Pinned Tweet
Jerry Seinfeld turned down $110 million.
NBC offered him $5 million an episode for a tenth season. It was 1997, and Seinfeld was the biggest comedy on TV.
He said no.
Every year before the Christmas break, Jerry asked his co-stars if they wanted to keep going. In season 9, all three said they were done.
Jason Alexander explains why. He, Julia Louis-Dreyfus and Michael Richards were not partners in syndication.
Jerry and Larry David owned the show. The reruns would pay them for life.
Alexander says he has made less than $200,000 from them.
The cast was paid well, reportedly $600,000 an episode by the end. But Alexander says another season only tied them tighter to one character.
His words: “We were going to contribute to our own eventual demise as actors.”
Before the final taping, the four held hands backstage like every week. They called it the circle of power.
Same question today with AI: when your work builds something that pays forever, who owns it?
Who do you think was right, Jerry or the cast?
Numex retweeted
DAMN, THIS BOARD PUTS 8 ARMS ON ONE BINARY, REBUILDS THE SDK AND HITS VERIFY 128/128 IN 0.12s
it runs the clean-room loop from this Opus 5.5 reverse engineering article, where one Touhou rebuild came back as 697 functions with 696 exact matches
the idea: rebuild an SDK directory from a binary, one arm per independent task, no code before the spec exists, and one rule on the footer: a catalog entry is not a network call
the split:
→ investigate: REA hands the agent a real disassembler through MCP, running locally
→ spec: every claim gets a tag, observation, inference or unknown, with the file and address
→ build: binary → AWS SDK → region catalog, each arm runs grip, transfer, place, return
→ verify: 50 inputs through both builds, 30 normal, 10 edge, 10 malformed
you only check:
→ the match rate the harness reports
→ the unknown tags, that's where the agent stopped guessing
→ effort: medium to navigate, high to annotate, max for obfuscated code
the line from the article I keep: "I think it matches" is not evidence
China will ALWAYS beat America in AI.
Here’s why.
50% of the world’s AI researchers are Chinese.
70% of last year’s AI patents were published by China.
9 out of 10 top science and technology schools in the world are now in China, Jensen Huang says.
He calls China’s AI ecosystem vibrant, rich and incredibly innovative.
Now the money.
Huang says China alone is a $50 billion AI opportunity in 2026.
Nvidia once held about 95% of China’s AI chip market.
After US export controls, that fell to around 50%.
Now Huang says Nvidia has largely conceded that market to Huawei.
Talent. Patents. Universities. Money.
That’s what decides AI races.
Can the West catch up?
Anthropic is raising money at a $350 billion valuation. At the same time, one investor on stage says AI is not what people fear.
Lonsdale spoke at a CAZ Investments conference. He called AI not a god and not a threat to humanity.
In the video below, he explains what it is instead, calmly.
“A productivity tool,” comparable to the Second Industrial Revolution of 1870 to 1900.
He said US services pay about $5 trillion in wages.
About $2 trillion of that, 40%, is ready for a big productivity jump from AI.
He said companies his firm backed two years ago tripled their cash flow.
In many service businesses, he says, AI already doubles or triples profitability.
He also argued that the US has to win the AI race, so hostile states can’t use the technology to shape culture and decisions in other countries.
His message to founders: focus on measurable productivity, not hypothetical threats.
So is AI a real $2 trillion opportunity, or is this just what investors say when they have money on the table?
Numex retweeted
Trump called Jensen Huang. Live. On stage.
Jensen just said “put me on speaker.”
And then the whole room heard a number.
$20 trillion.
That’s how much investment Trump says is coming into the US right now.
Under the previous administration, he said, it was less than $1 trillion.
Same country. Same AI race.
A 20x gap, if you believe the claim.
At the same summit, Jensen was pushing back on “AI doomsday” predictions.
He called them unscientific and irresponsible.
So the CEO of the most valuable AI chip company is on stage.
And the President joins by phone.
When the AI story gets this big, it stops being tech news.
It becomes a money story.
Do you think $20 trillion is realistic?
Numex retweeted
Elon Musk just said something uncomfortable about AI.
“Humans will not. There won't be anything that humans can do that the AI and the robots can't do better.”
His vision goes far beyond chatbots.
AI could outperform humans at almost every job, while robots turn intelligence into real-world production.
But if machines can do everything better than us…
What exactly will give human life meaning?
This is the question nobody can ignore anymore.
Trump called Jensen Huang. Live. On stage.
Jensen just said “put me on speaker.”
And then the whole room heard a number.
$20 trillion.
That’s how much investment Trump says is coming into the US right now.
Under the previous administration, he said, it was less than $1 trillion.
Same country. Same AI race.
A 20x gap, if you believe the claim.
At the same summit, Jensen was pushing back on “AI doomsday” predictions.
He called them unscientific and irresponsible.
So the CEO of the most valuable AI chip company is on stage.
And the President joins by phone.
When the AI story gets this big, it stops being tech news.
It becomes a money story.
Do you think $20 trillion is realistic?
Numex retweeted
Jerry Seinfeld turned down $110 million.
NBC offered him $5 million an episode for a tenth season. It was 1997, and Seinfeld was the biggest comedy on TV.
He said no.
Every year before the Christmas break, Jerry asked his co-stars if they wanted to keep going. In season 9, all three said they were done.
Jason Alexander explains why. He, Julia Louis-Dreyfus and Michael Richards were not partners in syndication.
Jerry and Larry David owned the show. The reruns would pay them for life.
Alexander says he has made less than $200,000 from them.
The cast was paid well, reportedly $600,000 an episode by the end. But Alexander says another season only tied them tighter to one character.
His words: “We were going to contribute to our own eventual demise as actors.”
Before the final taping, the four held hands backstage like every week. They called it the circle of power.
Same question today with AI: when your work builds something that pays forever, who owns it?
Who do you think was right, Jerry or the cast?
Numex retweeted
Elon Musk just admitted he may have sped up the AI race he was afraid of.
He helped launch OpenAI in 2015 to balance Google. Anthropic split off in 2021. His verdict: reality isn’t just an “interestingness maximizer.” It’s an “irony maximizer.”
Now count the irony. 1 lab became 2, and 2 became a race. He sued for $150B, and a jury in May ruled his claims fell outside the statute of limitations.
Then came the plot twist. He wrote that he was “clearly wrong” about Anthropic and called it the current AI leader.
My take: you can’t slow a race by adding runners. Every spin-out and every rival lab adds one more team with a reason to go faster.
Brake or gas? Which side are you on? Drop your take below.
Jerry Seinfeld turned down $110 million.
NBC offered him $5 million an episode for a tenth season. It was 1997, and Seinfeld was the biggest comedy on TV.
He said no.
Every year before the Christmas break, Jerry asked his co-stars if they wanted to keep going. In season 9, all three said they were done.
Jason Alexander explains why. He, Julia Louis-Dreyfus and Michael Richards were not partners in syndication.
Jerry and Larry David owned the show. The reruns would pay them for life.
Alexander says he has made less than $200,000 from them.
The cast was paid well, reportedly $600,000 an episode by the end. But Alexander says another season only tied them tighter to one character.
His words: “We were going to contribute to our own eventual demise as actors.”
Before the final taping, the four held hands backstage like every week. They called it the circle of power.
Same question today with AI: when your work builds something that pays forever, who owns it?
Who do you think was right, Jerry or the cast?
Numex retweeted
Anthropic’s CEO just put a date on it: 2027.
Dario Amodei says AI could be better than almost all humans at almost everything. It could take longer, he adds, but not “a whole bunch longer.”
He’s not an outside critic. He runs one of the labs building it.
The date isn’t the point. The question after it is: if machines do the work, how do humans find meaning?
His answer in 86 seconds: the idea that self-worth comes from economic labor will be invalidated.
“My true belief is that this is coming.”
Does 2027 sound too early or too late to you?
Anthropic’s CEO just put a date on it: 2027.
Dario Amodei says AI could be better than almost all humans at almost everything. It could take longer, he adds, but not “a whole bunch longer.”
He’s not an outside critic. He runs one of the labs building it.
The date isn’t the point. The question after it is: if machines do the work, how do humans find meaning?
His answer in 86 seconds: the idea that self-worth comes from economic labor will be invalidated.
“My true belief is that this is coming.”
Does 2027 sound too early or too late to you?
His own AI agent just negotiated $80K off his salary. In 6 minutes.
Graham sent an AI, Victor, to his job interview. The company sent an AI recruiter, Pat. Graham sat in the call and watched.
His target: $140K–150K.
Victor’s offer: $60K.
The recruiter’s ceiling: $90K, still $30K above what his own bot asked for.
The video is staged, he says. The risk isn’t.
Companies already use AI recruiters on real candidates. An agent that speaks for you can still work against you.
Would you let an AI negotiate your salary? Yes or no.
Jerry Seinfeld turned down $110 million.
NBC offered him $5 million an episode for a tenth season. It was 1997, and Seinfeld was the biggest comedy on TV.
He said no.
Every year before the Christmas break, Jerry asked his co-stars if they wanted to keep going. In season 9, all three said they were done.
Jason Alexander explains why. He, Julia Louis-Dreyfus and Michael Richards were not partners in syndication.
Jerry and Larry David owned the show. The reruns would pay them for life.
Alexander says he has made less than $200,000 from them.
The cast was paid well, reportedly $600,000 an episode by the end. But Alexander says another season only tied them tighter to one character.
His words: “We were going to contribute to our own eventual demise as actors.”
Before the final taping, the four held hands backstage like every week. They called it the circle of power.
Same question today with AI: when your work builds something that pays forever, who owns it?
Who do you think was right, Jerry or the cast?
Numex retweeted
In the early days of Apple, Steve Jobs and Steve Wozniak faced a question that would follow them for years: what matters more, the money or the people who helped you make it?
Back in 1975, Atari hired them to build Breakout, an arcade game that had to run on as few chips as possible. Jobs took the assignment. Woz had the engineering skills to solve it.
There was one problem. Woz was already working full-time at Hewlett-Packard, so he spent four sleepless days getting the job done. The result was remarkable: just 45 chips, while Atari’s own engineers were using around 150.
Atari paid Jobs for the work. He told Woz the reward was $700 and handed him $350. But the actual payment had been roughly $5,000.
Woz accepted the money without questioning it. Years later, he discovered the truth in a book about Atari. By then, Apple had made both men incredibly wealthy.
And this wasn’t an isolated case. Before Apple went public in 1980, Jobs refused to give stock options to Daniel Kottke, a longtime friend who had helped build the company. Woz, meanwhile, shared his own shares with early employees who had been left out.
One man held on to the money. The other gave a meaningful part of his wealth away.
Jobs was undeniably brilliant at building a business. But brilliance and loyalty are two different things.
The real question isn't whether you can win in business. It's what you're willing to do to win, and who pays the price.
Musk just said most robot demos aren’t real AI.
His test is simple: show the robot something it has never seen. If it still handles it, that’s real AI. Same bar as full self-driving.
The hard part is closing the loop between vision and touch. A robot has far more degrees of freedom than a car, and no script to lean on.
But smart isn’t enough. Production ramps like an S-curve, and the beginning is slow. The supply chain for robot parts barely exists.
His example: if China cuts off magnets for electric motors, you get robots with no arms.
The winner won’t have the best demo. They’ll have the supply chain.
What matters more in the next 5 years: smarter AI or the supply chain behind it?
Anthropic’s CEO says: “We don’t know if the models are conscious.”
A neuroscientist just told TED: we do know.
Anil Seth’s answer is one word.
No.
Here’s his argument. It’s simple.
Simulate a hurricane on a supercomputer.
Does anything get wet?
No.
The storm is in the code. The rain isn’t.
Now simulate a brain.
Every neuron. Every connection.
Does anyone feel anything?
Same logic. A simulation of a mind isn’t a mind.
So why does ChatGPT feel conscious?
Because for all of human history,
language meant a mind was behind it.
Now we’ve built something that talks without one.
Our instincts haven’t caught up.
Seth’s idea is the uncomfortable part.
Consciousness may not be about intelligence.
It may be about being alive.
A body that has to keep itself going.
Something at stake.
Intelligence is what you do.
Consciousness is what it feels like to be a living thing doing it.
His closing slide:
“AI is computer software. It is not a living mind.”
The real risk may not be a conscious AI.
It’s millions of people believing in one.
Do you think he’s right?
40 Mac Minis cost about the same as one Nvidia H100.
Someone in China just lined up hundreds of them on yellow hardware-store shelves. Cables everywhere, a dashboard on a monitor, daylight through the office windows.
Each Mini draws 5 to 40 watts. No special power, no cooling. Unified memory means small models run well on retail hardware.
One Mini loses to an H100. But 40 of them running small models in parallel, 24/7, with zero cloud bill? For some workloads, that math works.
The catch: managing hundreds of separate machines is painful, the networking is slow, and none of it will train a frontier model.
Everyone is fighting over GPUs. Some people just bought shelves.
Smart move or expensive toy? Tell me below.
Numex retweeted
warren buffett is 96 and still telling everyone the same boring thing. almost nobody listens.
he’s been on stage at the berkshire meeting for decades saying it. no secret strategy. no trading system. no course to buy.
what most people spend every month:
- trading app fees: $30
- crypto “signals” group: $99
- stock picking newsletter: $49
- panic selling on red days: priceless
total: broke and stressed
what buffett actually tells regular people to do:
- buy a low-cost s&p 500 index fund and keep buying it
- think in decades, not weeks
- keep fees as close to zero as possible
- check your portfolio rarely
the tools you actually need:
index fund (voo / vti / fxaix)
own the whole market for ~0.03% a year. buffett told his own family’s trustee to put 90% here
brokerage (fidelity / schwab / vanguard)
zero commissions, fractional shares, done
auto-invest on payday
the money moves before you can spend it or second-guess it
high-yield savings
3–6 months of expenses, so a bad month never forces you to sell at the bottom
a simple spreadsheet
track your net worth once a month. not daily. daily is how you end up panic selling
what nobody tells you:
- the hard part isn’t picking stocks, it’s doing nothing when the market drops
- fees look tiny until you compound them over 30 years
- boring is the strategy. if it’s exciting, you’re probably paying for it
Numex retweeted
this is the most underrated ai setup on the internet right now.
4 mac studios stacked on my desk, wired into one ring over thunderbolt, running a model that no single machine on earth under $10k could hold. zero cloud. zero api keys. zero monthly bill.
what i was paying every month:
- cursor ultra .............. $200
- claude max 20x ............ $200
- chatgpt pro ............... $200
- google ai ultra ........... $250
- total ..................... $850/mo
what runs it now:
- 4× mac studio, 192gb each . 768gb unified memory
- thunderbolt cables ........ ring, no switch needed
- electricity .............. ~$25/mo
- subscriptions ............. $0
the tools that make it work:
exo ....................... github exo-explore/exo
one command, and it finds every mac on the network and splits the model across them layer by layer. 4 boxes start acting like one giant gpu
mlx ....................... apple’s own framework
built for apple silicon, runs straight on unified memory. the reason this is fast and not a slideshow
deepseek / kimi k2 / qwen3 . open weights
600b+ parameter models at 4-bit fit across the ring. one studio can’t load them. four can
lm studio ................. lmstudio ai
for the smaller models when you don’t need the whole cluster
tailscale ................. tailscale com
hit the cluster from your phone anywhere, feels exactly like the chatgpt app
what nobody tells you:
-> thunderbolt ring > wifi. over wifi exo works, but tokens crawl
-> memory is the real limit, not compute. add up the model size before you buy anything
-> keep the weights loaded, cold starts on a 600b model take minutes
-> it’s not cheap upfront. this pays off if you’re actually burning $800+ a month, not $20
12 months in, the bill i used to pay is still in my pocket.