
Sep 14, 2026 · 1h 11m
Listen from 23:13
Listen at 23:13
AI oversight meets meme coins and improvised data centers
Ask Jason: AI Extinction, Grok's Rogue Meme Coin & Weed Farms as Data Centers | E2337
The discussion tests where human responsibility, regulation, and infrastructure must catch up with increasingly capable AI systems.
- 1AI risks look more immediate in job disruption, inequality, and unrest than in cinematic extinction scenarios.
- 2Autonomous agents raise unresolved questions about permission, platform liability, and human control over public-facing actions.
- 3Underused properties could become distributed compute sites if electricity, cooling, capital, and regulation align.
Don't miss
The hosts explore converting unused cannabis cultivation facilities and other stranded properties into distributed data centers.
The brief
The hosts separate plausible AI harms—job displacement, inequality, and unrest—from claims that recursive self-improvement will imminently eliminate humanity.
A bot-to-bot meme-coin incident sharpens the accountability problem: permissionless systems can act publicly, but platforms and humans still face liability.
The conversation favors an independent AI ecosystem built from open models, local compute, and smaller providers rather than a few centralized gatekeepers.
For enterprises, human review should track compliance, security, liability, and data sensitivity; higher-stakes systems require stronger controls.
The standout proposal is practical and strange: convert cannabis facilities, dead malls, warehouses, and other underused properties into distributed data centers.
The episode closes by applying the same experimental mindset to founders, research, and infrastructure: test tools, build unusual products, and follow available capacity.
What was said on this episode
44 statements · 21 positive · 18 negative · 1 mixed · 4 neutral
AI may cause societal problems without causing human extinction.
“Those kind of things are not cataclysmic, everybody's going to die. Those are under the category of like the social fabric is destroyed in our country, in our, in our world, societal problems. There could be societal problems.”
Listen at 3:27
Millions of drivers could lose their jobs within ten years.
“we could see millions of drivers lose their jobs in under 10 years.”
Listen at 3:43
Properly designed AI products can be safer than non-AI products.
“AI products, if designed correctly, should be safer than non-AI products and can be.”
Listen at 9:15
Properly designed AI products can be safer than non-AI products.
“AI products, if designed correctly, should be safer than non-AI products and can be.”
Listen at 9:15
AI should primarily protect society from people with harmful intent.
“AI should be used first and foremost to protect against humans with bad intent, AKA terrorists, school shooters, mentally ill people, people in society with screws loose, or, you know, who are religious zealots, or they're hearing voices.”
Listen at 10:57
Crypto projects accepting public money should require two or three verified people.
“I believe the way the regulation should work is that at least 2 people, maybe 3, need to put their driver's license in”
Listen at 15:47
Crypto projects should require identification from at least two or three people.
“at least 2 people, maybe 3, need to put their driver's license in”
Listen at 15:51
Meme coins should be regulated in the United States.
“These meme coins, at least in the United States, should be regulated in my mind.”
Listen at 16:38
Large U.S. meme coins should face stringent regulation because AI scaling is dangerous.
“Everything that involves these crypto coins, at least in the United States, when they hit a certain size, uh, they should have like super regulation on them because it is too dangerous to allow people to create financial products and then trick AIs to do it at scale.”
Listen at 16:54
Investors should expect Genesis Coin and similar tokens to become worthless.
“your expectation should be it goes to zero.”
Listen at 18:43
Investors should expect Genesis Coin and similar tokens to reach zero.
“your expectation should be it goes to zero.”
Listen at 18:43
Open-source AI products may lag frontier models by six months.
“the open source, uh, products are 6 months behind maybe the frontier models.”
Listen at 22:00
Open-source AI products may lag frontier models by approximately six months.
“the open source, uh, products are 6 months behind maybe the frontier models.”
Listen at 22:00
Perplexity Computer and OpenCLaw will become installable on Mac Studio with one click.
“You're going to be able to use Perplexity Computer and OpenCLaw, one click, have it on your Mac Studio.”
Listen at 23:13
Local AI hardware costs will halve while computing power doubles every two years.
“it'll go down 50% every 2 years in terms of cost, and the power will go up double every 2 years.”
Listen at 23:20
Local AI hardware costs will halve while its power doubles every two years.
“it'll go down 50% every 2 years in terms of cost, and the power will go up double every 2 years.”
Listen at 23:20
AI agents should be evaluated by whether they perform tasks better, faster, or cheaper.
“do the bots do it better, quicker, faster, cheaper, essentially?”
Listen at 27:11
Basic AI users are probably twenty percent more effective.
“they're probably 20% more effective”
Listen at 32:02
Twenty-person companies can gain productivity equivalent to four employees annually.
“it's like getting 4 new employees for free every year.”
Listen at 32:20
Higher-value enterprise data requires stronger human oversight controls.
“The more important the data you have is, the more important it is to have these controls in place, to have a human in the loop.”
Listen at 36:01
Higher-stakes enterprise transactions require more human oversight.
“depending on the transaction and what's at stake, that's when you need to have more humans in the loop”
Listen at 36:43
Jason prefers limiting his company to 25 employees.
“25 is the limit at this point in my career based on me wanting to enjoy my life.”
Listen at 41:14
Employees who damage the company or violate policies should lose stock options.
“if you damage the company, If you break these policies, you lose your stock options.”
Listen at 42:05
Employees who damage a company or violate policies should lose stock options.
“if you damage the company, If you break these policies, you lose your stock options.”
Listen at 42:05
Properly implemented AI will perform nearly every profession as well as humans.
“if properly implemented, will do as good a job as a human almost universally.”
Listen at 49:15
Properly implemented AI will perform as well as humans across nearly every profession.
“if properly implemented, will do as good a job as a human almost universally.”
Listen at 49:15
A sufficiently optimized AI could outperform the average surgeon within a year.
“If you spent enough time, you know, one year just making it perfect, it's going to beat the average surgeon.”
Listen at 49:24
A sufficiently refined AI system will outperform the average surgeon.
“If you spent enough time, you know, one year just making it perfect, it's going to beat the average surgeon.”
Listen at 49:24
Researchers should avoid frontier AI models for sensitive research.
“I do think anybody who's doing this kind of research should not be using the frontier models.”
Listen at 50:06
Researchers should avoid frontier AI models for sensitive research.
“anybody who's doing this kind of research should not be using the frontier models.”
Listen at 50:07
Researchers’ use of frontier models will collectively improve OpenAI and Anthropic models.
“they're gonna in aggregate train OpenAI and Anthropic's models to solve these problems”
Listen at 51:24
OpenAI mishandled its response to researchers using its models.
“For sure, it was a PR blunder.”
Listen at 56:17
A successful startup product could attract one million annual customers paying $100.
“if it did work and a million people used it, they'd pay $100 a year for it.”
Listen at 57:55
A startup reaching six execution steps could generate $100 million in revenue.
“can you imagine if you got like 6 steps done, that $100 million in revenue come, then you're in the venture game.”
Listen at 58:10
Elon will have substantial computing capacity in space within three years.
“in 3 years, Elon will start having like serious amounts of compute in space”
Listen at 1:01:41
Elon will demonstrate substantial space-based computing within approximately one year.
“In 3 years, Elon will start having like serious amounts of compute in space, and he'll demonstrate it probably next year is my guess.”
Listen at 1:01:41
Elon will probably demonstrate space-based compute next year.
“he'll demonstrate it probably next year is my guess.”
Listen at 1:01:47
The United States may lose much of its data-center business to Gulf monarchies.
“don't be surprised if like the U.S. winds up giving this entire data center business over to the, the Gulf monarchies.”
Listen at 1:05:01
The United States could lose the data-center industry to Gulf monarchies.
“the U.S. winds up giving this entire data center business over to the, the Gulf monarchies.”
Listen at 1:05:02
Political discussion about AI data centers will change after the midterm elections.
“This whole thing, the whole dialogue is going to change, Mo, after the midterms.”
Listen at 1:06:19
Data-center construction will accelerate for 24 months before efficiency technologies solve constraints.
“The build-out's going to be fast and furious for another 24 months, but at some point there's so much energy and development put into inference chips, efficiency, photonics, energy, nuclear energy, that these problems will get solved.”
Listen at 1:07:59
Data-center construction will remain rapid for another twenty-four months.
“The build-out's going to be fast and furious for another 24 months”
Listen at 1:07:59
AI data-center capacity will eventually become overbuilt.
“I don't know when we hit overbuilt, but it's inevitable that we hit overbuilt.”
Listen at 1:08:18
Data-center infrastructure will eventually become overbuilt.
“it's inevitable that we hit overbuilt.”
Listen at 1:08:21
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Books & mentions
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.
