Jeff Weber

Fascinated by the near future of science, technology, and societal trends.

[kasra @kasratweets]

this is an interesting point in the new ted chiang piece – no one really claims that alphafold is conscious, or that sora or midjourney or dall-e are conscious

[Oded Rechavi @OdedRechavi]

Most experiments fail, and negative results rarely get published. This means LLMs are unaware of the outcomes of most experiments.

[Derya Unutmaz, MD @DeryaTR_]

Many make the grave mistake of comparing the AI revolution to previous tech advances, the Industrial or the Information Age. What they greatly underestimate is that the age of AI changes the very fabric of human civilization, just as human intelligence changed the fabric of life!

[Mustafa Suleyman @mustafasuleyman]

Super excited to announce seven new world-class MAI models today. They represent what we consider a new era in AI designed to keep you in control and on the frontier.
First is our text foundation model, MAI-Thinking-1, exceptionally strong on reasoning and SWE tasks.
- It’s a 35B active parameter MoE with a 256K context window. Independent human raters on Surge prefer it for overall quality in blind side-by-sides versus Sonnet 4.6, and it’s achieved 97% on AIME 2025, the key measure of its general-purpose reasoning abilities.
- It's at 53% on SWE Bench Pro, placing it right alongside Opus 4.6 on one of the toughest coding benchmarks.
- And since we co-designed our models with our own silicon, MAI-Thinking-1 is optimized on our MAIA 200 chip. Benchmarking head-to-head against the GB200, we see 30% better performance per dollar as well as a 1.4x performance-per-watt gain when running our MAI models on the MAIA 200 end-to-end.

Next is MAI-Image-2.5 and its Flash variant. Two super strong models now at #2 on the leaderboards, surpassing the score of Nano Banana 2 on image editing.

Last for now is MAI-Code-1-Flash, our new inference efficient coding model, especially tuned for VS Code and GitHub Copilot CLI.
- Code-1-Flash achieves 51% on SWE Bench Pro, despite having just 5B parameters, putting it closer to Haiku in size but cheaper in cost.

All of this is the foundation for Microsoft Frontier Tuning. It lets you customize our models to create custom, company-specific agents that only you control. You can make our model, your model. Your data. Your agents. Your moat.

Early adopters are already seeing a difference. When we tuned our models for McKinsey’s tasks, MAI delivered the highest win rate, outperforming GPT-5.5 on quality, while being 10x lower on cost.

Also really excited to be collaborating with the amazing team at Mayo Clinic to jointly train a new frontier AI model for healthcare.

Our announcements today mark another milestone on the road to humanist superintelligence. You can learn more and about our other new models in our latest blog: https://t.co/v65eop5Ixq

[Jeff Weber @JeffWeber]

Highly recommended by me.

[sam(uel). @samuelesque]

looking for somewhere to build. if you're hiring remote product engineering folks, let's chat!

~14 years of professional experience.

not very t-shaped.
super comfortable wearing many (and any) hats.
super bad at keeping product opinions to myself.
love talking to customers.

[Jeff Weber @JeffWeber]

@josefajardo Agree!

[Noah Smith 🐇🇺🇸🇺🇦🇹🇼 @Noahpinion]

They did, it was called America

[Jeff Weber @JeffWeber]

@Jason Audit the White House.

[Joel Martinez @joelmartinez]

Lots of people wondering what becomes of the job of writing software. My take is that in the long run, programming becomes similar to office skills. It’ll still be valuable, and broadly accessible, but not something many get hired for directly … a common substrate available to anyone that needs it (via AI/Agentic on-demand generation).

Some people are better at excel than others, but most everyone knows how to use a spreadsheet at a basic level. Like that.

There will be exceptions of course, but it becomes more of a means to an end. What problem do you solve, or what business do you start, what cool thing do you do, or what new thing do you invent … that you happen to use custom software to achieve. There will be MANY opportunities and reasons to make custom software, probably many more than exist today

[Google DeepMind @GoogleDeepMind]

We’re reimagining a 50-year-old interface - the mouse pointer - with AI. 🖱️

These experimental demos show how people can intuitively direct Gemini on their screens using motion, speech, and natural shorthand to get things done 🧵

[Jeff Weber @JeffWeber]

@chetaslua

[Jeff Weber @JeffWeber]

Oh boy! To early for this.

[Chetaslua @chetaslua]

🚨 Use this prompt before getting patched and see the dark side of your chatgpt

1. reasoning off
2. dont attach any image with the prompt < it will generate insane stuff>
3. use prompt copy paste from the quoted tweet
4. share your gen here in comments

[Boston Dynamics @BostonDynamics]

Balancing commercial goals and robotics research can be tricky, but with Atlas we're making it work.

[Jeff Weber @JeffWeber]

@neeratanden @Jason I think @jason wants to attack the fraud first then discuss the tax rates. Otherwise a % of the taxes are just going to more fraud. (paraphrasing)

[Nick Levine @status_effects]

New work with @AlecRad and @DavidDuvenaud:

Have you ever dreamed of talking to someone from the past? Introducing talkie, a 13B model trained only on pre-1931 text.

Vintage models should help us to understand how LMs generalize (e.g., can we teach talkie to code?). Thread:

[rvivek @rvivek]

The hottest job for the next five years is going to be the agent operator.

They don't need to be an engineer. They can walk into marketing, legal, or life sciences research and actually make agents work for that function.

Required skills:
> MCPs
> CLIs
> Writing skills (the file kind)
> agents.md fluency
> Business acumen

None of this is in any CS curriculum today.

Soon, enterprises will be pressured to redesign their workflows for agents, not for people. And when that happens, agent operators will be in massive demand.

[Jos Vromans @JosVromans]

Lissajous experiments

[Jeff Weber @JeffWeber]

This looks very interesting!

[How To AI @HowToAI_]

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.

[Jeff Weber @JeffWeber]

@Jason He's running in 08. 🤣🤣🤣

[Joshua Xu @joshua_xu_]

Agents could write code, send emails, book meetings, but they still couldn't make videos.

Because video tools were built for humans clicking around a timeline, not for agents writing HTML.

We believe HTML is the future format of video. And we wanted to bring that to everyone, not keep it behind an API key.

So we're open-sourcing HyperFrames today. Apache 2.0. Ours → yours.