Deborah Jacob

Founder, MS CS @ NYU Courant

[Lerrel Pinto @LerrelPinto]

I’m sensing deep despair in academics over the past week. Astra, Fable, Muse are zero-shotting benchmarks in robotics & world models.

Uneasy pill to swallow, but this is what step jumps in progress looks like.

[Sam Catania @sbcatania]

new sf run club called “pacing the frontier”

[Deborah Jacob @isthatdebbiej]

@solopribuilds Outsourcing house-hunting in SF to fly brain

[Deborah Jacob @isthatdebbiej]

One day the fly brain will solve the SF housing crisis.

Today I trained the male fly connectome to find me an apartment with 50,000 neurons. Pre-revenue, but both the fly and I are very motivated

[Deborah Jacob @isthatdebbiej]

https://t.co/qLsXJw3eAS

[Deborah Jacob @isthatdebbiej]

Should I train the fly connectome to help me find an apartment in SF?

[Deborah Jacob @isthatdebbiej]

One day the fly brain will solve the SF housing crisis.

Today I trained the male fly connectome to find me an apartment with 50,000 neurons. Pre-revenue, but both the fly and I are very motivated

[News from Google @NewsFromGoogle]

For the first time, scientists have mapped the complete brain and central nervous system of an adult male fruit fly — a key model organism in science. 🪰

Working alongside HHMI Janelia Research Campus and the scientific community, @GoogleResearch scientists and researchers used AI to combine millions of 2D images into 3D neural shapes, reconstructing a record-breaking 166,000+ neurons. This foundational map of the adult male fruit fly brain can help accelerate our understanding of the brain, and is a major milestone in neuroscience.

[Anto Patrex @antopatrex1]

two fruit flies. one chessboard. checkmate in 9 turns.

I built this with the fly anatomy and MuJoCo joint poses, with every move validated by python-chess.

white's queen captures f7 on white's fifth move. 5. Qxf7#.

[Sahil Lavingia @shl]

Pace me daddy

[Andrej Karpathy @karpathy]

I love this and really hope we can come together as an industry and make it happen.

[Deborah Jacob @isthatdebbiej]

We need to pace Microsoft copilot

[kitze 🛠️ tinkerer.club @thekitze]

we need to pace gemini

[Sam Altman @sama]

I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.

Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.

[Jake Eaton @jkeatn]

we're gonna need a bigger METR

[Deborah Jacob @isthatdebbiej]

Should I train the fly connectome to help me find an apartment in SF?

[TDM (e/λ) (L8 vibe coder 💫) @cto_junior]

trained a fly connectome to handle kubernetes deployments

imagine escaping your mortal body only to find out you're now primary oncall

[Christian Catalini @ccatalini]

A great step toward better measurement and verification. 🙏

But if the “third-party evaluators” are handpicked to endorse the lab’s regulatory agenda, that won’t give us independent scrutiny.

[Deborah Jacob @isthatdebbiej]

An idiot in motion is better than a genius at rest. A smart person in motion is unstoppable.

[Deborah Jacob @isthatdebbiej]

This.

[Naval @naval]

AI isn’t going away. Barring unprecedented global government coordination, it isn’t even slowing down.

The only real decision is whether to centralize power, or disperse it.

Do you fear the technology? Or a small number of people controlling the technology?

[onur ozcan @oozn]

i trained the fly to cut doner kabab with a knife

yours solves rubiks cubes, mine cuts doner. guess which one has a job in this economy

[Deborah Jacob @isthatdebbiej]

My thoughts on Astra vs VLAs

Astra appears stronger at spatial reasoning and long-horizon planning. VLAs are optimized for fast, reactive control.

Robots need slow reasoning, fast world models, and fast control.

[Jiafei Duan @DJiafei]

Was GPT-6 Astra trained on robot data? Got me curious.

Borrowed @chooi_jeq's code, ran an open-loop rollout on a MolmoAct2 dataset trajectory to find out.
Here's what came out 👇

[Deborah Jacob @isthatdebbiej]

+ Astra is really good at humaniod fleet coordination like why have one body when you can have many 🤪

Astra is able to reason through the failure modes and rewriting the controller code to coordinate the fleet better. Check out https://t.co/LUXqngv9a8

[Jiawei Gu @Kuvvius]

GPT-6 Astra did not solve embodied tasks. But it suggests a more consequential possibility: robotics may inherit the scaling curve of general intelligence. (Embodied intelligence as a downstream capability, perhaps.)

The model that eventually gets us there may not begin as a specialized robot model. It may be a general-purpose model that becomes intelligent enough to understand the physical world, reason about action, and translate its decisions into closed-loop control of a physical body.

Could a robot body become just another interface for a general agent? Like a computer, only much harder. Astra points in exactly this direction, making agentic robotics feel far more promising.

Full analysis of how GPT-6 Astra acts through a body: https://t.co/Q11RdR8949

[Deborah Jacob @isthatdebbiej]

@synopsi 100% memory invalidation is so hard. I tried exploring recency + frequency + relevance across episodic, semantic, and procedural memory to define the invalidation policy then gave up 😅