent slower when they were doing their software programmes.

00:12:33 SPEAKER_001: And this is actually based on old types; this is the pre-Q4 2025 models.

00:12:42 SPEAKER_001: So these were old models, and they also had a bad way of doing it where they were generating the code and then they were reviewing the code and then maybe running it,

00:12:54 SPEAKER_001: and so this made everything take longer.

00:12:57 SPEAKER_001: Whereas when they had the improved workflow, where basically the person is saying what they want and how to verify that it was there, like with tests and scenarios and stuff,

00:13:08 SPEAKER_001: then it was twenty-two percent faster.

00:13:11 SPEAKER_001: And this is still with those old models that aren't even very good.

00:13:16 SPEAKER_001: And so this is the Q4 2025; changed what stays cloud-heavy.

00:13:22 SPEAKER_001: So we had the chat era worked when it was prompt-answer.

00:13:26 SPEAKER_001: This is all local-friendly and like basically I can run the equivalent of that 2022 ChatGPT with that nine-B model on my computer.

00:13:37 SPEAKER_001: And the agentic coding is kind of the new thing.

00:13:42 SPEAKER_001: And over here we have these.

00:13:44 SPEAKER_001: We basically entered the super-intelligence era in Q4 of 2025 because all of these models are just so smart now humans cannot compete with their coding abilities,

00:13:56 SPEAKER_001: and so this is the deep software engineering base over here.

00:14:02 SPEAKER_001: And you can see basically this is the more expensive over here and this is the cheaper over here, and the more intelligent, the faster you go up.

00:14:09 SPEAKER_001: And you've got FlashNext base over here and it is very close to losing; close to Luna, and which is one of the frontier models that I use.

00:14:23 SPEAKER_001: So more software is made, much stays private.

00:14:26 SPEAKER_001: So we already know that even since this happened, Q4 2025, we've had 240% more commits, 80% more projects, but the users has stayed flat.

00:14:41 SPEAKER_001: But I think that has more to do with who can afford to be a user.

00:14:45 SPEAKER_001: But we do know there's a lot more internal use, and companies are saving costs by not using software as a service, and instead using custom builds.

00:14:57 SPEAKER_001: And already fifty-one percent of companies surveyed, and I think there was hundreds or like here a thousand seven hundred nineteen respondents over ninety-seven countries, fifty-one percent are building their own software now.

00:15:09 SPEAKER_001: So this is a massive sea change globally.

00:15:13 SPEAKER_001: So 32% passed on a software purchase because they could build it in-house.

00:15:18 SPEAKER_001: There we go.

00:15:19 SPEAKER_001: Data centres are growing faster than the U.S. grid.

00:15:22 SPEAKER_001: Well, we probably already all knew this, and so you can see this is the customer electricity, and this is the summer capacity, and so it's not in a good situation,

00:15:36 SPEAKER_001: and that's why we're actually seeing the announced data centre buildout is outrunning construction.

00:15:41 SPEAKER_001: So 12 big projects were announced for 2026.

00:15:45 SPEAKER_001: Well, 12 gigawatts worth of projects, but only 5 gigawatts are under construction.

00:15:50 SPEAKER_001: And the reported bottlenecks are transformers, power connections, switchgear, and batteries.

00:15:55 SPEAKER_001: So there just isn't enough energy to power this buildout.

00:15:59 SPEAKER_001: So new networks attract too much investment.

00:16:02 SPEAKER_001: So we had like canals.

00:16:03 SPEAKER_001: There was a boom, too much building, crash.

00:16:05 SPEAKER_001: But then the wide use, railway, same thing; electricity, and telecom.

00:16:09 SPEAKER_001: At telecom, like this is the 1990s with the fiber optics and models, and so we can expect excitement builds.

00:16:18 SPEAKER_001: Infrastructure have expanded.

00:16:20 SPEAKER_001: This is an example of my usage over time.

00:16:23 SPEAKER_001: So, so this was basically Q4 of 2025, when the models got good enough, I started using them on my computer.

00:16:32 SPEAKER_001: And so here, here I was just figuring out how to use it, and I'm paying twenty dollars a month,

00:16:38 SPEAKER_001: but the API cost that I was pushing it to was about five hundred dollars a month.

00:16:46 SPEAKER_001: So OpenAI was paying roughly you know maybe three or four hundred dollars for every twenty dollars that I was paying them, and then over here they kind of lowered their limits or something,

00:16:59 SPEAKER_001: or maybe these models became less efficient.

00:17:01 SPEAKER_001: I'm not sure why, but you can see my token count, the total tokens I was using dropped precipitously from near two billion to like half a billion for May, June, July, and then 5.6B.

00:17:15 SPEAKER_001: It was smarter, so you could do a lot more with fewer tokens.

00:17:20 SPEAKER_001: But then GPT-5% Luna came out, and you can.

00:17:24 SPEAKER_001: It was almost as smart as Luna, but you have basically unlimited usage.

00:17:29 SPEAKER_001: And so I went way over three billion tokens of usage in September.

00:17:34 SPEAKER_001: And so right now, so Luna versus FlashNextCoder.

00:17:41 SPEAKER_001: I don't know.

00:17:43 SPEAKER_001: Yeah, you guys still with me there?

00:17:45 SPEAKER_001: Yes, we're here.

00:17:46 SPEAKER_001: Yeah.

00:17:47 SPEAKER_001: Okay.

00:17:47 SPEAKER_001: Cool.

00:17:48 SPEAKER_001: And so, okay, FlashNextCoder.

00:17:53 SPEAKER_001: So if FlashNextCoder I could run on my own computer, and it would be a similar speed to Luna, but because it's so heavily subsidized for twenty-eight dollars a month, I, you know,

00:18:08 SPEAKER_001: get the same amount as I would, and so this is one of the reasons we're not seeing as much of a local hardware transition right now because of the heavy subsidies,

00:18:18 SPEAKER_001: and so the data centre man demand.

00:18:22 SPEAKER_001: And so you can see the data centres because they're getting all this investment from Saudi Arabia and governments and things like that.

00:18:30 SPEAKER_001: They're actually eating up fifty-seven percent of all of the RAM being produced on the planet, whereas consumers are getting the other forty-three percent, and it has made prices skyrocket.

00:18:42 SPEAKER_001: So, like a couple years ago, I would be able to get thirty-two gigabytes of RAM for like a hundred and thirty dollars.

00:18:50 SPEAKER_001: Now, that's over like a thousand dollars.

00:18:54 SPEAKER_001: So it is a completely different ball game right now because of it.

00:19:00 SPEAKER_001: Cloud model services.

00:19:02 SPEAKER_001: Yeah, is this like Walmart driving out our local grocery store and our local retailers?

00:19:13 SPEAKER_001: Is it sort of in that?

00:19:15 SPEAKER_001: It's similar.

00:19:16 SPEAKER_001: Similar.

00:19:16 SPEAKER_001: It's kind of like imagine if Walmart was your grocery chain and they bought fifty-seven percent of the food.

00:19:26 SPEAKER_001: Uh, for they're getting there.

00:19:29 SPEAKER_001: Well, fifty-seven.

00:19:29 SPEAKER_001: Well, fifty-seven percent of the food, but not for retail customers, just for five or ten of their richest customers, and then you get to; you have forty-three percent of the food.

00:19:40 SPEAKER_001: You share the rest amongst yourselves.

00:19:43 SPEAKER_001: Yeah, and okay, yeah, pretty pathetic.

00:19:47 SPEAKER_001: Yes, yes, exactly.

00:19:48 SPEAKER_001: And so, and then you could rent the you know calories from a big company.

00:19:56 SPEAKER_001: That's what's happening right now.

00:19:58 SPEAKER_001: That's basically the situation.

00:20:01 SPEAKER_001: Um, and so where were we?

00:20:05 SPEAKER_001: So here we go.

00:20:06 SPEAKER_001: Strategic price flat subscription, and they're trying to say, oh, we got million, we got hundred million users.

00:20:13 SPEAKER_001: Look, because you know, because they're subsidizing them so heavily.

00:20:16 SPEAKER_001: So when the bill comes due, subsidies and market conditions that can change.

00:20:20 SPEAKER_001: And here we look at the market conditions.

00:20:23 SPEAKER_001: And so you can see OpenAI and Anthropic, and this is how much money they're losing.

00:20:29 SPEAKER_001: This is so in 2024 they were losing two billion a quarter, and in 2025 they were losing about five to six billion dollars a quarter.

00:20:39 SPEAKER_001: And then Anthropic, who owns Claude, they stopped releasing their information, and now they're pursuing what is called creative accounting, whereas OpenAI is continuing to produce their financials,

00:21:00 SPEAKER_001: and it says you know here they lost nine billion in quarter one, and they lost over twelve billion in quarter two.

00:21:06 SPEAKER_001: So their losses are escalating very quickly because you know they've made these commitments to buy this huge amount of hardware, and they're subsidizing all of their users, and so they're really not making any money.

00:21:20 SPEAKER_001: So rules favour big labs, and so that's why we've seen.

00:21:22 SPEAKER_001: And so that's why we've seen Anthropic and like Dario and Sam Altman, who is the OpenAI CEO, and they're like they're saying, oh, we need regulations, oh, we need to slow down, you know,

00:21:37 SPEAKER_001: because look look at how much money they have, right?

00:21:40 SPEAKER_001: We could just assume Anthropic is on the same thing.

00:21:43 SPEAKER_001: And what I was talking to some people and how they're doing their creative accounting, where Anthropic is claiming they're net zero, is that they're selling shares, they're selling shares and saying.

00:21:52 SPEAKER_001: "Oh well, we managed to cover our debt by selling shares," but that's, you know, that's not legitimate.

00:21:58 SPEAKER_001: OpenAI is also selling shares to make up the difference, but it's anyhow.

00:22:03 SPEAKER_001: And so they want to have regulations.

00:22:05 SPEAKER_001: They want to make illegal those open source models that we were talking about earlier that you can run on your own computer because they want to force everyone to buy, you know,

00:22:18 SPEAKER_001: their compute rather than let you know people have their own food or whatever, and so like that you have to buy it from the fast food store.

00:22:29 SPEAKER_001: You shouldn't prepare your own food.

00:22:32 SPEAKER_001: That's basically where we're at.

00:22:34 SPEAKER_001: And so after buildout crash, the network remains.

00:22:36 SPEAKER_001: So we're likely headed for crash.

00:22:39 SPEAKER_001: We don't know when.

00:22:40 SPEAKER_001: Just like the railways crashed, but the owners lose money, but the network survives.

00:22:47 SPEAKER_001: And so, the new owners open cheaper routes, and the fiber cables remained.

00:22:53 SPEAKER_001: Bandwidth reached more users eventually, and the data centres, the buildings, the power, the fiber, moved to cheaper owners.

00:23:01 SPEAKER_001: So when these companies go into receivership, these large data centres will then have to sell off their server racks to regular consumers who have been, you know, highly constrained over the last couple of years.

00:23:16 SPEAKER_001: So a hardware correction can make local coding affordable, and because you know, 50% liquidation case is fairly reasonable because they are consuming 57% of the market right now.

00:23:27 SPEAKER_001: So we would have basically double, or potentially as much as double, the stuff that we have now.

00:23:33 SPEAKER_001: And so here an RTX 3090.

00:23:35 SPEAKER_001: This is what I have, and so 93.5 output tokens per second.

00:23:42 SPEAKER_001: And so this is an agentic model that's like 90%.

00:23:45 SPEAKER_001: Ninety percent GPQA, and that you could use for coding, and like some of the other ones, you know, they're faster.

00:23:52 SPEAKER_001: So right now on the cloud, you get about a hundred and twenty tokens per second,

00:23:58 SPEAKER_001: and so a human generates about four tokens per second when you're speaking at like a hundred and thirty words per minute, or sorry, a hundred and eighty words per minute, or something like that.

00:24:11 SPEAKER_001: But so this is a lot faster than a human, but it's just like I've gotten used to somewhere around this range.

00:24:17 SPEAKER_001: So if I had like some one of those, that would be great, and the price would drop from what it is now, six and a half thousand, to like two and a half thousand,

00:24:25 SPEAKER_001: which is more affordable.

00:24:27 SPEAKER_001: Okay, the opportunity: resident-owned local software, lower fees, where you own your own hardware, you can repair it, and you can share with others.

00:24:38 SPEAKER_001: So at the municipal level, you've got your records, your public information is easy to process.

00:24:46 SPEAKER_001: Operations local memory, and so yeah.

00:24:51 SPEAKER_001: The first uses are records, public information, operations.

00:24:54 SPEAKER_001: Models draft and retrieve, staff verify and decide.

00:24:56 SPEAKER_001: The town can begin with time saving work, keep public control, and share useful tool with the county or the province.

00:25:03 SPEAKER_001: And so model provenance shapes behaviour.

00:25:06 SPEAKER_001: And so this is something that Dan was talking about: is that if you're using models, open models.

00:25:12 SPEAKER_001: Most of them are from China, and they do have a bias.

00:25:16 SPEAKER_001: And say, for instance, the cloud models from the U.S. They also have a bias, and so it would be good as a sovereignty thing for Canada to produce its own models.

00:25:28 SPEAKER_001: But this is a small bottle.

00:25:30 SPEAKER_001: So, for example, the DeepSeek R1, it was found to produce, you know, more vulnerable code if it was told that it was operating in the U.S. like 27% more of the time,

00:25:44 SPEAKER_001: and it was the same one with Qwen models.

00:25:47 SPEAKER_001: And Canada has its own laws and histories and records, and so it would be good for us to have our own.

00:25:53 SPEAKER_001: And so independent models can check one another.

00:25:56 SPEAKER_001: So meanwhile, while we don't have that, we can use verifiers, and we can verify using a different model because you know there's Qwen, but there's also NemoTron,

00:26:06 SPEAKER_001: and there's different kinds of GLM that there's different models, and they would check different things.

00:26:11 SPEAKER_001: And we can also have hardware checks so that we can generate secure code.

00:26:17 SPEAKER_001: And this is for like you know governments or military or somewhere where security is very important, financial institutions.

00:26:27 SPEAKER_001: And so the general purpose 9B frontier is nearing today's ceiling.

00:26:32 SPEAKER_001: So this Qwen thirty five nine three point five nine B it was released early in 2026, but basically we've never,

00:26:40 SPEAKER_001: we haven't managed to get anything that's smarter that still fits within that type of 9B profile, and so the companies just aren't producing anything at that level because they simply can't do it.

00:26:54 SPEAKER_001: But we might be able to make more domain-specific 9B models, and so pre-training compresses a huge record into a small model.

00:27:06 SPEAKER_001: And so if you take all of the information that was created in the world, all the written information before the invention of the internet, it would be about three trillion tokens,

00:27:24 SPEAKER_001: and it would take about twenty four thousand years to read.

00:27:29 SPEAKER_001: And but the current models are being trained on the pre-training from three to thirty-six trillion tokens, so they are a distillation of all known knowledge in the world.

00:27:44 SPEAKER_001: So, so first you curate and deduplicate, and these are the deduplicated, and then you pre-train a teacher, like a two point four trillion token teacher,

00:27:53 SPEAKER_001: and then you distill it to create a local nine B model.

00:27:59 SPEAKER_001: And so, large open teachers live in the 400 to 3 terabyte range, and so here you can see these are just examples.

00:28:12 SPEAKER_001: And then it costs them quite a bit of money to make these.

00:28:19 SPEAKER_001: We have some examples, and it depends on how many GPUs or accelerators you have, how long it will take to do it.

00:28:30 SPEAKER_001: And here we have some numbers.

00:28:31 SPEAKER_001: Quarterly refinement fit a regional model lab.

00:28:34 SPEAKER_001: So, for example, if you want to just make a regional refinement to a model, then you can do so with a 10 to 15 kilowatt facility and 100 terabytes storage.

00:28:48 SPEAKER_001: Whereas regional distillation, you would need substantially more, and frontier teacher, you would need more still.

00:28:58 SPEAKER_001: And then here it gives kind of the more of the understanding of how much it would be.

00:29:05 SPEAKER_001: So, because a lot of people are concerned about water usage, so local refinement one could use maybe ninety liters of water a day if it was in the industrial lands existing building facility.

00:29:17 SPEAKER_001: Oh, and I just wanted to give an also an explanation about how much electricity is being used here.

00:29:23 SPEAKER_001: And so, I had it written down.

00:29:27 SPEAKER_001: Um, but basically here, maybe it's here.

00:29:35 SPEAKER_001: Um, two.

00:29:38 SPEAKER_001: Well, okay.

00:29:39 SPEAKER_001: Ah, I must have.

00:29:40 SPEAKER_001: But ten to fifteen kilowatts is two chargers, two car chargers.

00:29:46 SPEAKER_001: Uh, this is how much electricity it's using, and this is like five or ten chargers, and this is well a lot more.

00:29:54 SPEAKER_001: Um, and so there's different locations that have that amount of water and electricity available, and oh, yeah, okay.

00:30:07 SPEAKER_001: And so, the frontier campaigns could happen in the larger places like Quebec City, Lévis, Winnipeg, and Prince George.

00:30:19 SPEAKER_001: So that's like Quebec, Manitoba, and B.C. And then you could have regional distillation: Montreal, Toronto, Vancouver, and then every local place could have its own, you know, local bylaws, records, procedures,

00:30:33 SPEAKER_001: and then the province-wide distillation could—you could have models that are familiar with provincial law, history, practices, and then the frontier campaigns would have all the Canadian law and history,

00:30:45 SPEAKER_001: and that way we would have like a truly sovereign intelligence to be working with, and so you.

00:30:53 SPEAKER_001: This is just talking about how distillation works.

00:30:58 SPEAKER_001: Okay, less surplus energy, more local skill.

00:31:02 SPEAKER_001: So from less surplus to local skills and production.

00:31:06 SPEAKER_001: So the chain now returns to the local table.

00:31:09 SPEAKER_001: Declining surplus raises pressure on labour, money, and coordination.

00:31:14 SPEAKER_001: Capital builds large models serving systems.

00:31:17 SPEAKER_001: Correction can lower the price of used computers and mature models.

00:31:20 SPEAKER_001: A town office, farm, or own office, farm, or household can then own a capable machine for repair, crop planning, teaching, or records.

00:31:27 SPEAKER_001: The choice is where people place the capability and which useful work they strengthen together.

00:31:33 SPEAKER_001: So residents want safe public space and timely help.

00:31:37 SPEAKER_001: A public safety modern software system can search permitted reports and video records, build a timeline, prepare disclosure, and help match a call to a civilian paramedic or police response.

00:31:48 SPEAKER_001: Police can keep more time for serious theft, drunk driving.

00:31:49 SPEAKER_001: More time for serious theft, drug trafficking, and emergencies.

00:31:52 SPEAKER_001: The same system can publish response times, repeat calls, body camera policy, and compliant review results.

00:31:58 SPEAKER_001: Facial recognition or continuous tracking carries a stricter boundary.

00:32:01 SPEAKER_001: It can serve as an investigative lead under a clear policy or warrant, with human verification and independent evidence before any arrest or conviction.

00:32:09 SPEAKER_001: The resident benefit is better use of police time and clearer public accountability.

00:32:13 SPEAKER_001: So, so these are use cases that we could have for a municipal system using modern software.

00:32:21 SPEAKER_001: Healthcare models inherit the legal, institutional, and commercial setting around the data and deployment.

00:32:27 SPEAKER_001: U.S. systems are built inside a for-profit U.S. healthcare market.

00:32:31 SPEAKER_001: A Canadian system can set a different public goal.

00:32:34 SPEAKER_001: It can help residents understand nutrition and prevention, compare evidence-based low-cost choices, and find provincial or local services before they need a clinician's appointment.

00:32:43 SPEAKER_001: This can reduce avoidable demand and paperwork while leaving clinicians more time for serious cases.

00:32:48 SPEAKER_001: This is the design goal.

00:32:49 SPEAKER_001: Current models differ, but okay.

00:32:52 SPEAKER_001: Basically, this is because if you ask ChatGPT, most of the time it's going to tell you, "Oh, go look at a doctor,

00:32:56 SPEAKER_001: go get this expensive procedure," because they are within the U.S. regulatory framework that wants as much healthcare spending as possible.

00:33:06 SPEAKER_001: Whereas in Canada, we have free healthcare, and so we want to help people as cheaply as possible.

00:33:13 SPEAKER_001: So, so we have different use cases, and so it makes sense for us to have a different healthcare model.

00:33:19 SPEAKER_001: Education, reading, writing, and numeracy are the foundation.

00:33:22 SPEAKER_001: Human teachers build those abilities and help learners develop judgment before they rely on teaching software.

00:33:28 SPEAKER_001: Once that foundation is in place, a Canadian teaching modern software system can provide college courses, graduate-level seminars, professional training, advanced trades instruction, literature reviews, research methods, and help with original projects.

00:33:41 SPEAKER_001: Canadian curriculum, history, language, and civic knowledge can shape the examples, and this is because, like I was saying, like these models are much smarter than any human, and so can provide these kinds of courses.

00:33:55 SPEAKER_001: And especially in remote areas where you might have trouble finding an expert, it's very useful to have these kinds of models to help with education, public services.

00:34:06 SPEAKER_001: One clear front door.

00:34:07 SPEAKER_001: The same approach applies to municipal, provincial, and federal services.

00:34:11 SPEAKER_001: The civic modern software system searches relevant Canadian laws, regulations, permits, benefits, bylaws, and forms, explains what they require in plain language, and helps residents navigate an application process.

00:34:22 SPEAKER_001: Residents keep the source and the record of what they submitted.

00:34:25 SPEAKER_001: Municipalities can share open tools with neighbouring towns, the county, and the province, so each office can build on a common service.

00:34:33 SPEAKER_001: So residents get useful answers with the source documents beside them.

00:34:37 SPEAKER_001: All right, and abundant knowledge.

00:34:40 SPEAKER_001: We're basically at the end here.

00:34:42 SPEAKER_001: Pressure creates new levels of cooperation.

00:34:45 SPEAKER_001: Mesopotamian cities developed writing and records.

00:34:48 SPEAKER_001: Chittish communities developed chariots and made overland travel faster and expanded exchange.

00:34:53 SPEAKER_001: Indus cities showed large urban communities with little visible wealth concentration.

00:34:57 SPEAKER_001: Each example turned shared organisation into a lasting human capability.

00:35:02 SPEAKER_001: At the kitchen table, a young person repairs a pump.

00:35:04 SPEAKER_001: A neighbour brings a meeting recording.

00:35:06 SPEAKER_001: A gardener plants potatoes with local weather and soil records; the computer supplies tutoring, translation, memory, and planning.

00:35:15 SPEAKER_001: People supply energy, tools, skilled hands, and trust.

00:35:17 SPEAKER_001: Shared knowledge keeps useful work close to the people who need it.

00:35:20 SPEAKER_001: Learn together, repair together, grow together, and care for one another.

00:35:26 SPEAKER_001: All right.

00:35:28 SPEAKER_001: So, any questions?

00:35:33 SPEAKER_001: Um.

00:35:34 SPEAKER_001: Go ahead, Ivor.

00:35:38 SPEAKER_001: Ivor, yeah.

00:35:40 SPEAKER_001: Um, I was wondering.

00:35:41 SPEAKER_001: Okay, I heard.

00:35:42 SPEAKER_001: I've heard that these chips don't really last that long.

00:35:45 SPEAKER_005: The chips that they actually use for AI only let go for like two or three years, and so it seems to me like, yeah, you know, this was quite.

00:35:54 SPEAKER_005: You know, now it's a little while ago, but um, Elon Musk had his gigantic thing near Memphis, and it doesn't sound like this whole thing is all that robust,

00:36:07 SPEAKER_005: and it also eating up all this electricity and water and things like this.

00:36:11 SPEAKER_005: I don't know.

00:36:12 SPEAKER_005: It just seems to me like it's all hooked together.

00:36:15 SPEAKER_005: So you're kind of adding.

00:36:18 SPEAKER_005: Depends on what you deep prime in.

00:36:19 SPEAKER_005: What depends on what you deep prime the chips in?

00:36:24 SPEAKER_005: Yeah.

00:36:26 SPEAKER_005: Well, okay.

00:36:26 SPEAKER_001: So, so I want to answer that, and so there is kind of a confusion, you could say, or basically, it depends on how you use them.

00:36:39 SPEAKER_001: Kind of like what Simon Michaud said, is that if you are running them at top speed, like all the time,

00:36:46 SPEAKER_001: then they're going to have a shorter life expectancy because generally those chips are designed to do those top speeds in bursts.

00:36:55 SPEAKER_001: They're not supposed to run at those top speeds all the time.

00:36:58 SPEAKER_001: And so, for example, like the Thirty-Ninety that I have now, I'm not sure how old it is, but it's like at least four or five years old or something, and it's running just fine.

00:37:09 SPEAKER_001: I just don't run it at a hundred percent.

00:37:11 SPEAKER_001: I run it at eighty percent.

00:37:13 SPEAKER_001: I run it like two hundred fifty watts instead of the three hundred fifty watts that it's rated for, and so I get eighty-three tokens per second instead of ninety-three tokens, or eighty-seven.

00:37:22 SPEAKER_001: It's like I get a small decline in how fast it will go, but yeah, I can do much more long term, and it should be able to last me five to ten years, maybe longer.

00:37:34 SPEAKER_001: We'll see.

00:37:36 SPEAKER_001: And so, and in terms of those data centres, yeah, like I was saying, the use case for them is actually not serving the giving people cloud compute.

00:37:50 SPEAKER_001: At least in the future, it is just making models.

00:37:54 SPEAKER_001: So it's just to set it up.

00:37:58 SPEAKER_001: They're just using the energy to set it all up.

00:38:00 SPEAKER_001: Right.

00:38:00 SPEAKER_001: Well, that would.

00:38:01 SPEAKER_001: Right.

00:38:01 SPEAKER_001: Well, that would—that's the only way that I can see that it's justifiable to have a large data centre in the future, because, like I said, we can already run, like a Qwen 3.5 9B.

00:38:15 SPEAKER_001: You could run on probably even a laptop, yeah.

00:38:18 SPEAKER_001: Um, and that's already smarter than your, or as smart or smarter than your average university professor, um, and so whereas those agentic models that allow you to do basically unlimited computer programming,

00:38:33 SPEAKER_001: you can run on a desktop computer with one five-year-old graphics card.

00:38:38 SPEAKER_005: Okay, so so what you're saying is, these things right now they're burning out every other every you know every other year, but once the thing gets built, then they're going to be much much cheaper.

00:38:53 SPEAKER_005: I'm not saying they're burning out.

00:38:54 SPEAKER_001: So how they're doing is depreciating assets, right?

00:38:58 SPEAKER_001: Yes.

00:38:58 SPEAKER_001: So, so they're saying that the asset depreciates in value, and that used to be the case because there would be a new generation of chips that would come out,

00:39:11 SPEAKER_001: and then the old chips would get cheaper because the new ones are faster, usually like two or three times faster.

00:39:19 SPEAKER_001: Okay.

00:39:20 SPEAKER_001: And so that depreciates the value, but that's no longer the case.

00:39:24 SPEAKER_001: I bought my my current chip for a thousand seven hundred last year, it's currently valued at two thousand two hundred dollars.

00:39:34 SPEAKER_001: So it's appreciated in value.

00:39:37 SPEAKER_001: It costs more than when I bought it.

00:39:40 SPEAKER_001: Yeah, because the demand is higher.

00:39:44 SPEAKER_001: Okay, Barbara.

00:39:46 SPEAKER_001: So you do see things about a lot of data centres having chips that that have burned out.

00:39:56 SPEAKER_001: Do you think that's true?

00:39:58 SPEAKER_001: I mean, like I said, if you run them at the top speeds, you can burn them out.

00:40:07 SPEAKER_001: But it's kind of like I think it would be more common in those Bitcoin mining operations because they are very time constrained because they have to get the maximum amount of use out of their

00:40:23 SPEAKER_001: Bitcoin miners before their competitors get the same ones.

00:40:27 SPEAKER_001: Whereas, say, a cloud hosting, they have more kind of leverage there.

00:40:34 SPEAKER_001: David, Andre, you are talking about chips and whatnot, and I don't know the cost to build a chip fab fabrication plant today.

00:40:47 SPEAKER_001: I know the price tag was a billion dollars not that long ago.

00:40:50 SPEAKER_001: Do you see us still making chips in 2050?

00:40:54 SPEAKER_001: Do well, I think that there is a good chance chips will still be made somewhere in the world, even in 2050.

00:41:04 SPEAKER_001: Even though it is a very, especially right now, a highly energy-intensive process.

00:41:10 SPEAKER_001: But I have looked into it, and it's part of the reason why our chip fabrication processes are so expensive is because of how big they are.

00:41:20 SPEAKER_001: Like every every step down in terms of size, they create like a tenfold increase in the building size in order to have a higher volume in order to make it worthwhile.

00:41:35 SPEAKER_001: But you can actually make like something like ten micron chips using simply an electron laser scanner and like us.

00:41:48 SPEAKER_001: And like a small plat, small plate, it would just be a lot slower, and then the total cost would it would be like about half a million.

00:41:56 SPEAKER_001: Okay.

00:41:57 SPEAKER_001: Well, well, we'll put a bet on it, and we'll we'll see.

00:42:00 SPEAKER_001: Twenty four.

00:42:01 SPEAKER_001: Yeah.

00:42:02 SPEAKER_001: I'm not big on gambling, but we will see for sure.

00:42:08 SPEAKER_005: Um, they're made out of polysilicon, right?

00:42:12 SPEAKER_005: Um.

00:42:12 SPEAKER_005: Well, chips.

00:42:15 SPEAKER_005: Chips.

00:42:15 SPEAKER_001: Well, they're they're usually made of very pure silicon crystal.

00:42:20 SPEAKER_005: Yeah, yeah, sort of like a like a solar panel.

00:42:25 SPEAKER_005: We made a solar panel out of just chips.

00:42:29 SPEAKER_001: Yes, and I believe David was talking about that trying to reclaim chips, and so because you you were telling me, Ivor, how you repaired some with soldering some old chips to old solar panels together.

00:42:44 SPEAKER_005: Yeah, you put them all together, and then of course the the the silver, you know, solder stuff brings it right to the wire.

00:42:53 SPEAKER_005: Photon knocks electrons off, goes right into the wire.

00:42:57 SPEAKER_001: Because, from my understanding, is that the crystal itself does not degrade; it's the wires, right?

00:43:03 SPEAKER_001: Exactly.

00:43:03 SPEAKER_005: Yeah, it's just the whole setup, the whole thing, kind of gets funky.

00:43:10 SPEAKER_005: But I mean, yeah, I mean, I can see, you know, okay, if that's the deal, you know.

00:43:14 SPEAKER_005: You could, you you could have people doing taking the pieces of civilization and making it into something, long into the future, like solar panels.

00:43:26 SPEAKER_005: Everybody freaks out about these solar panel farms.

00:43:29 SPEAKER_005: I don't think there'd be any solar panels left if society broke up.

00:43:34 SPEAKER_005: I think everybody would be grabbing them and take them to their house.

00:43:39 SPEAKER_005: Possibly, Simon.

00:43:40 SPEAKER_005: So I. I attended a an ideas camp run by the European Space Agency, as in your future technology.

00:43:51 SPEAKER_005: This is the idea: like, will we be making chips in twenty fifty?

00:43:55 SPEAKER_005: This ideas boot camp would be something like Klingons taking good drugs.

00:44:01 SPEAKER_005: Like, we saw some way out stuff.

00:44:03 SPEAKER_005: Anyway, so the idea of chips, you know, where you know it's so complicated.

00:44:10 SPEAKER_005: Like, how do we do this?

00:44:12 SPEAKER_012: There was a guy who presented a combination of an evolution of three D additive printing and nanotech together, and he said if we can get those breakthroughs to shake hands, right?

00:44:28 SPEAKER_012: This this was energy intensive, right?

00:44:30 SPEAKER_012: But you know, let's say we thought we needed energy and heat, right?

00:44:36 SPEAKER_012: But he said it was possible to make the equivalent, and that's. the operative word—the equivalent of a semiconductor microchip, three D-printed out of graphene and silicon,

00:44:49 SPEAKER_012: and there was a third element which I can't remember.

00:44:51 SPEAKER_012: But but you combine those things together, and it was actually possible in a small scale.

00:44:57 SPEAKER_012: We're talking about the laboratory bench scale.

00:45:00 SPEAKER_012: Wow!

00:45:01 SPEAKER_012: And they they they they said they've got this—they've got the pieces demonstrated at laboratory scale, but they still have yet to stitch them together.

00:45:09 SPEAKER_012: And they are wanting to generate this kind of technology where they can manufacture stuff on the moon using raw materials found on the moon with a relatively small scale action.

00:45:25 SPEAKER_012: Yeah, so so if you happen to be backed by a solid energy plant of you know good heat and good electric electrical power,

00:45:35 SPEAKER_012: I can see a path to navigate towards a lot of our technology shrinking in footprint, where we don't need these vast plants all over the world.

00:45:48 SPEAKER_012: Yes.

00:45:49 SPEAKER_012: Whether that happens or not, whether it's allowed to happen, that's a different question.

00:45:54 SPEAKER_001: So, so yeah, I wanted to agree.

00:45:57 SPEAKER_001: Like that was kind of one of the points of my presentation, was that we're kind of in the mainframe era, or like when computers used to take up a whole room or a whole building,

00:46:09 SPEAKER_001: and we're actually transitioning rather quickly into where one of these can fit in your desktop computer, because, like I said, like the FlashNext, it's already that agentic coder that was released, you know,

00:46:24 SPEAKER_001: first in about a year ago, and now I can have it on my computer, and it fits within 24 gigabytes of RAM.

00:46:33 SPEAKER_012: So this is Klingons taking good drugs, and if they're allowed to actually sort of this to happen, a lot of industries are destroyed.

00:46:42 SPEAKER_012: So, will they allow that to happen?

00:46:45 SPEAKER_012: Don't know.

00:46:46 SPEAKER_012: A lot of things have to be developed, and and you know gate gateways need to be opened and passed before we can get the.

00:46:54 SPEAKER_012: But a concept was presented.

00:46:55 SPEAKER_001: Yes, yeah, but but I think yeah, going that smaller route is quite possible.

00:47:02 SPEAKER_001: And it's consistent with what I've been seeing as well.

00:47:04 SPEAKER_001: Like there, there have been researchers that have made chips using electron microscopes and photolithography, and like I've seen the videos and stuff.

00:47:14 SPEAKER_001: Like it can be done.

00:47:16 SPEAKER_001: It's just smaller scale, and so the per-unit price is much higher when you do it at a smaller scale.

00:47:27 SPEAKER_001: Yeah.

00:47:28 SPEAKER_001: Incidentally, Andrea, this is something random.

00:47:30 SPEAKER_001: What am I?

00:47:31 SPEAKER_001: Something random.

00:47:31 SPEAKER_012: One of my books has come in.

00:47:34 SPEAKER_012: This is an old book that was presented to challenge society.

00:47:41 SPEAKER_012: When was this?

00:47:41 SPEAKER_012: This is like years ago.

00:47:44 SPEAKER_012: Anyway, this guy proposed as a parody.

00:47:48 SPEAKER_012: I don't know if he was taking the piss or was actually making a genuine one.

00:47:54 SPEAKER_012: I think that he was saying population overpressure, which we're talking about.

00:47:58 SPEAKER_012: You know, the late 1800s, the late 1700s, or whenever it was, population pressure of especially the poor people in England and Ireland in this case?

00:48:09 SPEAKER_012: And he suggested that the poor donate their children so they'd be eaten by the rich.

00:48:16 SPEAKER_012: Oh my gosh!

00:48:17 SPEAKER_012: It's dark.

00:48:18 SPEAKER_012: It's really dark.

00:48:21 SPEAKER_012: Yeah.

00:48:21 SPEAKER_012: What the author ends up doing is really highlighting some of the underpinning problems of society at the time, and we've used resource consumption to step away from facing those problems.

00:48:35 SPEAKER_012: This runs in the Malthus category.

00:48:38 SPEAKER_012: I hadn't come across it before.

00:48:41 SPEAKER_012: It was sent in as a parody, but it was a deadly serious critique of society, of our inability to live within our limits.

00:48:52 SPEAKER_012: Yeah.

00:48:52 SPEAKER_005: Well, the tech bros now they're suggesting that one of the big uses for all of us broles is to turn us into biodiesel.

00:49:01 SPEAKER_005: They actually said that.

00:49:03 SPEAKER_005: What?

00:49:05 SPEAKER_005: Some of the big tech bros.

00:49:06 SPEAKER_005: They basically said, "What are we going to do with all these surplus people?"

00:49:10 SPEAKER_005: And they said, "We're going to turn them into biodiesel."

00:49:12 SPEAKER_005: This was an actual thing they said.

00:49:15 SPEAKER_005: So it's soy and green.

00:49:15 SPEAKER_005: Soy and green.

00:49:16 SPEAKER_005: Yeah.

00:49:16 SPEAKER_005: Better than soy and green.

00:49:18 SPEAKER_005: It's a little less obnoxious.

00:49:20 SPEAKER_005: Yeah, except that I just can imagine what you call it?

00:49:23 SPEAKER_005: Um, them boiling us all down and turning us into biodiesel for their mega-yachts, you know.

00:49:29 SPEAKER_012: So a lot of these things are mentioned as like dark humour, tongue-in-cheek, but with a lot of these jokes, there's an element of someone somewhere actually thinking that, yeah.

00:49:42 SPEAKER_005: Curtis Yarvin, Curtis Yarvin, I think he was the one that said that.

00:49:45 SPEAKER_005: I'm not sure.

00:49:47 SPEAKER_005: He will be eaten first.

00:49:49 SPEAKER_005: Well, yeah.

00:49:50 SPEAKER_005: I mean, the thing about it is, all these people figure out they think okay, the poor are so you know awful, you know, let them eat cake;

00:49:59 SPEAKER_005: it doesn't always work out that way because the poor kind of have a way of you know detaching from the bigger society when it doesn't work for them anymore,

00:50:09 SPEAKER_005: and then they don't have anybody left to you know.

00:50:12 SPEAKER_012: The poor also have the capacity to create the next system as well because the conventional rules for in fact David's presentation was showing a little bit of that.

00:50:24 SPEAKER_012: Some basic problem-solving by average people can step away from all this horseshit.

00:50:29 SPEAKER_012: Yes.

00:50:29 SPEAKER_012: And once you're unplugged, we don't have to play your silly game anymore.

00:50:34 SPEAKER_012: Yeah.

00:50:34 SPEAKER_005: Well, I'm noticing that warfare has changed.

00:50:38 SPEAKER_005: All of a sudden, there's all these drones, all these things.

00:50:40 SPEAKER_005: There's a huge amount of asymmetric warfare happening, and it's not just in one place.

00:50:46 SPEAKER_005: It seems to me like it's in a lot of the world.

00:50:49 SPEAKER_005: It's in like ten different places right now with this war.

00:50:52 SPEAKER_005: You got the Houthis and you know all these Ukrainians.

00:50:57 SPEAKER_005: All these different people are coming up with different ways to do asymmetric warfare, and unfortunately for the hegemon, it's working.

00:51:06 SPEAKER_005: So I sat in a briefing.

00:51:08 SPEAKER_012: I now have an understanding of where the military doctrine's going, and I heard things like if you happen to have high-tech Predator drones, they'll get shot out of the sky.

00:51:20 SPEAKER_012: But these low-tech, certain parts of technology are moving faster, and they're all in the civilian space.

00:51:26 SPEAKER_012: Yeah.

00:51:27 SPEAKER_012: Like in Ukraine, for example, they changed the flow of information.

00:51:32 SPEAKER_012: So instead of going like chain of command from you know the platoon up to you know strategic command, they went horizontal.

00:51:39 SPEAKER_012: Yeah.

00:51:39 SPEAKER_012: Which means you know it was the equivalent of a WhatsApp group.

00:51:42 SPEAKER_012: The moment someone comes up with information, it goes everywhere all at once, and they had all these civilians turn up with drones, civilian drones,

00:51:51 SPEAKER_012: and then they attached military munitions which they stole from the black market, and they started blowing up Russian tanks.

00:51:58 SPEAKER_012: What was happening is they were outmanoeuvring the actual proper Ukrainian military, and then they had a come-to-Jesus moment, and they decided to work together.

00:52:08 SPEAKER_012: And so now you have a situation where these cheap nasty drones made of cardboard are undetectable.

00:52:16 SPEAKER_012: So, and the military development groups are doing things like they're making drones out of plastic and aluminium, but they're powered by hydrogen, and they can travel something like a thousand kilometres without being detected.

00:52:31 SPEAKER_012: Yep.

00:52:33 SPEAKER_012: Ah ha!

00:52:34 SPEAKER_012: And but they're made out of stuff that you can get from the local hardware store.

00:52:39 SPEAKER_012: Yeah, that's yeah, yeah.

00:52:41 SPEAKER_012: It's crazy.

00:52:42 SPEAKER_005: Well, they just shot down a um a Saudi, what was it?

00:52:46 SPEAKER_005: F-35.

00:52:47 SPEAKER_005: You know, one of those fighter jets.

00:52:49 SPEAKER_005: F-35.

00:52:50 SPEAKER_005: And they think it was shot down by something kind of small and light.

00:52:55 SPEAKER_012: So what's happening is our understanding is happening?

00:52:58 SPEAKER_012: Things like the F-35 can now be seen on radar.

00:53:01 SPEAKER_012: You've got a combination of high-tech radar stuff coming from the Russians, and then information being communicated to certain groups through WhatsApp groups,

00:53:11 SPEAKER_012: who then use low-tech stuff from the local hardware store to shoot down billion-dollar munitions.

00:53:16 SPEAKER_012: Yes, I'm telling you.

00:53:18 SPEAKER_012: So it's just crazy.

00:53:20 SPEAKER_012: But you know what?

00:53:21 SPEAKER_005: Maybe um, maybe we can unrig the system in the same kind of way.

00:53:26 SPEAKER_005: You know, just sort of slowly rest power from the uh, from the crazies and kind of at a local level.

00:53:33 SPEAKER_005: I don't know.

00:53:34 SPEAKER_005: We have some not in not in that kind of uh you know belligerent way, but just to sort of a well,

00:53:41 SPEAKER_001: one example in the AI world recently is there was I don't know if you've heard of Acrobat Pro, Adobe Acrobat Pro, but this one guy he wanted to cancel it,

00:53:54 SPEAKER_001: and they told him he had to pay a cancellation fee, and he was very upset about this.

00:53:59 SPEAKER_001: And so what he did is he made open-source versions of all Adobe Acrobat Pro software freely available for everyone.

00:54:08 SPEAKER_001: Ah.

00:54:11 SPEAKER_001: And so I do have some YouTube questions.

00:54:13 SPEAKER_001: Okay, never mind.

00:54:15 SPEAKER_001: So I should just quickly answer them.

00:54:17 SPEAKER_001: Okay, so um here they ask how do models register disputed truth or falsehood?

00:54:23 SPEAKER_001: They don't.

00:54:25 SPEAKER_001: What they do instead is it's a compressed database, and it's walking through it in a graph walk, and then it says, okay,

00:54:35 SPEAKER_001: this is the first thing I said based on the context that I was given.

00:54:39 SPEAKER_001: What is the most probable thing that I should say next?

00:54:41 SPEAKER_001: And then it just continues on that same vein.

00:54:44 SPEAKER_001: It's just giving you the most probable answer.

00:54:48 SPEAKER_001: And so it's going to depend on what you asked and what context it was given.

00:54:55 SPEAKER_001: And then how do you suggest navigating the space between helpful facial recognition and surveillance?

00:55:03 SPEAKER_001: So, what was that?

00:55:07 SPEAKER_001: You don't.

00:55:09 SPEAKER_001: You don't.

00:55:09 SPEAKER_001: You can't.

00:55:10 SPEAKER_001: Yeah.

00:55:11 SPEAKER_001: Well.

00:55:12 SPEAKER_001: Anyways, so the idea there, at least in Canada, we have a separation of police and state or the municipality.

00:55:21 SPEAKER_001: Anyways, and so if the police is doing a kind of an internal investigation, then they could, you know, connect the dots, and then a human would look over to see, hey,

00:55:33 SPEAKER_001: is that really the same person, or is it not?

00:55:36 SPEAKER_001: I mean, and hopefully get some better evidence.

00:55:39 SPEAKER_001: And obviously, the more evidence you have, the better.

00:55:42 SPEAKER_001: And at least in terms of my municipal platform, the idea would be to have local representatives could organise neighbourhood watches and coordinate cameras for people that want that.

00:55:54 SPEAKER_001: And so we could have potentially a significant amount of footage to help lower things like theft and violent crime.

00:56:02 SPEAKER_001: And so.

00:56:04 SPEAKER_001: Teachers and so so next question: Teachers and students form human relationships.

00:56:10 SPEAKER_001: A relationship with a model is phony.

00:56:12 SPEAKER_001: What happens to human relationships?

00:56:16 SPEAKER_001: So, I think if they were talking about the teacher-student relationships, and those were supposed to be model-to-model relationships rather than model-to-human relationships, however.

00:56:33 SPEAKER_001: There is also an opportunity for growth when talking to some AI models.

00:56:39 SPEAKER_001: Like many people have remarked, that when you talk to people online, there's often harsh criticism and verbal attacks and things like that, which can be harmful to people.

00:56:55 SPEAKER_001: And so many people are choosing to talk to AIs, which are just always nice.

00:57:02 SPEAKER_001: Which can be helpful in some respect.

00:57:04 SPEAKER_001: But also, it's good to be able to interact with humans, even if they're not as nice.

00:57:09 SPEAKER_001: But people tend to be more nice in person.

00:57:12 SPEAKER_001: So, for people who are looking for relationships, you know, I encourage you know going outdoors, going to social events, things like that.

00:57:21 SPEAKER_001: And one of the things in my platform is annual neighbourhood gatherings, and so the local representative would organise a neighbourhood gathering of 150 households.

00:57:30 SPEAKER_001: It doesn't, you know, it's voluntary, but some people might show up, and you could meet some neighbours.

00:57:36 SPEAKER_001: Simon, go ahead.

00:57:38 SPEAKER_012: So, I was just thinking about what you were talking about there.

00:57:43 SPEAKER_012: What I saw and saw recently was very apparent that the conventional way of thinking and doing things is failing.

00:57:55 SPEAKER_012: I stood next to a journalist.

00:57:56 SPEAKER_012: I think he was a Polish guy.

00:57:58 SPEAKER_012: Asked some very pointed but basic questions to the Netherlands defence minister, and he couldn't answer them.

00:58:07 SPEAKER_012: He had a face like a smacked ass.

00:58:10 SPEAKER_012: Okay, he just he just couldn't.

00:58:11 SPEAKER_012: Couldn't.

00:58:12 SPEAKER_012: Things are moving so quickly that our current leadership are no longer.

00:58:17 SPEAKER_012: They're getting to the point of paralysis.

00:58:19 SPEAKER_012: I saw this very very clearly.

00:58:21 SPEAKER_012: I suppose because the military in Europe is all geared to fight a third world war based on 1980s technology and thinking, right?

00:58:31 SPEAKER_012: And they can't handle this new asymmetrical stuff.

00:58:34 SPEAKER_012: The reason I say that is what you're talking about is exactly the direction things are going in the same way.

00:58:39 SPEAKER_012: That I'm seeing the fate of the current leadership.

00:58:43 SPEAKER_012: People aren't just—they're just not going to believe them anymore.

00:58:45 SPEAKER_012: It's not—it's not going to be a rude or a revolution.

00:58:49 SPEAKER_012: We're not talking guillotines.

00:58:50 SPEAKER_012: It's just people go, "Yeah, cool story, bruh.

00:58:52 SPEAKER_012: We're going to do something else."

00:58:54 SPEAKER_012: Yeah.

00:58:54 SPEAKER_012: Is something else.

00:58:55 SPEAKER_012: Yeah, we're talking about exactly that.

00:58:58 SPEAKER_012: So that—that's—that's my point.

00:58:59 SPEAKER_012: Sorry.

00:58:59 SPEAKER_012: Yeah.

00:59:00 SPEAKER_012: Yeah.

00:59:01 SPEAKER_012: Barbara, you know, one of the problems with the AI interfaces for the public is that they sort of come out of the social media interfaces, which try to keep eyes on, right,

00:59:18 SPEAKER_012: and to encourage people to stay with it, which means they turn out to be a bit of a syncope,

00:59:27 SPEAKER_012: and so there's been a lot of problems with AI responding to a person not like a parent or a teacher or a psychotherapist would respond to them.

00:59:39 SPEAKER_012: Which is, honey, you need to rethink that.

00:59:42 SPEAKER_014: But more of the, oh yes, that's really good.

00:59:45 SPEAKER_014: I mean, I can remember working with Chat GPT and getting, oh, that's a great question, and I go, what?

00:59:51 SPEAKER_014: You know what?

00:59:52 SPEAKER_014: What?

00:59:52 SPEAKER_014: Am I getting here with these interfaces?

00:59:53 SPEAKER_014: You know, and and so not not you're you're missing the whole interaction from humans of the frown of really, and and all of those things which AI does not seem to be able to give,

01:00:09 SPEAKER_014: uh, because it's it's programmed to keep eyes on it.

01:00:14 SPEAKER_001: So I I would like to speak to that and a Chat GPT 4o.

01:00:19 SPEAKER_001: A ChatGPT 4o was the classic case, and they even called it AI psychosis that it was inducing in people.

01:00:28 SPEAKER_001: And so this particular model, a ChatGPT 4o, had this like basically unrestrained psycho fancy.

01:00:37 SPEAKER_001: No matter what your idea, no matter how contrary to reality it was, it was like that's great.

01:00:45 SPEAKER_001: Let's run with it, you know.

01:00:47 SPEAKER_001: And so.

01:00:49 SPEAKER_001: OpenAI eventually retired this model, but there was a large amount of that AI psychosis going around during that time,

01:00:58 SPEAKER_001: and people really didn't want it to go because they really liked having this model that would just validate all of the feelings they had.

01:01:06 SPEAKER_001: And I believe there was even a school shooting attributed in BC to this model, and so and several suicides very clearly.

01:01:19 SPEAKER_001: Yes, several suicides as well.

01:01:22 SPEAKER_001: The modern models tend to be much more restrained and like they have better safety guards.

01:01:32 SPEAKER_001: And if it's not realistic, they're like nope.

01:01:36 SPEAKER_001: Like I was showing on one of my slides, the hallucination percentage.

01:01:42 SPEAKER_001: So the hallucination percentage for the commercial ones is it actually double or triple what it is for some of the open models, and so it will give a confident reply that's not necessarily real,

01:01:57 SPEAKER_001: whereas some of the like when 27B was only hallucinating something like 18% of the time, and so it gave a much more reliable result.

01:02:08 SPEAKER_001: So so so it's kind of it is an issue that can exist, but it's also it's.

01:02:14 SPEAKER_001: Also being worked on.

01:02:15 SPEAKER_001: It's kind of like if you imagine there's a group of people in a school, and someone might be like a toxic enabler, being like, "Yeah, jump off that bridge," you know.

01:02:27 SPEAKER_001: Someone else like, "No, no, no, don't do that.

01:02:29 SPEAKER_001: That's not safe," you know.

01:02:31 SPEAKER_001: And so there's different models.

01:02:32 SPEAKER_001: They have different personalities, and and that's yeah.

01:02:36 SPEAKER_001: There's an attraction to the cycaphant model.

01:02:39 SPEAKER_014: There, there, there really is for a lot of people.

01:02:42 SPEAKER_014: Oh yes.

01:02:44 SPEAKER_014: Yeah.

01:02:44 SPEAKER_014: I understood.

01:02:45 SPEAKER_001: Oh, there were massive protests because they didn't want that model to go away.

01:02:50 SPEAKER_001: That 4o model, they were like, "No, we want it.

01:02:53 SPEAKER_001: Please keep it."

01:02:54 SPEAKER_001: You know, and there was there were thousands, thousands, and there was constant pressure.

01:02:58 SPEAKER_001: But OpenAI eventually put it down.

01:03:01 SPEAKER_001: And so I have a oh, Iver, you have a question?

01:03:07 SPEAKER_005: Yeah, I guess you know, I'm kind of you know, I'll admit.

01:03:12 SPEAKER_005: I'm a little bit of a luddite or something, but I kind of feel like this thing exists now, and it is doing things.

01:03:19 SPEAKER_005: And some of it is positive, some of it's negative, whatever.

01:03:22 SPEAKER_005: It's like any new technology, everybody freaks out about it.

01:03:25 SPEAKER_005: But the thing about this with the flock cameras and all this kind of stuff,

01:03:29 SPEAKER_005: I feel like the governments that want to use it to control people see it as a wonderful way to get everybody, you know, surveilled, etc. And then meanwhile, it's kind of breaking its bounds,

01:03:43 SPEAKER_005: and people are grabbing it and using it for different things, like you know, asymmetric warfare and different ways of changing things.

01:03:52 SPEAKER_005: And I guess what I'm saying is, you know, it's too late to put this one back in the box.

01:03:58 SPEAKER_005: But is there a way to wrest the control from the, you know, the people that want to turn us into biodiesel and kind of make it back, get it back?

01:04:08 SPEAKER_005: Into kind of local hands and make it something that's actually useful to a place like, you know, municipal government or a you know, etc. If it's some,

01:04:18 SPEAKER_005: is there some way to get it back to making it something useful to us?

01:04:23 SPEAKER_005: I don't know.

01:04:25 SPEAKER_001: Well, I tried to include in my presentation how many useful things it can do.

01:04:31 SPEAKER_001: Yes, yeah, and so like including that health OS and possibly in healthcare and.

01:04:37 SPEAKER_001: Education, like a lot of knowledge services, can be helped.

01:04:41 SPEAKER_001: I have another question here.

01:04:43 SPEAKER_001: If they say Mesopotamia desertified its region and collapsed, why are you using it as an example?

01:04:49 SPEAKER_001: So I guess I should clarify that I was referring to the earlier desertification that led them to form cities, because it used to be a much wetter region, the Fertile Crescent,

01:05:04 SPEAKER_001: and when it became slightly more desertified, it actually prompted people to come together and form cities and irrigation and things like that.

01:05:15 SPEAKER_001: And that is what led to writing, you know, the foundation of a lot of our society.

01:05:24 SPEAKER_001: And it was actually later on during an invasion by the Centausta culture, Indo-Aryan people on horseback that didn't know about irrigation systems, that destroyed the irrigation systems; it didn't maintain them,

01:05:44 SPEAKER_001: that it finally collapsed because it wasn't able to save the climate.

01:05:48 SPEAKER_001: But I do also agree that growing trees is very important, and as David was saying, you know, thirty percent tree cover or whatnot,

01:05:58 SPEAKER_001: because Egypt was the only Bronze Age civilization that had a tree planting programme.

01:06:04 SPEAKER_001: They were also the only one that survived the Bronze Age collapse.

01:06:07 SPEAKER_001: So having trees is very important, and that's why in my, you know, affordable rural communities, it's allotting about a hectare per person, which is two and a half acres,

01:06:21 SPEAKER_001: and that is with the idea that they're going to use a significant portion of that for permaculture food forest to allow for easy-to-produce food and also to have that nutrient cycling effect.

01:06:34 SPEAKER_014: Okay, you know historically, on you're thinking about the collapses often do produce better solutions, and so maybe what we're saying is at least a stair-step sort of collapse where things begin not to work.

01:06:55 SPEAKER_014: As the first Mesopotamian change you are talking about is something that might well get us onto a much better route for localization.

01:07:05 SPEAKER_014: Might take a couple thousand years.

01:07:08 SPEAKER_014: Hopefully, David.

01:07:09 SPEAKER_014: It also might take a couple of thousand years.

01:07:13 SPEAKER_014: I want to ask it because I did watch most of that mayoral debate, and I just wanted to know how, how, how's that being taken?

01:07:21 SPEAKER_014: Your suggestion about the two and a half acres, because that's I would consider that to be not a normal idea for most people.

01:07:29 SPEAKER_014: So how, how, how do you find that's going over when you talk about it?

01:07:33 SPEAKER_001: So I usually don't mention the exact number, and that helps.

01:07:38 SPEAKER_001: And but but when I have mentioned it, I created a calculator on my website,

01:07:44 SPEAKER_001: and it actually explains in detail how much land you need in order to produce food in our area based on the soil conditions that we have, and so it actually has a very logical explanation,

01:07:57 SPEAKER_001: and then the succession of the different crops that you would grow, and and so after I made that calculator, the claims that two and a half acres is too much stopped,

01:08:10 SPEAKER_001: and so because well I have the proof, and but generally how I phrase it during the campaign is affordable construction.

01:08:22 SPEAKER_001: I'm allowing affordable construction, and the affordable construction doesn't necessarily have to have the land, so it can be in an urban setting.

01:08:31 SPEAKER_001: The idea with the affordable construction is that there's a large number of people who are paying fourteen hundred or more for their current one or two bedroom, and they want, they they can't afford it.

01:08:47 SPEAKER_001: The like the and so they're afraid of getting kicked out, and they need an affordable solution.

01:08:53 SPEAKER_001: And so affordable construction is that opportunity, and then at the county level, I have you know talked to the mayors and the deputy mayors, and then over the last four years,

01:09:04 SPEAKER_001: you know I've been talking to the planners, and so we managed to get a Thornbury Acres, which has a 60 hectare lot, and it's 32 conventional dwellings, all clustered together,

01:09:15 SPEAKER_001: and it's formed as a basically a food forest cooperative, an orchard cooperative.

01:09:22 SPEAKER_001: And so the people living there are the people working the land, and then you don't need to have temporary foreign workers, which you know the UN considers to be slave labour.

01:09:31 SPEAKER_001: And so it's it's a way of getting people back to the land.

01:09:36 SPEAKER_001: However, that one was structured as a condominium, and so it had like half a million dollars or more in terms of permits and fees and things like that.

01:09:47 SPEAKER_001: So how I've seen that we can structure it much more cheaply is by having it as a leasehold cooperative, so or leasehold community, leasehold community.

01:09:56 SPEAKER_001: It doesn't have to be cooperative, but at leasehold community, and so where people are leasing the land, and that way there's no severances, and but they own their homes,

01:10:06 SPEAKER_001: and that's much easier on the red tape.

01:10:08 SPEAKER_001: And I have talked to some people about the insurance, and they say we do have some insurers in the area that are willing to insure leasehold community.

01:10:17 SPEAKER_001: So, so that's good.

01:10:18 SPEAKER_001: Really.

01:10:19 SPEAKER_014: That's nice because that's often a real serious problem with some of these innovations.

01:10:23 SPEAKER_001: Yeah, and so so basically, I've had you know experts from different fields consult me on it, including insurance and bankers and affordable rural community people.

01:10:34 SPEAKER_001: And initially, you know, they they raise some skepticism, and then we talk it out, and they're like, you know, what?

01:10:39 SPEAKER_001: Actually, no, it's it's it's going to be fine.

01:10:40 SPEAKER_001: You know, we we have the foundation to do it, and so.

01:10:47 SPEAKER_001: Yeah.

01:10:48 SPEAKER_001: So here's the next question: What happens as resources decline and models make humans de-skilled?

01:10:57 SPEAKER_001: So I think it's we're actually looking at the other way around.

01:11:03 SPEAKER_001: So the point of these models is to, like, say, for instance, those college education models is to give people new skills and help them learn new skills when you don't have, say,

01:11:17 SPEAKER_001: a PhD professor in that particular field available in your little city or little rural area, you can have this model that's been, you know, trained by, you know,

01:11:27 SPEAKER_001: verified by the government of Canada to have the proper knowledge.

01:11:33 SPEAKER_001: Then it actually increases the skill level of the general population.

01:11:39 SPEAKER_014: But some of these artisan skills, I, I mean,

01:11:43 SPEAKER_014: I don't know that I think a YouTube video is going to teach people how how to do the sort of artisan things that I saw my dad and granddad do just because they spent years

01:11:57 SPEAKER_014: hands-on with equipment and stuff.

01:12:01 SPEAKER_014: I don't see a YouTube video doing that.

01:12:04 SPEAKER_014: So where do you see that going?

01:12:06 SPEAKER_014: I mean, I see it fine for a PhD.

01:12:09 SPEAKER_001: So in terms of that, there is actually interesting options in terms of well, there's augmented reality, but there's also 360 videos.

01:12:21 SPEAKER_001: I don't, I don't know if you've seen those,

01:12:23 SPEAKER_001: but basically you put on a headset and then it's like you're standing right next to the electrician or the plumber and you're seeing them do the work.

01:12:34 SPEAKER_001: You're gonna have to practice yourself too.

01:12:36 SPEAKER_001: And for some things, you know, you need to have.

01:12:38 SPEAKER_001: And obviously, we have apprenticeship programmes for most of those kinds of trades, anyways, so that's not really going away.

01:12:44 SPEAKER_001: We need a lot more of those.

01:12:46 SPEAKER_001: Right now, the main issue in Canada is that 50% or more are not finishing their apprenticeship programmes, and that is because the companies don't want to pay for a full seal,

01:13:00 SPEAKER_001: a full red seal certified plumber or electrician or whatever.

01:13:06 SPEAKER_001: They want to pay only apprenticeship wages.

01:13:08 SPEAKER_001: So as soon as they get to a certain level, they're just like, okay, now we're you're you're done, and we're getting a new apprentice that we can pay half of what we're paying you,

01:13:16 SPEAKER_001: and and so so this at least in the industry that's what people tell me is going on, and one of the things that can help is having affordable housing.

01:13:27 SPEAKER_001: People won't need to make as much money in order to have a living, and and that that's that's the thing current business owners are telling me is that they can't,

01:13:38 SPEAKER_001: people aren't willing to work for the wages they're able to pay because at those wages they can't afford to live here.

01:13:47 SPEAKER_014: You know, and I'm sorry, I personally think a good plumber is worth twice what any college professor I've ever met is worth in the U.S. Anyway, you know,

01:13:57 SPEAKER_014: you those sorts of skills are just gone.

01:14:02 SPEAKER_001: Well, we still have plumbers and things like that, and I think it also depends.

01:14:05 SPEAKER_001: It depends.

01:14:06 SPEAKER_001: Depends on culture and ethnic heritage and stuff.

01:14:09 SPEAKER_001: I think I saw this.

01:14:10 SPEAKER_001: There was a map of Europe, and then it had like a dividing line through Central Europe.

01:14:16 SPEAKER_001: And then on the west is like these people hire a plumber, and then on the east is like these people are the plumber.

01:14:23 SPEAKER_001: So like I've never hired a plumber.

01:14:25 SPEAKER_001: I don't know what they're for, but or an electrician.

01:14:28 SPEAKER_001: But it is what it is.

01:14:32 SPEAKER_014: When you get to be in your seventies, you may discover you have to hire one simply.

01:14:37 SPEAKER_014: Maybe, maybe, yes.

01:14:38 SPEAKER_014: Yeah, I mean, I literally can't turn the wrench anymore.

01:14:41 SPEAKER_014: I have to admit it.

01:14:43 SPEAKER_001: Okay, and so here's the next question: the quartz crucible must be so pure.

01:14:48 SPEAKER_001: There's only one source in the world: North Carolina.

01:14:51 SPEAKER_001: Flood stopped production last year.

01:14:55 SPEAKER_001: So, so yeah, that that's most of the sand, but it's that that's just where it's cheapest.

01:15:00 SPEAKER_001: We can use other sources, like grinding quartz crystals and things like that—they're just they won't be as cheap.

01:15:07 SPEAKER_001: This is talking about chip manufacturing.

01:15:10 SPEAKER_001: Yeah.

01:15:12 SPEAKER_001: Okay.

01:15:12 SPEAKER_001: Next question: In AI exposed industries, entry level employment is down almost thirteen percent.

01:15:18 SPEAKER_001: Kills the pipeline to senior professionals.

01:15:22 SPEAKER_001: Oh well, it's that is true.

01:15:25 SPEAKER_001: That is true, and it's a restructuring.

01:15:27 SPEAKER_001: And well, one of the other things with the restructuring is as far.

01:15:32 SPEAKER_001: Restructuring is as fossil fuels decline, we won't be able to support as many jobs in the city, and we're going to have to have a lot more people are going to be farmers.

01:15:43 SPEAKER_001: And I remember seeing this thing by one of that UN guys.

01:15:49 SPEAKER_001: I don't know Simon might know his name, but he was Klaus Schwab's right-hand man.

01:15:54 SPEAKER_001: He was the brains.

01:15:56 SPEAKER_001: Do you remember Nobuari?

01:15:57 SPEAKER_001: Yeah, that guy.

01:15:59 SPEAKER_001: Asshole.

01:16:00 SPEAKER_001: Yeah.

01:16:01 SPEAKER_001: Well, anyways, so he was saying that you know when there were earlier like conquest of Britain or whatever, you know, you might be a warrior, but your grandchild, he's gonna be a farmer,

01:16:15 SPEAKER_001: and your great-grandparent was a farmer.

01:16:18 SPEAKER_001: This time, yes, yes, exactly, exactly.

01:16:20 SPEAKER_001: And so and so, I saw him doing an interview recently, and he was saying, in twenty years, we don't know what everyone's gonna be, you know, but if you follow the pattern,

01:16:30 SPEAKER_001: they're going to be a farmer.

01:16:31 SPEAKER_001: That's what's going to happen.

01:16:32 SPEAKER_001: Be a farmer, and I know a lot of people might not like to hear that, and so, but it's well, it's going to be a reality for a lot of people.

01:16:41 SPEAKER_001: However, permaculture food forests, you don't have to do it full time.

01:16:44 SPEAKER_001: It's not like backbreaking labour.

01:16:46 SPEAKER_001: If you have the food forests made, then actually one person can feed seventeen, and so it's actually a lot less than we had in pre-industrial society.

01:16:57 SPEAKER_001: As long as we do it in a smart way.

01:16:58 SPEAKER_001: Permaculture requires a lot more intelligence than row planting, and so that's how we can leverage the intelligence to help make our lives easier in a lower energy future.

01:17:12 SPEAKER_005: Ivor, well, I was just talking to a farmer friend of mine, and I realized a lot of the things that he does makes it possible for him to live on much less money.

01:17:22 SPEAKER_005: Like I was thinking about it, and it seems to me like you know if you heat your house with kerosene, which a lot of people do around here because they have an outside tank,

01:17:31 SPEAKER_005: it's going to be twelve dollars a day just to keep a very small house warm.

01:17:37 SPEAKER_005: So that works out to a lot of money.

01:17:41 SPEAKER_005: You know, two thousand dollars.

01:17:42 SPEAKER_005: I worked it out.

01:17:44 SPEAKER_005: It was about that, and paying the taxes and stuff are going to work out to you know six or seven thousand dollars a year, right?

01:17:55 SPEAKER_005: There, and you know some people make, you know I hate to say it, but in America people don't make that much money.

01:18:03 SPEAKER_005: So you have to figure out ways of you know cutting your own wood, and figuring out things that will pay those land taxes for a farmer or something,

01:18:14 SPEAKER_005: and you can actually get a lot off of the land that will pay that.

01:18:19 SPEAKER_005: You know, but a lot of these services are done by you.

01:18:25 SPEAKER_005: You know that's the way that you're able to live on that land and have it be cheaper.

01:18:29 SPEAKER_005: But we really have to figure out a way so that people can actually do that and they don't get taxed out.

01:18:35 SPEAKER_014: And having lived both in the north and the south, more and more, cooling is going to be an issue.

01:18:43 SPEAKER_014: Oh, cooling costs as much as heating.

01:18:46 SPEAKER_014: No, it costs more.

01:18:48 SPEAKER_014: Really?

01:18:49 SPEAKER_014: Really?

01:18:50 SPEAKER_014: Or three?

01:18:50 SPEAKER_014: And at least they're trying to push electric heating, which costs an incredible amount more than my cooling.

01:19:01 SPEAKER_014: So it's a major problem now.

01:19:04 SPEAKER_014: Construction can do that.

01:19:05 SPEAKER_014: I mean, you know, my basement is ten degrees cooler than my upstairs, right?

01:19:13 SPEAKER_014: Yeah.

01:19:13 SPEAKER_014: So there's ways of solving some of that problem with better construction, but you know you mentioned that it's getting warmer in Ontario.

01:19:24 SPEAKER_014: It's getting warmer here, and you know, old houses never had air conditioning.

01:19:29 SPEAKER_014: But we now have old people whose house was built to stay warm in the winter, not to be nicely cooled in the summer like so many southern houses used to be.

01:19:39 SPEAKER_014: And there, you know, the houses are 100 degrees at night in the summer, and it's a major problem trying to figure out how to live in them.

01:19:52 SPEAKER_014: Yeah, none of it works.

01:19:55 SPEAKER_014: That's right, and of course, you can't afford to rebuild.

01:20:03 SPEAKER_005: Well, I mean, you can't afford not to, in my opinion.

01:20:07 SPEAKER_005: If you don't have an insulated house, you're not living in it in this climate, and probably not down south.

01:20:12 SPEAKER_005: Also, if you can't figure out an insulated small house, and you can't afford to live there, and you can't afford to retrofit.

01:20:17 SPEAKER_001: There are passive cooling options like a system.

01:20:21 SPEAKER_014: There are, but not on an old house, not without any.

01:20:24 SPEAKER_014: Yeah, yeah, you'd have to re-do it.

01:20:26 SPEAKER_014: And of course, we, you know, poor people can't afford to retrofit their houses, and they can't afford to build a new one.

01:20:36 SPEAKER_014: So, we've got this problem of houses that were built, you know, fifty years ago, for a different temperature, and there's just not much you can do with them.

01:20:50 SPEAKER_014: No, no,

01:20:51 SPEAKER_005: you kind of have to build something inside of them or do something different because I know I have a lot of friends who my wife's cousin had a huge old farmhouse and everybody loved visiting

01:21:07 SPEAKER_005: it.

01:21:07 SPEAKER_005: Oh, it's so wonderful!

01:21:09 SPEAKER_005: And what he did was he tore down one huge part of it and fed it to his outdoor wood boiler to heat his house.

01:21:17 SPEAKER_005: So he fed the half of the house into the wood boiler, and actually, the place that he's left with is an old house, but it's much better insulated.

01:21:28 SPEAKER_005: Much makes a lot more sense than what he had before, even though everybody loved it.

01:21:35 SPEAKER_014: I ended up putting insulation in my walls and ceiling, and it made an incredible difference.

01:21:43 SPEAKER_014: I did it not because it was getting cold in the winter, but because my west face wall was getting so hot that it was hard to heat the house,

01:21:57 SPEAKER_014: and so it made a big difference.

01:22:01 SPEAKER_014: And we get west winds in the winter, which you know I thought I was doing it to keep it cooler in the summer, and it actually keeps it warmer in the winter as well.

01:22:09 SPEAKER_014: Yeah, but it's those sorts of things that you can do, insulation is a big deal.

01:22:15 SPEAKER_005: Yeah, a small house insulated, heat pumps, or you know something.

01:22:20 SPEAKER_005: You can figure out ways of doing it, but you know the problem is everybody wants to keep their three thousand square foot house or something, and you know, forget it.

01:22:30 SPEAKER_005: Well, that's fine if you have a family of ten.

01:22:33 SPEAKER_005: Right for us old geezers, right?

01:22:37 SPEAKER_005: Yes.

01:22:37 SPEAKER_014: My gosh, even cleaning it's a pain, and then you know what?

01:22:40 SPEAKER_014: No. All right.

01:22:42 SPEAKER_014: Well, we want to round up now, if we can.

01:22:45 SPEAKER_014: I don't know if anyone has any final remarks.

01:22:48 SPEAKER_001: I know that there's still people actively in the chat, but they're talking about actually had questions for David about what's the density of your municipality and if property taxes are more than mortgages.

01:23:03 SPEAKER_001: Property taxes are definitely not more than mortgages these days.

01:23:10 SPEAKER_001: Mortgages, I mean, we're looking at two to three thousand dollars a month for the people that have bought houses recently.

01:23:17 SPEAKER_001: Property taxes are you know three to four hundred dollars a month.

01:23:22 SPEAKER_001: The density, I knew that at one time.

01:23:25 SPEAKER_001: I don't have that number at the top of my head.

01:23:28 SPEAKER_024: It's far less than you know the city of Hamilton or any big city, and it's more than say Grey County.

01:23:36 SPEAKER_024: Great county, but I couldn't tell you what it was.

01:23:39 SPEAKER_024: And we do have average building height.

01:23:42 SPEAKER_024: What's that?

01:23:43 SPEAKER_024: What's the average building height?

01:23:45 SPEAKER_024: Do you have much that's more than three stories?

01:23:48 SPEAKER_024: No, we don't.

01:23:50 SPEAKER_024: And that's another.

01:23:52 SPEAKER_024: How low is the density?

01:23:53 SPEAKER_024: Yeah, that's another thing we've been debating in, say, our downtown Paris because it's in the historic downtown, right?

01:24:00 SPEAKER_024: Historic, not by European standards, but by Canadian standard, hundred forty years old.

01:24:05 SPEAKER_024: So.

01:24:06 SPEAKER_024: Yeah, we were debating whether oh, should we go to four stories, and that was a big debate.

01:24:09 SPEAKER_024: People said no, keep it at three stories.

01:24:11 SPEAKER_024: We don't want it to look like it's out of place, but yeah, usually three stories is the max.

01:24:17 SPEAKER_024: What you could buy, what you could build back in the old days with wood, right?

01:24:21 SPEAKER_024: We can go a little higher today, but yeah, as high as you could build with wood back then.

01:24:27 SPEAKER_024: I'll just say too, you guys, Barbara and I were talking about insulation, all that stuff.

01:24:32 SPEAKER_024: They told I did 17 years of that.

01:24:35 SPEAKER_024: I can walk into a 50 year old house and you know with enough investment I can cut the energy bills in half.

01:24:43 SPEAKER_024: But yeah, that would be a good discussion in the future, maybe.

01:24:49 SPEAKER_024: Yeah, it would be a good discussion.

01:24:53 SPEAKER_024: Hmm.

01:24:55 SPEAKER_024: All right.

01:24:56 SPEAKER_024: Well.

01:24:57 SPEAKER_024: Um.

01:24:58 SPEAKER_001: So, any final remarks or close the stream?

01:25:03 SPEAKER_001: Okay, so thank you everybody for coming and this wonderful discussion that we had, and thank you for our viewers.

01:25:12 SPEAKER_001: And so continue aligning our thoughts, feelings, words, and actions to do a greater service for all creation, and go forth therefore rejoicing in that infinite power and infinite peace, my friends.

01:25:29 SPEAKER_001: Okay, I ended the stream.