Efe retweeted
Barcola doit être entrain de s'arracher les locks il a quitté le tyrannique PSG pour le cadavérique Liverpool et ses 2 concurrents au PSG sont out pour la saison il aurait été titu en tong
Efe retweeted
🚨 José Mourinho: "Could I play Espi-Mbappé together?
Yes, but when we lose the ball, I'll have to be the one to start running on the pitch and I'll kick the h*ll out of the opposition player on the counter."
Efe retweeted
This the kinda stuff Pep would say after we beat a new team 7-0
Alvaro Arbeloa on #ManCity under Pep Guardiola/Enzo Maresca: "They have some similar ideas, but it's not fair for Maresca [to judge him] because he's just arrived after three, four months. He has some new players, obviously they are amazing players, an amazing team...
"I watched the three games they lost [in a row], they had so many chances to win. They lost, but I don't know if they deserved to lose. Every new manager needs time, even if you have amazing players like Maresca has. For me, he's a really good coach, and he will prove again!"
Efe retweeted
🚨 | Désiré Doué (21) has torn his right ACL; the PSG and France international winger could be out for the rest of the season. (L'Eq)
getfootballnewsfrance.com/20…
Chamath reveals his AI megacorp thesis: “Each megacorp is going to go from the application layer all the way down to the dirt.”
“I have a theory. I think that there are going to be nine to 10 megacorps that get built over the next decade because of AI. And I think that each megacorp is going to go from the application layer all the way down to the dirt.“
“What is Anthropic doing? With all their money they're already going down to the dirt and then they're going up to the application stack. OpenAI doing the same thing. SpaceX doing the same thing, they own one of the best apps in the game, Cursor, they own all the way down to the dirt. Meta owns all the way down to the dirt all the way up to the app stack.”
“So I think what AI is creating essentially is haves and have nots. The haves are the folks that have these critical assets that can be very tightly vertically integrated. And I think that there's going to be 10 of these folks.”
“You’re going to have this really rich app ecosystem, a lot of open source models and some closed source models, and then you're going to use a lot of distributed silicon.”
“And then you're going to need this really important constraint which is the land and the power. So I think if you view it that way, you're going to see these 8, 9, 10 conglomerates, and now the market is betting who is a conglomerate and who is not. And that more than anything I think explains the flows over the last few weeks and probably like the last year even.”
“It doesn't mean anything. What does it mean to make a product in a world where you have a trillion agents that you can spin up on demand to make anything you want? What is a product? I don't know what a product is. What is that?”
Efe retweeted
Jamie Carragher called Cole Palmer the best player in the league in a season where Salah single handedly took the club he claims to support to a 20th league title 😂
Efe retweeted
How is it that any sentient adult can claim to have been blindsided by the EU side? Did they sleep through Michel Barnier from 2016 to 2021?
EXCLUSIVE
* Senior allies of Andy Burnham say he is determined to take Britain back into the EU on latest episode of The State or It with @patrickkmaguire and @lara_spirit
* “He wants to rejoin, that’s the truth of it. He feels it very deeply and thinks he is on the right side of public opinion,” one ally tells @patrickkmaguire. “The idea that this is being driven by anyone other than him or it’s just clever positioning or tactical is wrong. It’s just what he believes”
* How Andy Burnham walked into a Brexit ambush after member states demanded tuition fees issue was resolved before the deal could be implemented. It blindsided the British side and left bad blood
* There is frustration some in govt with Hamish Falconer. The concern is that the British side underestimated how important this is to the EU - partly because so many figures in Brussels send their kids to university here. There is a significant degree of self-interest involved
* We now enter a year of discussions about the Brexit options. The No 10 approach will be the reverse of “project fear” - talking up the benefits of closer ties at every turn. They argue that the Tories and Reform UK are now campaigning for the status quo, which polls suggest people are far from happy with
* But there will be trade offs - financial payments, free movement etc. The EU is clear these are non-negotiable. Closer alignment may be popular now, but will that change in it’s a debate about tje costs involved?
podcasts.apple.com/gb/podcas…
Efe retweeted
OAI will never IPO
Okay now think what happens to OpenAI who still needs to raise a ton of capital for all the commitments if Anthropic is trading below their last funding round (1.25T)
Bank of $NVDA is now closed and Softbank is already levered to the gills. Who is going to give them billions?
Efe retweeted
It’s genuinely insane how Griezmann, who lost the UCL final, Euro final and the Ballon d’Or to Ronaldo in 2016, still speaks so highly of him, while Bruno scores against Wales and points towards the captain’s armband. I just hope the Netherlands absolutely destroy those rats
Efe retweeted
$AAPL at $5 trillion, +27% YTD and barely off all time high, watching the AI carnage, while Siri AI takes 12 seconds of thinking to set a timer.
Efe retweeted
TPUs are helping Google $GOOG win the AI race by enabling a model-silicon recursive self-improvement (RSI) loop.
Up until recently, RSI was mostly a concept that applied to AI models helping make better AI models. I believe Google is changing that narrative. It's becoming clear that model-silicon co-design is the future of AI, and RSI has a lot more leverage in the model-chip feedback loop.
Because of scaling laws, the AI race has always been more of a compute acquisition race. Model design is important, but secondary to compute. More compute gets you better results, while better models help you extract more intelligence at lower costs. So the race really comes down to first who can accumulate compute at the cheapest price, and then who can use it most efficiently.
TPUs are arguably the biggest leverage Google has in the AI race. Internally, TPUs are 30-40%+ cheaper than comparable $NVIDIA setups, even after paying Broadcom and TSMC, since Google isn't paying NVIDIA's ~75% gross margins. When frontier AI companies are pouring hundreds of billions into compute, that gap means Google gets way more compute per dollar with much better ROIC, and that compounds every cycle.
Selling TPUs to outside buyers like Anthropic $ANT, Meta, and Oracle also lets Google spread its chip R&D across more volume. Because they can afford thinner margins than NVIDIA, they can undercut on price while still keeping the best economics for themselves.
But I'd argue those substantial cost savings aren't even the biggest edge for Google. As I said in my last post, I believe every frontier lab will eventually become a chip company because otherwise you leave too much efficiency on the table.
Because Google knows Gemini's roadmap in advance, they can strip out the extra features a GPU needs and design chips strictly around Gemini's bottlenecks. For example, TPU 8i triples on-chip memory and redesigns chip-to-chip networking specifically for long-context MoE reasoning, giving them up to 80% better performance per dollar than Ironwood.
On the flip side, the Gemini team designs the next model around the TPU's exact layout before the chip is even made. TPUs are much more rigid than GPUs, so outside models designed for NVIDIA can lose half their utilization in key layers on a TPU, while Gemini is tailored to fit it.
Outside buyers like Anthropic and Meta still buy TPUs because the pricing and supply make it worth adapting their workloads, but they can't match the co-design efficiency Google gets by building Gemini and the chip together.
NVIDIA can't go that far because it doesn't own the models. It can't force thousands of customers to redesign their models around its chip, so it has to keep its chips flexible instead of tuning for one model. In fact, a core part of NVIDIA's success historically has been that they built a general-purpose chip that could be used by gamers, crypto miners, and different AI labs.
Jensen used to argue custom chips move too slowly for fast-changing models, and that's clearly not holding up anymore. Google went from 2-3 year TPU cycles, to yearly releases, to shipping 4 different 8th-gen systems just this year (8t, 8i, 8tx, and 8i Max), each tuned for a different job.
Having both the chips and the models is what gives Google a real shot at RSI:
• Before leaving Google, Jeff Dean said AlphaEvolve came up with a circuit going straight into next-gen TPUs, calling it "TPU brains helping design next-generation TPU bodies."
• AlphaChip cuts chip layouts from weeks to hours, and AlphaEvolve sped up part of Gemini's training code by 23% (taking 1% off total training time).
• Amin Vahdat, who runs AI infra, said TPU 8 was co-designed for Gemini, and this week Google showed that building models around TPUs can boost training throughput by up to 57%.
• Sergey said their models are now helping build the next ones, and AI already writes over 75% of DeepMind's production code (with humans still reviewing it).
• Gemini 4 Argon agents also freed up 300+ TB of data center memory on their own and wrote custom TPU kernels running 8x to 15x faster.
• Sundar even told investors on the Q3 call to expect "continued acceleration" over the next 12 months because of RSI.
For most of 2026 Google looked behind and even missed their June Gemini 3.5 Pro target. Then Gemini 4 Argon went live internally, and over the last 7 weeks they've shipped 4 Gemini 4 models (Argon, Carbon, Hydrogen, and Neon).
Neon is only their mid-tier model, yet it beat Opus 5.5 and GPT-6.2 Sol on almost every launch benchmark (only trailing on 3), hit 95% on long context vs 67% and 80%, and cut hallucinations to 14% vs 50%+. On Artificial Analysis it's also about 36% cheaper per task than Opus (Sol is cheaper, though with much higher hallucination rates).
In tech, even a slightly faster rate of improvement compounds into winner-take-most. Right now it looks like Google is improving faster than everyone else.
OpenAI and Anthropic are obviously still serious contenders, especially if eventually scale their own custom chips, though neither has Google's full stack today. NVIDIA was the one company that already had the hardware edge to close this loop from the other side by making huge bets on frontier open-weight models but that is now in question without Jensen there to take those risks. Google here has a real shot at pulling away and in the RSI world if you pull-away, it'll be very difficult for others to catch up as the distance keeps widening.
Ok I do have a view on NVIDIA without Jensen.
I’ll tell you the real reason $NVDA will likely fail without him and why the frontier labs and hyperscalers are the biggest winners. (Elon has been in a really good mood recently. Have you noticed?)
The biggest threat to NVIDIA is its own customers, the frontier labs. NVIDIA has to become a frontier lab faster than frontier labs are becoming chip companies. And with Jensen out, they likely won't maintain the momentum to get there.
With recursive self-improvement (RSI), every frontier lab and hyperscaler must become a chip company or lose the AI race. It's that simple, and they've all laid the groundwork (Google TPUs, OpenAI and Meta building custom ASICs with Broadcom, Anthropic building an in-house silicon team, and SpaceX/xAI/Tesla building Terafab).
That's because software-only RSI has clear physical limits. Software can reduce how much computation and data movement a model needs, but it can't change the chip's physical bandwidth, memory capacity, or interconnect. Custom silicon gives labs another way to improve performance: designing the memory system, compute, and networking around their own workloads.
To get the real compounding effect of RSI, you have to co-design the model and the silicon so it becomes a two-way feedback loop: smarter models design better custom chips, and those chips train and run the next model faster. You can't do that on off-the-shelf NVIDIA GPUs because you get zero proprietary edge buying the same chip as everyone else, and getting your next-gen architecture into NVIDIA's silicon means sharing your roadmap with them years before tape-out (which they then sell to your rivals). Plus, AI now automates writing compilers and porting CUDA kernels to other chips, which kills the CUDA moat.
Jensen argued custom ASICs aren't more cost-efficient than GPUs because models change too fast for fixed hardware, and partners like Broadcom charge fat margins too. But hardware doesn't fall behind model changes when AI co-designs the next model and chip together instead of humans designing them separately (which Google and OpenAI are already doing). And labs aren't building custom silicon just to save on NVIDIA's 75%+ gross margins. Closing that hardware-software loop is the main goal, though the cost savings are still massive since RSI requires 24/7 internal inference where custom ASICs are way cheaper and more power-efficient than general-purpose GPUs.
That's why NVIDIA has been trying so hard to become a frontier open-weight lab, buying Groq (fast inference IP), Poolside + Nemotron (internal coding models), and Hugging Face (distribution). If closed labs win, enterprises just call their APIs running on the labs' own custom silicon and NVIDIA gets cut out. Subsidizing free open-weight models is the only way to make enterprises buy or rent NVIDIA GPUs to run the models themselves.
With Jensen gone, it's unclear if the Hugging Face deal even survives DOJ approval (can a GPU company own the whole stack?), and a non-founder is unlikely to take the massive risks needed to become a frontier lab: putting billions of capex into internal clusters and free open-weight models instead of just selling GPUs, while competing directly with its own biggest customers $GOOG, $META, $SPCX, $ANT, $MSFT
Efe retweeted
what i’ve come to understand and hold true in my head but am always revisiting…
1. we don’t have enough raw resources (memory, cpu, power, data centers) - this won’t go away for perhaps up to 6 to 10 years
2. There is minimal to no moat for the prop labs; except selling poison and cure. They will make money selling expensive expensive cyber tokens in response to the security problems created by AI
3. AI fundamentally has changed work and there’s no going back. The box is open and businesses need fewer humans to run a businesses.
4. AI is an economic munition. If a country depends upon AI (and they will) and their access is suspended then a country will be economically crippled by such sanctions.
5. Openweight models are typically three months behind prop labs. They are a critical counter to $ant.
6. Countries must develop their own sovereign open weight models which codify their culture/customs to protect/encode their nations value structures to protect against cultural homogenization.
7. Distillation is competition and fair game. How it is done is should be the focus of conversation (not “if” it’s allowed) - ie distillation currently happens through theft/credit card fraud - that isn’t good!
Though tough thoughts am pondering about:
8. Training on pre-existing IP is perhaps not theft of IP. The same way that a human reading a book, seeing an idea, learning concepts and then remixing that concept isn’t a copyright violation. Mathematics isn’t IP/copyright theft.
Efe retweeted
I hate the Nigerian education system mehn, this is so evil
This can only mean one thing, Robert Pattinson is about to replace Ronaldo in the Portuguese national team
Efe retweeted
cedric diggory died at 17,
edward was stuck at 17,
mickey 17,
james bond is 007,
cedric's shirt number was 07
he's the 7th batman and the 7th bond,
robert pattinson on some voodoo shit dawg
😭😭😭😭
Efe retweeted
"While Intuit’s revenue increased 12% year over year, Mailchimp revenue declined 1% to $313 million."
3x ARR and Mailchimp unlike Airtable isn't growing
But, it might be easier to turn around if make it more agent-friendly
And Bending Spoons will quickly raise prices, ofc
2021: Intuit buys Mailchimp for $12 billion
2026: Intuit sells Mailchimp to Bending Spoons for $3.68 billion cnbc.com/2026/11/19/intuit-m…




































