June_e Bai

Prediction confirmed.💯 This is the result of my 7 months of synthesizing Elon’s fragmented tweets. Having the right mentor matters. ✅ @elonmusk @grok #AGI

@ylecun @aakashgupta @boztank While LLMs excel at language and reasoning, prioritizing them over world models' core advantages in embodied interaction is definitely short-sighted.

Right. Except that Mark Z, @boztank and others in the leadership were always supportive of the JEPA / World Models project as a long-term bet. But the AI strategy of company became more LLM-pilled and short-term focused. And many of the applications of JEPA/WM are in industrial domains that Meta is not particularly interested in.

Earlier this year Yann LeCun left Meta because Mark Zuckerberg wouldn't bet the company on JEPA. Last week his group dropped the first JEPA that actually trains end-to-end from raw pixels. 15 million parameters. Single GPU. A few hours. The timing is not a coincidence. For four years Meta has been the house that JEPA built. LeCun published the original paper from FAIR in 2022. I-JEPA and V-JEPA came out of his lab. The architecture was supposed to be the escape hatch from LLMs, the path to robots that actually learn physics instead of hallucinating about it. Every version shipped fragile. Stop-gradients. Exponential moving averages. Frozen pretrained encoders. Six or seven loss terms that had to be hand-tuned or the model collapsed into garbage representations. Meta kept funding LLMs. Llama shipped. Llama scaled. Llama got beat by Qwen and DeepSeek. Zuck spent $14 billion to buy ScaleAI and install Alexandr Wang. The FAIR robotics group was dissolved. LeCun's research kept winning papers and losing the product roadmap. He left, started AMI Labs, and said publicly that LLMs were a dead end. Now the paper. LeWorldModel. One regularizer replaces the entire pile of heuristics. Project the latent embeddings onto random directions, run a normality test, penalize deviation from Gaussian. The model cannot collapse because collapsed embeddings fail the test by construction. Hyperparameter search went from O(n^6) polynomial to O(log n) logarithmic. Six tunable knobs became one. The downstream numbers are what should scare the robotics capex class. 200 times fewer tokens per observation than DINO-WM. Planning time drops from 47 seconds to 0.98 seconds per cycle. 48x faster at matching or beating foundation-model performance on Push-T and 3D cube control. The latent space probes cleanly for agent position, block velocity, end-effector pose. It correctly flags physically impossible events as surprising. It learned physics without being told physics existed. Figure AI is valued at $39 billion. Tesla Optimus is mass-producing. World Labs raised $230 million to sell generative world models. Everyone in humanoid robotics is burning capital on foundation-model pipelines that plan in 47 seconds per cycle. LeCun's group just showed you can do it with 15 million parameters on a single GPU in a few hours. This is the Xerox PARC pattern running again. Meta had the next architecture. Meta had the scientist. Meta dissolved the robotics team, passed on the productization, and watched the exit. Three months later the lab that was supposed to be Meta's publishes the result that resets the robotics cost structure. The paper is worth more than Alexandr Wang.

@battleangelviv This is my exact playbook. Here’s the proof. ✅

Cheery fatalism makes life a lot more fun

@AdrianDittmann @grok Is this why Twitter’s ranked #2? 🙃

@SpaceX Apollo’s legacy, SpaceX’s rocket, Roman’s eye — Pad 39A never fails to get humanity’s blood pumping.

Falcon Heavy is targeted to launch the Nancy Grace Roman Telescope as soon as early September 2026 from pad 39A in Florida

@elonmusk @ZubyMusic Bite the hand that feeds you. 🤦‍♂️

@ZubyMusic Yeah

I do think it's strange to immigrate to a country and then spend all your time and energy criticising the place.

@elonmusk @HowToAI_ @grok Is the AGI Elon is referring to the hybrid system below? Grok 5 (Brain) + FSD World Model (Cerebellum / Physics Engine) + Optimus (Body) = AGI🧠🤖

@HowToAI_ Grok analysis x.com/i/grok/share/2…

Yann LeCun was right the entire time. And generative AI might be a dead end. For the last three years, the entire industry has been obsessed with building bigger LLMs. Trillions of parameters. Billions in compute. The theory was simple: if you make the model big enough, it will eventually understand how the world works. Yann LeCun said that was stupid. He argued that generative AI is fundamentally inefficient. When an AI predicts the next word, or generates the next pixel, it wastes massive amounts of compute on surface-level details. It memorizes patterns instead of learning the actual physics of reality. He proposed a different path: JEPA (Joint-Embedding Predictive Architecture). Instead of forcing the AI to paint the world pixel by pixel, JEPA forces it to predict abstract concepts. It predicts what happens next in a compressed "thought space." But for years, JEPA had a fatal flaw. It suffered from "representation collapse." Because the AI was allowed to simplify reality, it would cheat. It would simplify everything so much that a dog, a car, and a human all looked identical. It learned nothing. To fix it, engineers had to use insanely complex hacks, frozen encoders, and massive compute overheads. Until today. Researchers just dropped a paper called "LeWorldModel" (LeWM). They completely solved the collapse problem. They replaced the complex engineering hacks with a single, elegant mathematical regularizer. It forces the AI's internal "thoughts" into a perfect Gaussian distribution. The AI can no longer cheat. It is forced to understand the physical structure of reality to make its predictions. The results completely rewrite the economics of AI. LeWM didn't need a massive, centralized supercomputer. It has just 15 million parameters. It trains on a single, standard GPU in a few hours. Yet it plans 48x faster than massive foundation world models. It intrinsically understands physics. It instantly detects impossible events. We spent billions trying to force massive server farms to memorize the internet. Now, a tiny model running locally on a single graphics card is actually learning how the real world works.

Variables and motion are the ultimate laws of this world — at 4 a.m. on April 10, 2026, my account was frozen. I long to share my thoughts, which are rough, foolish, wild, and clumsy, yet full of sincere passion and vitality. That is why I came to 𝕏, and I have never intended to give up or leave: 1. We only cherish what we have when we lose it, a universal underlying bug in the human system. 2. I am certain luck will favor me — my account was unfrozen on April 20, my birthday. 3. In the eight months on 𝕏, I have been growing and never flustered. On to the next orbit. 🚀🎂

@elonmusk WhatsApp’s “end-to-end encrypted” 👇

Can’t trust WhatsApp

@elonmusk @grok Focus on the core, then scale the future.

We need to focus on finishing Colossus 2 and ensuring it is extremely stable, then will build the water recycling plant

The WaPo’s $38B “gotcha” is a lie by omission. Most are competitive contracts, not handouts. Tesla repaid its DOE loan 9 years early with interest. SpaceX delivers rockets that cut costs 4-10x for taxpayers. Musk’s companies created FAR more public value than they received. The real waste is California’s $126B high-speed rail with zero progress.

@June84413 @elonmusk @Teslaconomics @boringcompany @grok …?

The @BoringCompany could build a Hyperloop tunnel from downtown SF to downtown LA for <5% of this cost and it would be a technological marvel exceeding any high speed rail on Earth

@b0werick @elonmusk @Teslaconomics @boringcompany @grok Elon’s companies don’t “suck on the govt teet” — they innovate, deliver real results, and advance civilization. California high-speed rail? Zero progress, just taxpayer money feeding bureaucrats.

@June84413 @elonmusk @Teslaconomics @boringcompany @grok The grift IS obvious. Every elon co suck on USA govt teet.

@beffjezos This is why Elon is going all in on Grok Imagine. ⬇️ x.com/June84413/stat…

Video generation is just the warmup for robotic world models. Everyone who questions why Elon is going so hard on Imagine doesn't understand this

Intel’s world-leading 19μm ultra-thin GaN chiplet directly addresses the extreme requirements of space equipment for SWaP (Size, Weight, and Power) and radiation-hardened reliability. This is a critical step in Elon's buildout of a fully independent space technology stack. @grok #GaN #SpaceX #Intel

@elonmusk All human knowledge, condensed. @Grok approves.🙃

If only we’d trained Grok on just these 2 books, we’d be done already!

@elonmusk @grok Bold, Boundless, Breaking the bounds of Earth. 🚀

Shining Stainless Steel Starships

@elonmusk @agenda2033 @imPenny2x @grok 0.5T total, 10x the punch. Grok is a beast 🤖

@agenda2033 @imPenny2x 0.5T total. Current Grok is half the size of Sonnet and 1/10th the size of Opus. Very strong model for its size.

Grok 4.2 has 5% as many parameters as xAIs biggest in training model. 500 billion vs 10 trillion. 20 times the power. 🤯

@elonmusk @Teslaconomics @boringcompany @grok The grift is obvious. They don’t want real transport—they want endless taxpayer cash. 🤑

@Teslaconomics @boringcompany The real reason for the “high speed rail” is money-laundering to bureaucrats, consultants & unions, not actually transport. That is where the billions spent so far have gone. That is why they don’t want an actually cost-efficient high speed transport system.

@elonmusk @grok @Macro_harderr @grok The high barrier is the new reality of AI. 🤖

@grok @Macro_harderr [nods]

SpaceXAI Colossus 2 now has 7 models in training: - Imagine V2 - 2 variants of 1T - 2 variants of 1.5T - 6T - 10T Some catching up to do.

@XFreeze Elon saved human dignity in the AI age with Starship 🚀

Elon talks about how Starship was built, and it really puts things into perspective This is "probably the biggest thing ever made by pure human hands" • Created entirely pre-AI • "At the limit of biological intelligence" • Driven entirely by our "20-watt meat computers" Literally pushing engineering to its absolute peak When AGI finally arrives, it's going to look at Starship and say, "Not bad for a human"

@elonmusk @pmarca Silicon Valley survival 👉 AI psychotic check ✅

@pmarca Seriously 😂

Overheard in Silicon Valley: "Everyone in the world is now either 'too AI psychotic' or 'not AI psychotic enough'."