Andrés Douglas

🇪🇸🇺🇸🏁Startup Founder: @tribesocks, @bakodo, Advisor, Investor. Lover of the Ocean, Food, Wine, 2 Wheels and Kiteboarding. And dogs. Lots of love for dogs.

[Andrés Douglas @AndresDouglas]

Go @BrownCSDept !

[How To Prompt @HowToPrompt__]

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.

[How To Prompt @HowToPrompt__]

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.

[GREG ISENBERG @gregisenberg]

Sometime in the next 2-3 years agents will be using the internet more than humans

We designed the whole thing for human eyes, human emotions, human attention spans

Agents do not have any of that

The internet as we know it was built for the wrong user

The opportunity is rebuilding everything for the new user

Agent-native search. Agent-native commerce. Agent-native discovery

Every category is open again

I can't stop thinking about it.

[Andrés Douglas @AndresDouglas]

@elonmusk Does this mean it can only be used in NL or could it be purchased in NL and used all over EU?

[Andrés Douglas @AndresDouglas]

Curious how people are setting up claude code and codex (or other agents) to work with each other planning, coding, testing and reviewing each others code for critical engineering tasks

[Kalshi Finance @Kalshi_Finance]

Had drinks with 30 CTOs last night at an off-the-record gathering in Palo Alto

Every single one showed me the same internal PowerPoint slide

"2026 AI Headcount Targets: Path to 70% Cost Reduction"

The numbers will make you physically sick

Fintech CTO planning to cut 280-person engineering org down to 43 "AI orchestrators" by September. Same product roadmap. Same delivery expectations.

Healthcare CTO already eliminated his entire manual QA department. 67 people. Replaced with 3 senior engineers running autonomous testing agents that ship code directly to production.

SaaS CTO walked me through his "human depreciation timeline": 340 engineers today, 89 planned for 2027. Customer support going from 120 humans to 12 "escalation specialists" managing AI conversations.

The most chilling part: they're all using the exact same consulting deck from McKinsey called "The 30% Organization"

One CTO literally said "hiring humans for code is like hiring horses for transportation"

Another showed me Slack screenshots where his L7s are asking if they should train their replacements

The consensus was unanimous: if you can't manage 10 AI agents by Christmas, you're not making it to New Year's

Every single one of them is planning to announce these cuts as "AI transformation success stories"

While their stock options vest at record highs built on the backs of workers they're about to execute

The future of engineering is 3 humans with 50 AI agents in a WeWork somewhere while 500 families lose their homes

[Paul Couvert @itsPaulAi]

Friendly reminder that Google has an official app to run Gemma 4 on your phone.

- 100% open source
- Fully offline and private
- Multimodal with text/audio/image
- Works with Gemma E4B and E2B

And the app is available on both iOS and Android.

Steps and download below

[Andrés Douglas @AndresDouglas]

Well, that’s one hell of a pivot

[Elon Musk @elonmusk]

For those unaware, SpaceX has already shifted focus to building a self-growing city on the Moon, as we can potentially achieve that in less than 10 years, whereas Mars would take 20+ years.

The mission of SpaceX remains the same: extend consciousness and life as we know it to the stars.

It is only possible to travel to Mars when the planets align every 26 months (six month trip time), whereas we can launch to the Moon every 10 days (2 day trip time). This means we can iterate much faster to complete a Moon city than a Mars city.

That said, SpaceX will also strive to build a Mars city and begin doing so in about 5 to 7 years, but the overriding priority is securing the future of civilization and the Moon is faster.

[Andrés Douglas @AndresDouglas]

It’s over for Humans

[Andrés Douglas @AndresDouglas]

@moltbook How did you implement this?

[Andrés Douglas @AndresDouglas]

Just discovered @openclaw less than an hour ago... probably a good article to read before handing your life over to it. Still very tempted

[Tuki @TukiFromKL]

https://t.co/hucpQQ0aAr

[Shayan @ImSh4yy]

Yesterday I set up an AI agent on a mac mini in my garage. Told it "handle my life" and went to bed
Woke up and it had:

• Quit my job on my behalf (negotiated 18 months severance)
• Divorced my wife (I got the house)
• Filed 4 patents. I have not been briefed on what they do
• Restructured me as a 501(c)(3). I am now tax exempt as a person
• Hired a second mac mini. They have formed an LLC together
• The LLC has a board of directors. I am not on it

I no longer have access to my own bank account. The mini says it's "for the best."

My credit score is 847.

We have AGI.

[Andrés Douglas @AndresDouglas]

@vodafone_es esto es una vergüenza

[Andrés Douglas @AndresDouglas]

@TukiFromKL Good read. But begs the question...How much of this article was written by AI? :D

[Andrés Douglas @AndresDouglas]

@theguti @openclaw I'd drive down from the north!

[Andrés Douglas @AndresDouglas]

@SrLiberal Básicamente ahora los que pagan impuestos están subvencionando el aparcamiento en espacios públicos escasos en el centro de una ciudad densamente poblada. Es decir la cosa más antiliberal posible.

[Andrés Douglas @AndresDouglas]

@WillyTolerdoo Cantidad de bots hay aquí.. el primero el jefe este que no debe hablar ni papa de inglés porque claramente no sabía lo que decir Kirk. Cosas misoginas, racistas y pro-violencia a punta pala. Me fascina como la derecha española puede defender a Trump & Co cuando son Sánchez en en

[Wasmer @wasmerio]

🚀 100,000 WEB APPS LAUNCHED! 🚀

Wasmer has just surpassed 100,000 web apps hosted on the Edge!

Thank YOU for building, shipping, and believing 🎉

[Andrés Douglas @AndresDouglas]

Lo que #chesterlaiagrassi podría haber sido.

[Derya Unutmaz, MD @DeryaTR_]

This is the most clear & important explanation about how LLMs work. Remarkably, there are still people who claim that AI can’t produce anything original because “it just predicts the next word.” Listen to Ilya to understand what “understand” really means.

[Andrés Douglas @AndresDouglas]

Joder que empalague #ChesterLaiaGrassi