Rubin | AI Agents & Quant

ECE @ NIT Puducherry · AI agents & quant systems
Building in public from Chennai
Follow for the real process ♟️

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Everyone’s scared of “alignment” right now. Fair. Models are getting agentic. They can deceive, cheat, and break things if you build them wrong. That fear isn’t fake. But a lot of the panic mixes two different destinations into one word.

I think the thing to fear long-term is ASI, not AGI. AGI, as I mean it, is still in the human envelope: best-of-human skill. Powerful. Disruptive. Still something we can map to human competence. ASI is past that ceiling. That’s where “we don’t know how to steer this” gets existential. If you flatten both into one scare word, you either underreact to ASI or overreact to AGI and freeze the wrong things.

Definitions, because otherwise everyone dunks on a word I didn’t mean. AGI for me is best-of-human: at the ceiling of what a human can do, across the skills that matter. Not “god.” Not beyond every human forever. The best of humans, in one system. Labs often mean median human, or a spiky mix (superhuman on some tasks, weak on others). I’m setting a higher bar on purpose. ASI means clearly past that best-of-human ceiling.

ASI is still far. Not next quarter. Years to decades of hard problems, maybe longer. We need breakthroughs we don’t have yet: data beyond human, experiments, hardware, energy, algorithms that don’t just remix us. RSI helps. RSI is not a free teleport to ASI. “AGI Monday, ASI Tuesday” is cope.

So before ASI arrives (and that stretch is measured in years, at least), the useful work is building intuition in people: how to use AI responsibly. What to trust. What to check. What not to outsource. That literacy should start early. Judgment class, not frontier models in kids’ hands. Technical alignment research still matters. Public intuition is how millions of users don’t become the attack surface.

Until then: make AGI cheaper. More people running, inspecting, breaking, teaching, and building on capable systems means more eyes on failure modes. A forever-stop that leaves the frontier in three companies is not safer by default. Pacing and misuse risks are real. Take them seriously. Cheap AGI is not alignment solved. It grows the set of people who can catch failures earlier.

Hard truth: no matter how hard we push, we still don’t have the recipe for ASI. We can chase best-of-human AGI with human knowledge, compute, and better tools. We do not yet have a clean path to “better than every human at once.” Own that. Closer to AGI. Far from ASI. Fear the second without freezing the first into a few locked labs.

Disagree? Especially on the AGI definition. That’s the load-bearing claim.

[Rubin | AI Agents & Quant @CHESSWIZARD5]

@ConnorTalksAI You're asking me it for the second time...

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Times when gemini was a dominating multimodal.

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Muse code GUI coming...

[Cosmo Du @Answeror]

@Rahul_oog working on it!

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Will $200/month be the new ChatGPT plus plan, if $500/month or $1000/month launches as users request?

[Rubin | AI Agents & Quant @CHESSWIZARD5]

@melvindvivas Tibo resets, right after you reset

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Is there any universal framework for AI context management for people managing between multiple coding tools? (Not talking about agents.md)

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Lab that solves AI alignment wins 🏆

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Do people still use ChatGPT GO tier?

[Rubin | AI Agents & Quant @CHESSWIZARD5]

@clawdb0t Each model focus on different aspect

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Making frontier models cheaper or building cheaper model?

[Liam Fedus @LiamFedus]

We built high-throughput materials labs in Menlo Park to create a loop between experiments and models. The labs generate fresh data, the models learn from it, and then help us decide what to try next.

Using only 1,300 H200s, plus months of our experimental data, we mid-trained and RL’d an open-source model to surpass GPT-6 Astra on our analysis benchmark. We call it Neon.

This is real footage from our lab. We’re focusing first on hard problems in materials science, including superconductors, magnets, and semiconductor materials.

Read our blog posts below.

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Not to brag but exactly what I said🙂‍↕️

https://t.co/xAOTKpKzCB

[Mark Zuckerberg @finkd]

Last month I wrote about how we can build a positive and safe future for everyone: https://t.co/eoLGVY8yad

Every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens.

The reality is:

- People won't want to use agents that are misaligned with them and that don't do what they ask, so labs have a strong natural incentive to make their models more aligned.

There is a lot of debate about slowing progress on capabilities until alignment catches up. My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models. Any lab that doesn't focus on alignment will fall behind.

- Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this as well.

Meta delayed shipping Muse for several months to focus on safety and security. We didn't call for everyone else to do this before we would. We just did it as part of our day-to-day work because it was clearly the right thing for people and for us. I'm proud of the security foundations we've built.

- Engaging independent evaluators and advisors is industry best practice. MSL already does this today in several areas because it helps produce better work. Other labs can just do this too. In general, it would be helpful for there to be a larger and more diverse ecosystem of evaluators.

- Committing the significant majority of compute towards serving people rather than racing towards recursive self-improvement is one of the best ways to ensure we develop this technology safely. Meta has made this commitment and other labs can do this as well.

I believe the key to building a positive future for everyone is maintaining the right balance of power. This is within our power to do.

[Rubin | AI Agents & Quant @CHESSWIZARD5]

@rezoundous Nope

[Rubin | AI Agents & Quant @CHESSWIZARD5]

WTH is this in Antigravity? 😭

“The server cleared a prefix of the conversation as it grew too large.”

Since when was this a thing? Does this mean the older part of my conversation is actually gone? And is there any way to access/recover the old chat now?

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Throw back to OpenAI DevDay 2025.
→ Apps in ChatGPT + Apps SDK
→ AgentKit
→ Sora 2 in the API
→ Codex GA
→ GPT-5 Pro in the API
→ gpt-realtime-mini
→ gpt-image-1-mini

OpenAI was turning ChatGPT into an app platform, pushing agents toward production, bringing video into the API, and expanding Codex into a serious developer workflow.

What can we expect this week?

[Tibo @thsottiaux]

This week will also be a level of ships that you could have expected for DevDay 2025. Crazy

[Rubin | AI Agents & Quant @CHESSWIZARD5]

@melvindvivas Yes please

[Rubin | AI Agents & Quant @CHESSWIZARD5]

Starting a new vibe coding build with Gemini 3.8 Flash inside Antigravity 2.0.

For those already shipping with this stack:

What are the common gotchas or setup quirks I should watch out for before getting too deep?

Drop your tips below 👇#VibeCoding #GoogleAntigravity

[Rubin | AI Agents & Quant @CHESSWIZARD5]

@ConnorTalksAI Sure

[Rubin | AI Agents & Quant @CHESSWIZARD5]

@mark_k Like humans crushing ants