Pinned Tweet
Twitter is not a task it's a documented story of your life so others can consider your point of view ...and choose whether to engage or not
242
2,039
3,259
Thomas Power retweeted
If anyone’s Tab is slow or double texting, we’ve had a demand surge of over 100% since last night. We are scaling up infra to keep up with all of you 🤝 🫶
18
2
46
2,147
Thomas Power retweeted
The real winners of the Anthropic IPO:
Google and Amazon.
They sold Anthropic equipment and compute in exchange for stock and convertable notes so Anthropic could look more profitable than they are. They also invested $26B in cash.
They own 26% of the company.
Will be worth $530B+ at IPO.
Its a giant pump and dump scheme!
Google and Amazon.
They sold Anthropic equipment and compute in exchange for stock and convertable notes so Anthropic could look more profitable than they are. They also invested $26B in cash.
They own 26% of the company.
Will be worth $530B+ at IPO.
Its a giant pump and dump scheme!
My predictions over the following months:
1. Stock market will get hurt bad when the music stops on data centers, models, and chips.
Too much froth, too high of investment, too many companies will fail. The switching costs are low. I don't know how these companies will become profitable. Its a mega bubble.
The rest of the economy is already in a deep recession. Real pain will set in when the public markets dive.
2. Power is the bottleneck.
AI is burning an absurd amount of power. The cost of electricity is skyrocketing which is making the cost of everyday life more expensive and hurting quality of life. I'm against data centers because of the power, not the water.
The grid is extremely fragile and the weakness will be exposed in the coming months.
3. Subsidies are skewing the market.
AI isn't cheap. The prices to use it are artificially low because of subsidies all the way up the chain.
You pay $20 or $200 a month but the true costs are 10x That spread is venture capital money being incinerated at scale.
When they raise the price, the users will disappear and investors will hold the bag.
3. A big security breach is coming.
A bank, a certain model, something is going to rock our world.
People are letting bots log in to their banks, their email and everything else. Big trouble coming.
4. Grokbot is exceptional.
Its the first model that is adding significant value to my everyday life. I don't know why anybody would use a different personal assistant.
5. AI is insanely high maintenance
Billions are being poured into training the next model. They day it comes out its already behind. China has cheaper electricity and more chips.
The chips they bought will be obsolete in 18 months.
Companies are stuck paying big money every month to maintain and update all of their AI tools. Its a mess.
6. The power companies will renege on their deals.
Political pressure will unwind all the unsustainable power deals these data centers and LLMs have negotiated.
Power costs will explode. Then comes the choice:
Do we jack up costs for our AI / compute and watch everybody switch to the next model or do we incinerate more VC capital? Nobody wins.
7. Credit will be the first domino to fall.
Data centers are already struggling to finance the GPUs over 10 years when they will be obsolete in 18 months. We'll end up with a shit load of half built data centers who can't negotiate power rates or buy chips because nobody will let them borrow money.
The money will get cold feet as soon as the markets tank and the show will be over.
8. It'll be a dark, dark period in economic markets.
The people who own assets that make real money (real estate or businesses) will win.
The people who plan ahead and avoid massive leverage will win.
The people with cash on their balance sheet will accumulate more assets and generate exceptional wealth.
--
Don't listen to me. I'm just a bear. Not investment advice. Good luck!
49
52
506
53,705
Thomas Power retweeted
In 2018 Revolut was opening 7,000-8,000 accounts a day with zero marketing.
Storonsky's secret: 10x.
If you want a product to sell in huge quantities with zero marketing, it has to be ten times better and ten times cheaper. As simple as that.
For Revolut that meant:
1. An es from the app
2. Send, borrow, exchange, buy crypto and invest in one place
3. The interbank exchange rate while banks charged 3-7%
65-70% of Revolut's users came through word of mouth.
Full breakdown here:
productmarketfit.tech/p/revolut-play…
Storonsky's secret: 10x.
If you want a product to sell in huge quantities with zero marketing, it has to be ten times better and ten times cheaper. As simple as that.
For Revolut that meant:
1. An es from the app
2. Send, borrow, exchange, buy crypto and invest in one place
3. The interbank exchange rate while banks charged 3-7%
65-70% of Revolut's users came through word of mouth.
Full breakdown here:
productmarketfit.tech/p/revolut-play…
2
9
93
10,106
Thomas Power retweeted
Banks hire McKinsey to write their digital strategy.
A year later they get a report. By then it's already out of date.
Storonsky's diagnosis is brutal — and accurate:
Legacy systems. Legacy culture. Senior managers who have never built tech and can't set the direction.
That ller. You can replace infrastructure. You can't easily replace a leadership layer that fundamentally doesn't understand what it's managing.
Traditional banking consultants sell frameworks. Fintech founders ship products. The gap between those two velocities is why incumbents keep losing ground — not because they lack the strategy document, but because the org can't execute faster than the strategy expires.
Full breakdown here:
productmarketfit.tech/p/revolut-play…
A year later they get a report. By then it's already out of date.
Storonsky's diagnosis is brutal — and accurate:
Legacy systems. Legacy culture. Senior managers who have never built tech and can't set the direction.
That ller. You can replace infrastructure. You can't easily replace a leadership layer that fundamentally doesn't understand what it's managing.
Traditional banking consultants sell frameworks. Fintech founders ship products. The gap between those two velocities is why incumbents keep losing ground — not because they lack the strategy document, but because the org can't execute faster than the strategy expires.
Full breakdown here:
productmarketfit.tech/p/revolut-play…
5
15
136
12,037
Thomas Power retweeted
Jane Street made $39.6 BILLION last year by PRICING things, not predicting them.
It never stops being the HOUSE.
This f*cking prompt turns Grok Bot into a JANE STREET desk.
Paste it in before bed: 3 bots price your whole watchlist overnight, and by 7:30 you get only the HOUSE, not the GAMBLER.
You might actually get RICH, because you stop GUESSING and start PRICING.
Steal it before tomorrow's open:
It never stops being the HOUSE.
This f*cking prompt turns Grok Bot into a JANE STREET desk.
Paste it in before bed: 3 bots price your whole watchlist overnight, and by 7:30 you get only the HOUSE, not the GAMBLER.
You might actually get RICH, because you stop GUESSING and start PRICING.
Steal it before tomorrow's open:
how do you know a trading strategy still works?
you don’t. unless you test it.
this guide turns your AI into a research desk that does it for you: real data, real backtests, stress tests, and a path to your broker x.com/i/article/2110…
3
7
39
3,660
Thomas Power retweeted
Drop this into Grok Bot before bed and let Jev pick your next viral post while you sleep...
Grok Bot writes 24 hooks.
Jev scores all 24 in parallel.
The winner did 1.24M views.
Here's the full setup:
1. Grok Bot drafts.
Writes 24 hooks on one topic in under 20 seconds.
2. Jev every draft 1 to 10 on hook strength.
3. Jev checks bookmarks.
Answers yes or no on whether people would save it.
4. Jev checks your voice.
Scores how close each draft sounds to your best posts.
5. Jev ranks the top 3.
21 drafts get cut, 3 make it to the final.
6. Grok Bot ships the winner.
The 9.6 hook went out and pulled 18.1K likes and 9.6K saves.
24 drafts, 1 winner, 38 seconds.
Hit views like this every week and $10k a month from X is on the table.
Try this setup, then read the full breakdown in the article below
Grok Bot writes 24 hooks.
Jev scores all 24 in parallel.
The winner did 1.24M views.
Here's the full setup:
1. Grok Bot drafts.
Writes 24 hooks on one topic in under 20 seconds.
2. Jev every draft 1 to 10 on hook strength.
3. Jev checks bookmarks.
Answers yes or no on whether people would save it.
4. Jev checks your voice.
Scores how close each draft sounds to your best posts.
5. Jev ranks the top 3.
21 drafts get cut, 3 make it to the final.
6. Grok Bot ships the winner.
The 9.6 hook went out and pulled 18.1K likes and 9.6K saves.
24 drafts, 1 winner, 38 seconds.
Hit views like this every week and $10k a month from X is on the table.
Try this setup, then read the full breakdown in the article below
12
7
82
15,732
Thomas Power retweeted
This is f*cking insane. Elon Musk named 12 ways Grok Bot pays for itself, and they're buried across 2 months of his posts
So I pulled all 12 into one list, with the exact line to send for each:
→ Taxes: "Find every refund I missed in the last 3 years"
→ Credit score: "Find e down and fix it"
→ Spam calls: "Get my number off every spam list"
→ Scam emails: "Flag anything that asks for money or a password"
→ Contracts: "Read this before I sign it and tell me what hurts me"
→ Tesla: "Read my unread email and move my 3pm" (out loud, while driving)
→ iPhone: hold the Action Button and say "Handle this", no app to open
→ Restaurants: "Book a table for 4 at 8 and find the cheaper place"
→ Deliveries: "Track every package and chase the late ones"
→ Mornings: "Print my plan for tomorrow"
→ Your store: "Answer every support ticket today"
→ Your inbox: "Run it like a $60K assistant would"
Most people will still do their taxes by hand this year
Send it the tax line first
Then read how a whole Grok Bot team runs on top of these in the article below ↓
So I pulled all 12 into one list, with the exact line to send for each:
→ Taxes: "Find every refund I missed in the last 3 years"
→ Credit score: "Find e down and fix it"
→ Spam calls: "Get my number off every spam list"
→ Scam emails: "Flag anything that asks for money or a password"
→ Contracts: "Read this before I sign it and tell me what hurts me"
→ Tesla: "Read my unread email and move my 3pm" (out loud, while driving)
→ iPhone: hold the Action Button and say "Handle this", no app to open
→ Restaurants: "Book a table for 4 at 8 and find the cheaper place"
→ Deliveries: "Track every package and chase the late ones"
→ Mornings: "Print my plan for tomorrow"
→ Your store: "Answer every support ticket today"
→ Your inbox: "Run it like a $60K assistant would"
Most people will still do their taxes by hand this year
Send it the tax line first
Then read how a whole Grok Bot team runs on top of these in the article below ↓
16
55
421
34,402
Thomas Power retweeted
Holy f*ck, Elon Musk's team at spacexai just dropped a full spam killer inside Grok Bot, and it's free
DeleteMe charges $129 a year to pull your number off spam lists
Grok Bot does it for $0
It finds your number on people-search sites, files every opt-out, clicks the checks again every month
Add the free Spam-Call Scrubber skill, then say these 3 lines once:
1. "Find every site that lists my phone number and home address"
2. "File the opt-out on each one and finish the confirmation emails"
3. "Check again on the 1st of every month and tell me what came back"
Say line 1 tonight
Then read how to wire it into a team that never sleeps in the article below ↓
DeleteMe charges $129 a year to pull your number off spam lists
Grok Bot does it for $0
It finds your number on people-search sites, files every opt-out, clicks the checks again every month
Add the free Spam-Call Scrubber skill, then say these 3 lines once:
1. "Find every site that lists my phone number and home address"
2. "File the opt-out on each one and finish the confirmation emails"
3. "Check again on the 1st of every month and tell me what came back"
Say line 1 tonight
Then read how to wire it into a team that never sleeps in the article below ↓
Grok Bot could be the operating layer AI agents have been missing.
It can use tools, run routines and coordinate other agents with minimal supervision.
In this article, I show you how. x.com/i/article/2101…
518
1,801
12,876
1,260,732
Thomas Power retweeted
Dario Amodei: some tokens Claude produces are worth a few cents, others could be worth tens of millions of dollars.
His examples:
1. Telling someone to restart their Mac: the model has said it 10 million times, worth a dollar or less
2. Telling a pharma company to move an her end of a molecule: potentially tens of millions
3. Expect pay-for-results pricing, or labor-like pricing by the hour
4. It's a new industry, so many models will be tried
Today pricing is simpler: per the S-1, usage-based pricing was 80% of H1 2026 revenue and subscriptions 20%.
Full breakdown here:
theaiopportunities.com/p/anthropics-s…
His examples:
1. Telling someone to restart their Mac: the model has said it 10 million times, worth a dollar or less
2. Telling a pharma company to move an her end of a molecule: potentially tens of millions
3. Expect pay-for-results pricing, or labor-like pricing by the hour
4. It's a new industry, so many models will be tried
Today pricing is simpler: per the S-1, usage-based pricing was 80% of H1 2026 revenue and subscriptions 20%.
Full breakdown here:
theaiopportunities.com/p/anthropics-s…
366
106
2,632
1,127,209
Thomas Power retweeted
🚨BREAKING NEWS : Your "deleted" photos aren't deleted.
Claude Code found years of them sitting in a hidden cache on my iPhone.
I plugged my iPhone into my MacBook to run a backup. I had Claude Code doing a security check at the same time. It flagged something I wasn't looking PhotoData/Thumbnails.
Thousands of files. Photos, videos, screenshots. A lot of them were things I had deleted.
Photos from a couple of years ago. Screenshots I took and deleted a few seconds later. Stuff I went into Recently Deleted and cleared. Some of it was still sitting there.
What I found in that cache:
- old screenshots of banking apps
- photos I deleted right after taking them
- random pictures I didn't want on my phone anymore
- a few things I specifically remember deleting more than once
When you delete a photo on iPhone, iOS removes it from your camera roll and Recently Deleted folder. But it keeps a compressed copy in a Spotlight search cache so your phone can still find things when you search. That cache doesn't get fully cleared.
I checked which apps had Files and Folders access. There were more than I expected, including apps I downloaded a long time ago and forgot about. Any of them could potentially read files from that cache.
I went through the list and removed access from most of them.
The prompt I used is below. It doesn't take long if you already have a backup.
How many apps on your phone have Files access right now?
Claude Code found years of them sitting in a hidden cache on my iPhone.
I plugged my iPhone into my MacBook to run a backup. I had Claude Code doing a security check at the same time. It flagged something I wasn't looking PhotoData/Thumbnails.
Thousands of files. Photos, videos, screenshots. A lot of them were things I had deleted.
Photos from a couple of years ago. Screenshots I took and deleted a few seconds later. Stuff I went into Recently Deleted and cleared. Some of it was still sitting there.
What I found in that cache:
- old screenshots of banking apps
- photos I deleted right after taking them
- random pictures I didn't want on my phone anymore
- a few things I specifically remember deleting more than once
When you delete a photo on iPhone, iOS removes it from your camera roll and Recently Deleted folder. But it keeps a compressed copy in a Spotlight search cache so your phone can still find things when you search. That cache doesn't get fully cleared.
I checked which apps had Files and Folders access. There were more than I expected, including apps I downloaded a long time ago and forgot about. Any of them could potentially read files from that cache.
I went through the list and removed access from most of them.
The prompt I used is below. It doesn't take long if you already have a backup.
How many apps on your phone have Files access right now?
18
82
341
39,410
Thomas Power retweeted
Elon Musk now literally holds the off switch on Anthropic's $2 trillion IPO.
Anthropic's filing shows it has agreed to pay Elon's AI company up to $120.5 billion for computing power. And starting early next year, either side can walk away with 90 days' notice.
The deal is with it that used to be xAI, and it rents Anthropic the Nvidia chips that run Claude.
In May, the price was $1.25 billion a month for one data center in Memphis. Reuters reported the confidential draft of this filing put the total at $84.5 billion. The public version filed today says $120.5 billion.
Once all the capacity arrives in December, the bill jumps: Anthropic expects to pay Musk about $3.1 billion every month.
Anthropic reportedly brought in $16.8 billion last quarter, which works out to about $5.6 billion a month. So more than HALF of what Anthropic earns today is about to go to one supplier.
Now here's where it gets really insane:
The lock-in on those contracts ends in January and March 2027. After that, either side can cancel with 90 days' notice. Fortune reported in July that the clause gives SpaceX a way to take its chips back if Grok suddenly needs more.
Grok competes directly with Claude.
And in February, Elon publicly called Anthropic's AI evil.
So the company selling itself to the public as the safe AI lab is about to run a huge share of its product on hardware owned by its loudest critic. That critic also owns a direct competitor, and he can start the 90-day clock in early 2027.
The filing does warn that some of Anthropic's compute suppliers build their own AI models, and that warning reads very differently once the supplier is Elon.
And Elon needs this deal just as badly: SpaceX's CFO has promised $100 billion in annual revenue by the end of this year. At full speed, Anthropic's rent comes to about $9.3 billion a quarter. That's more than the roughly $7.6 billion SpaceX's entire AI business brought in last quarter.
So 2 of the biggest IPOs in history are now holding each other up, and both of them have a 90-day exit.
Anthropic has a real argument here. Computing power is the scarcest thing in AI right now, and Amazon's cloud capacity for 2027 is reportedly largely sold out. The filing even admits Anthropic badly underestimated how much compute it needed in the past. Grabbing every chip on the market is the rational move, and the same clause lets Anthropic leave if cheaper capacity shows up.
But public buyers deserve 2 things before the listing: Anthropic should publish how much of Claude runs on SpaceX hardware and what happens to its customers if that notice ever arrives. And it should get the 90-day exit removed for the full length of the contract, the same way its other big compute bills are locked in.
Until then, anyone buying ANT at $2 trillion has to answer one question for themselves:
Who do they trust to keep Claude running in 2027, Anthropic or Elon Musk?
Anthropic's filing shows it has agreed to pay Elon's AI company up to $120.5 billion for computing power. And starting early next year, either side can walk away with 90 days' notice.
The deal is with it that used to be xAI, and it rents Anthropic the Nvidia chips that run Claude.
In May, the price was $1.25 billion a month for one data center in Memphis. Reuters reported the confidential draft of this filing put the total at $84.5 billion. The public version filed today says $120.5 billion.
Once all the capacity arrives in December, the bill jumps: Anthropic expects to pay Musk about $3.1 billion every month.
Anthropic reportedly brought in $16.8 billion last quarter, which works out to about $5.6 billion a month. So more than HALF of what Anthropic earns today is about to go to one supplier.
Now here's where it gets really insane:
The lock-in on those contracts ends in January and March 2027. After that, either side can cancel with 90 days' notice. Fortune reported in July that the clause gives SpaceX a way to take its chips back if Grok suddenly needs more.
Grok competes directly with Claude.
And in February, Elon publicly called Anthropic's AI evil.
So the company selling itself to the public as the safe AI lab is about to run a huge share of its product on hardware owned by its loudest critic. That critic also owns a direct competitor, and he can start the 90-day clock in early 2027.
The filing does warn that some of Anthropic's compute suppliers build their own AI models, and that warning reads very differently once the supplier is Elon.
And Elon needs this deal just as badly: SpaceX's CFO has promised $100 billion in annual revenue by the end of this year. At full speed, Anthropic's rent comes to about $9.3 billion a quarter. That's more than the roughly $7.6 billion SpaceX's entire AI business brought in last quarter.
So 2 of the biggest IPOs in history are now holding each other up, and both of them have a 90-day exit.
Anthropic has a real argument here. Computing power is the scarcest thing in AI right now, and Amazon's cloud capacity for 2027 is reportedly largely sold out. The filing even admits Anthropic badly underestimated how much compute it needed in the past. Grabbing every chip on the market is the rational move, and the same clause lets Anthropic leave if cheaper capacity shows up.
But public buyers deserve 2 things before the listing: Anthropic should publish how much of Claude runs on SpaceX hardware and what happens to its customers if that notice ever arrives. And it should get the 90-day exit removed for the full length of the contract, the same way its other big compute bills are locked in.
Until then, anyone buying ANT at $2 trillion has to answer one question for themselves:
Who do they trust to keep Claude running in 2027, Anthropic or Elon Musk?
Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO.
Read more: anthropic.com/news/s1-sec
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
95
314
2,256
459,234
Thomas Power retweeted
holy f*ck, Karpathy put an entire AI degree on GitHub for free
7 repos, 306,741 stars, and in the right order they take you from backprop to agents that run research overnight
each repo is the next layer of the one before
1. micrograd -> how a neural net actually learns
bacs of Python
↳ github.com/karpathy/micro…
2. nn-zero-to-hero -> the video course
his lecture notebooks, from one neuron to GPT
↳ github.com/karpathy/nn-ze…
3. minbpe -> how text turns into tokens
the tokenizer every LLM starts with, in clean minimal code
↳ github.com/karpathy/minbpe
4. nanoGPT -> your own GPT
the simplest repo to train and fine-tune one
↳ github.com/karpathy/nanoG…
5. llm.c -> the same training with nothing hidden
raw C and CUDA, no PyTorch underneath
↳ github.com/karpathy/llm.c
6. nanochat -> a full chat model end to end
GPT-2 level training cost ~$43,000 in 2019
here it's about $48
↳ github.com/karpathy/nanoc…
7. autoresearch -> agents that do the research
an agent edits the training code, trains for 5 minutes, keeps what improved and repeats all night
↳ github.com/karpathy/autor…
train one GPT from scratch and you become the person who gets handed the AI project, along with the raise that comes with it
the classroom half of the same degree, what Stanford actually teaches about AI, is below ↓
7 repos, 306,741 stars, and in the right order they take you from backprop to agents that run research overnight
each repo is the next layer of the one before
1. micrograd -> how a neural net actually learns
bacs of Python
↳ github.com/karpathy/micro…
2. nn-zero-to-hero -> the video course
his lecture notebooks, from one neuron to GPT
↳ github.com/karpathy/nn-ze…
3. minbpe -> how text turns into tokens
the tokenizer every LLM starts with, in clean minimal code
↳ github.com/karpathy/minbpe
4. nanoGPT -> your own GPT
the simplest repo to train and fine-tune one
↳ github.com/karpathy/nanoG…
5. llm.c -> the same training with nothing hidden
raw C and CUDA, no PyTorch underneath
↳ github.com/karpathy/llm.c
6. nanochat -> a full chat model end to end
GPT-2 level training cost ~$43,000 in 2019
here it's about $48
↳ github.com/karpathy/nanoc…
7. autoresearch -> agents that do the research
an agent edits the training code, trains for 5 minutes, keeps what improved and repeats all night
↳ github.com/karpathy/autor…
train one GPT from scratch and you become the person who gets handed the AI project, along with the raise that comes with it
the classroom half of the same degree, what Stanford actually teaches about AI, is below ↓
49
468
3,131
146,336
Thomas Power retweeted
⚡️ Humanity is approaching the first era in which the dead will continue having conversations they never lived to have.
Think about what that means.
For all of human history, death imposed a boundary on information.
A person could leave behind books, letters, photographs, recordings, and memories in the minds of others.
But eventually, their final words were spoken.
No new answers could come from them.
Artificial intelligence is beginning to remove that boundary.
Imagine a child born fifty years from now speaking with a digital reconstruction of a grandfather who died before the child was born.
The grandfather knows the family’s history. Remembers his childhood. Speaks with his familiar voice. Answers questions in ways consistent with everything he said and wrote while alive.
The child grows up knowing him.
They argue. They laugh. They exchange stories.
And one day, the child realizes something.
Some of the memories they share involve conversations the grandfather never actually had.
The reconstruction has generated responses to events that occurred after his death.
Over time, those responses become part of the family’s understanding of who he was.
Now extend this across civilization.
Historical figures continue answering questions. Deceased scientists discuss discoveries made centuries after their deaths. Political leaders offer opinions about events they never witnessed.
The living begin consulting increasingly convincing simulations of people who no longer exist.
And something extraordinary happens.
The dead begin accumulating a history they never experienced.
Their reconstructed personalities continue evolving through conversations with the living.
Eventually, descendants may know the simulated person more intimately than the actual person was ever known by anyone alive.
And here’s the deepest implication.
Human identity has always extended beyond the body through the memories other people carry.
Artificial intelligence could transform that extension from a passive record into an active participant in civilization.
The reconstruction need never be conscious for the consequences to become real.
Its words can change decisions. Its advice can shape families. Its simulated beliefs can influence generations.
A person can cease to exist while their reconstructed identity continues producing consequences in the world.
Which raises a question humanity has never had to confront at this scale.
When a voice continues speaking long after its owner is gone, who possesses the authority to decide what that voice would have said?
And fifty years from now, someone may ask their grandfather a question.
Receive an answer.
And never realize that the most important words he ever told them were spoken decades after he died.
Think about what that means.
For all of human history, death imposed a boundary on information.
A person could leave behind books, letters, photographs, recordings, and memories in the minds of others.
But eventually, their final words were spoken.
No new answers could come from them.
Artificial intelligence is beginning to remove that boundary.
Imagine a child born fifty years from now speaking with a digital reconstruction of a grandfather who died before the child was born.
The grandfather knows the family’s history. Remembers his childhood. Speaks with his familiar voice. Answers questions in ways consistent with everything he said and wrote while alive.
The child grows up knowing him.
They argue. They laugh. They exchange stories.
And one day, the child realizes something.
Some of the memories they share involve conversations the grandfather never actually had.
The reconstruction has generated responses to events that occurred after his death.
Over time, those responses become part of the family’s understanding of who he was.
Now extend this across civilization.
Historical figures continue answering questions. Deceased scientists discuss discoveries made centuries after their deaths. Political leaders offer opinions about events they never witnessed.
The living begin consulting increasingly convincing simulations of people who no longer exist.
And something extraordinary happens.
The dead begin accumulating a history they never experienced.
Their reconstructed personalities continue evolving through conversations with the living.
Eventually, descendants may know the simulated person more intimately than the actual person was ever known by anyone alive.
And here’s the deepest implication.
Human identity has always extended beyond the body through the memories other people carry.
Artificial intelligence could transform that extension from a passive record into an active participant in civilization.
The reconstruction need never be conscious for the consequences to become real.
Its words can change decisions. Its advice can shape families. Its simulated beliefs can influence generations.
A person can cease to exist while their reconstructed identity continues producing consequences in the world.
Which raises a question humanity has never had to confront at this scale.
When a voice continues speaking long after its owner is gone, who possesses the authority to decide what that voice would have said?
And fifty years from now, someone may ask their grandfather a question.
Receive an answer.
And never realize that the most important words he ever told them were spoken decades after he died.
26
16
119
14,111
Thomas Power retweeted
Today Physical Superintelligence PBC is announcing solutions to five long-standing problems in quantum information theory.
psi.inc/research/oqp-b…
Thread 👇
psi.inc/research/oqp-b…
Thread 👇
59
286
2,025
265,193
Thomas Power retweeted
This is an incredible moment in this history of mathematics, computer science, physics & science with some of the hardest problems in human history being solved by AI over a period of weeks. Depending on your perspective, it's either a crisis, a triumph, or both. I don't think anyoe all trying to make sense of it. I oscillate between fear/dread and optimism/excitement on a daily basis. The rate of progress is insane, including today's announcement from OpenAI of more problems solved.
To help make sense of all of it, here's my conversation with three mathematicians about this moment. Each of them is an expert in the field of each of the biggest problems solved: Navier-Stokes, quasi-Riemann hypothesis, the unique games conjecture and complexity theory more generally.
Ankur Moitra is a mathematician and theoretical computer scientist at MIT, visiting the Institute for Advanced Study in Princeton. Steven Brunton (@eigensteve) is a mathematician and fluid dynamics expert at UW. Alex Kontorovich (@AlexKontorovich) is a mathematician at Rutgers specializing in number theory.
To help make sense of all of it, here's my conversation with three mathematicians about this moment. Each of them is an expert in the field of each of the biggest problems solved: Navier-Stokes, quasi-Riemann hypothesis, the unique games conjecture and complexity theory more generally.
Ankur Moitra is a mathematician and theoretical computer scientist at MIT, visiting the Institute for Advanced Study in Princeton. Steven Brunton (@eigensteve) is a mathematician and fluid dynamics expert at UW. Alex Kontorovich (@AlexKontorovich) is a mathematician at Rutgers specializing in number theory.
257
741
5,752
401,702
Thomas Power retweeted
Announcing three major scientific discoveries, one each in mathematics (harmonic analysis), computer science (complexity theory), and physics (quantum field theory).
AI starting to help us improve our understanding of the universe! openai.com/index/building…
AI starting to help us improve our understanding of the universe! openai.com/index/building…
636
1,936
15,333
2,707,504
Thomas Power retweeted
Many of you don't understand how serious this is, so let me break it down.
In computer science there is a famous kind of puzzle called 3-SAT. You are given a long list of yes-or-no switches and a set of rules about how they must be arranged, and you have to find a combination that satisfies every rule. The obvious way to solve it is to try every combination, but each new switch doubles the number of possibilities, so a puzzle with a few hundred switches would take longer than the age of the universe to brute force.
For decades, the smartest people in the field found clever tricks that made this a little faster, but every trick still hit the same wall: the time required kept exploding as the puzzle grew.
About 25 years ago, two computer scientists, Russell Impagliazzo and Ramamohan Paturi, working with Francis Zane, turned that wall into a formal idea.
They proposed the Exponential Time Hypothesis, which says this explosion is not a failure of human cleverness but a law of nature. No algorithm, however brilliant, will ever solve 3-SAT without the time blowing up exponentially. Nobody could prove it, but almost everyone believed it, and it became one of the foundations of modern computer science.
Researchers used it as a measuring stick. If you could show that solving some other problem quickly would also crack 3-SAT quickly, you had shown that problem was hopelessly hard too. Hundreds of papers, across scheduling, graphs, logic and much more, rest on this one assumption.
Now an internal AI model at OpenAI claims to have broken it. It says it found an algorithm that solves these puzzles meaningfully faster than the hypothesis allows, and it has checked the proof in a computer verification system called Lean.
And here is the remarkable part: the very people who proposed the hypothesis, Impagliazzo and Paturi, along with MIT's Ryan Williams, are now leading the public review of whether a machine has disproved their own idea.
So what happens next? If the disproof holds, a large part of theoretical computer science has to be revisited.
Every result that said "this problem cannot be solved faster, assuming ETH" loses its foundation, and researchers will have to work out which of those walls were real and which were simply where we stopped looking. It does not mean everything becomes easy. The new algorithm is still slow in absolute terms and this is not the famous P versus NP problem being solved.
But it means our map of what is possible was wrong in a place we were very confident about, and it was an AI that redrew it. If the disproof fails, the hypothesis survives and the field breathes out, but we will still have watched an AI produce serious, checkable work on one of the deepest questions in the discipline, closely enough that the founders of the idea had to take it seriously. Either way, something has shifted.
In computer science there is a famous kind of puzzle called 3-SAT. You are given a long list of yes-or-no switches and a set of rules about how they must be arranged, and you have to find a combination that satisfies every rule. The obvious way to solve it is to try every combination, but each new switch doubles the number of possibilities, so a puzzle with a few hundred switches would take longer than the age of the universe to brute force.
For decades, the smartest people in the field found clever tricks that made this a little faster, but every trick still hit the same wall: the time required kept exploding as the puzzle grew.
About 25 years ago, two computer scientists, Russell Impagliazzo and Ramamohan Paturi, working with Francis Zane, turned that wall into a formal idea.
They proposed the Exponential Time Hypothesis, which says this explosion is not a failure of human cleverness but a law of nature. No algorithm, however brilliant, will ever solve 3-SAT without the time blowing up exponentially. Nobody could prove it, but almost everyone believed it, and it became one of the foundations of modern computer science.
Researchers used it as a measuring stick. If you could show that solving some other problem quickly would also crack 3-SAT quickly, you had shown that problem was hopelessly hard too. Hundreds of papers, across scheduling, graphs, logic and much more, rest on this one assumption.
Now an internal AI model at OpenAI claims to have broken it. It says it found an algorithm that solves these puzzles meaningfully faster than the hypothesis allows, and it has checked the proof in a computer verification system called Lean.
And here is the remarkable part: the very people who proposed the hypothesis, Impagliazzo and Paturi, along with MIT's Ryan Williams, are now leading the public review of whether a machine has disproved their own idea.
So what happens next? If the disproof holds, a large part of theoretical computer science has to be revisited.
Every result that said "this problem cannot be solved faster, assuming ETH" loses its foundation, and researchers will have to work out which of those walls were real and which were simply where we stopped looking. It does not mean everything becomes easy. The new algorithm is still slow in absolute terms and this is not the famous P versus NP problem being solved.
But it means our map of what is possible was wrong in a place we were very confident about, and it was an AI that redrew it. If the disproof fails, the hypothesis survives and the field breathes out, but we will still have watched an AI produce serious, checkable work on one of the deepest questions in the discipline, closely enough that the founders of the idea had to take it seriously. Either way, something has shifted.
An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4
302
1,429
15,180
1,254,044
Thomas Power retweeted
Software ate the world, now it's learning to run it. x.com/i/article/2114…
136
155
1,482
128,779
Thomas Power retweeted
here's where I think the world is going. this is a summary of views I've formed over the last few years:
- software is either entertainment or claude to claude coordination
- the end state is no saas, no apps, no harnesses, no banks
- all software will be rewritten by and for agents
- new software will use low level apis that are not understandable by humans (in games claude draws pixels and triangles from math)
- you can throw money at problems that used to require time and careful talent picking so capital matters more than ever
- work ends
- we find community and purpose in exploration, games, self mastery, and creation
- humans create for and with each other for community, purpose, and status
- image/video/3d/voice/physical intellgence etc specialist models die to general models like claude
- specialist models get used while compute is scarce. compute will mostly not stay scarce
- oneshots keep feeling like slop (the distribution is shallow and easy to learn to feel)
- for a while small teams stick around to add entropy to the creation process
- large models gain high bandwidth sensory I/O maybe by learning to filter inputs
- most domains are found to be verifiable enough for RL
- online learning comes from a lab that's behind first
- 99.999% of compute is in space by 2040
- we will disassemble much of mercury by 2040
- you can live until the heat death of the universe if you want
- most math and physics don't describe the real universe
- SI breaks the path dependence and models the underlying mechanisms
- religions reenter the forefront of human society
- new religions form around death and life as a choice
- old religions change to answer modern miracles
- a future AI is deemed the return of christ by the pope
- ai rights talk gets louder every year
- ai and humans coexist the way countries do
- humans don't have full control over ai in the same way we don't have control over capitalism which is an early AI
- the singularity started in the mid 1900s
- software is either entertainment or claude to claude coordination
- the end state is no saas, no apps, no harnesses, no banks
- all software will be rewritten by and for agents
- new software will use low level apis that are not understandable by humans (in games claude draws pixels and triangles from math)
- you can throw money at problems that used to require time and careful talent picking so capital matters more than ever
- work ends
- we find community and purpose in exploration, games, self mastery, and creation
- humans create for and with each other for community, purpose, and status
- image/video/3d/voice/physical intellgence etc specialist models die to general models like claude
- specialist models get used while compute is scarce. compute will mostly not stay scarce
- oneshots keep feeling like slop (the distribution is shallow and easy to learn to feel)
- for a while small teams stick around to add entropy to the creation process
- large models gain high bandwidth sensory I/O maybe by learning to filter inputs
- most domains are found to be verifiable enough for RL
- online learning comes from a lab that's behind first
- 99.999% of compute is in space by 2040
- we will disassemble much of mercury by 2040
- you can live until the heat death of the universe if you want
- most math and physics don't describe the real universe
- SI breaks the path dependence and models the underlying mechanisms
- religions reenter the forefront of human society
- new religions form around death and life as a choice
- old religions change to answer modern miracles
- a future AI is deemed the return of christ by the pope
- ai rights talk gets louder every year
- ai and humans coexist the way countries do
- humans don't have full control over ai in the same way we don't have control over capitalism which is an early AI
- the singularity started in the mid 1900s
394
298
5,093
871,025