Amanpreet Singh

something new

[Amanpreet Singh @apsdehal]

@williamfalcon @kchonyc Congratulations Will on the great run!
I still fondly remember building our multimodal framework internally at FAIR on top of Pytorch Lightning many years. Thank you for your contributions!

[Amanpreet Singh @apsdehal]

@Madeactual That's a great idea and can improve accuracy and trust on direct system of record edits.

[Amanpreet Singh @apsdehal]

A year ago, everybody was pushing chat as a universal interface in their applications. Every system of record, every SaaS application.

But, as agents start to do actual work, the reverse is happening and the UI is now showing up embedded in the chat. 1/n

[OpenAI @OpenAI]

GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone.

Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.

[Amanpreet Singh @apsdehal]

wow

[Rikiya Takehi @rikiyatakehi]

We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.

We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: https://t.co/bOWCcFhCVq

[Amanpreet Singh @apsdehal]

Building for knowledge work? it's an important question for your platform: when should the user talk to an agent, and when should they be able to interact directly with what it made?

We have a strong thesis here and have built a solution around it. Would love to chat. DM! 9/n

[Amanpreet Singh @apsdehal]

Finally, there is another benefit here: implicit feedback.

Through user edits you can see what they wanted to change and why agent got it wrong. These become an implicit signal you can learn from, without making the person explain the same correction again. 8/n

[Amanpreet Singh @apsdehal]

Good article. It gives you a glimpse of how the interaction with agents will look in the future as they start doing more work.

Super Agents are emerging as a new interaction layer across the board: projects in Claude Code and Cursor also build on the same idea.

As we trust agents more, the bottleneck shifts from doing the work to reviewing the work. Over time, the super agent needs to be proactive while only highlighting and bubbling up new exceptions, questions, and edge cases to you.

[DeepSeek Harness @DeepSeekHarness]

https://t.co/gwwIHtyCRQ

[Gurnoor Singh @gurnoorsingh123]

Had a great time speaking at Bench Summit today about the challenges of evaluating AI agents in real-world environments!

Lots of interesting conversations with people building and evaluating agents at the frontier.

Thanks to @kaggle, @GoogleDeepMind, and @zapier for putting together a great event!

[Amanpreet Singh @apsdehal]

@Capslock404 Interestingly, humans have solved this problem. We are able to teach and transfer this tribal knowledge to the new hires very efficiently. We show each other how we work, ask clarifications and give feedback in the real time. The question is what does that look like for agents.

[Amanpreet Singh @apsdehal]

Eliciting the tribal knowledge in people's heads that was never written down anywhere is an art and the biggest barrier to the diffusion of AI in enterprises.

[gabriel @gabriel1]

"company brain" startups that construct information from internal messages and documents when they realize 90% of information only exist in peoples heads: 😭😭😵‍💫😵‍💫

[Amanpreet Singh @apsdehal]

https://t.co/Evd4yfKHSp

[Amanpreet Singh @apsdehal]

@stanichor785 @pangram It's handwritten. I have a series of notes on the same topic. My guess is the pattern around "it's not X, but Y" triggers it.

[Amanpreet Singh @apsdehal]

This is good timely note on trustworthiness and safety of enterprise agents.

As models and harnesses commoditize, we need to rethink how we can teach agents efficiently and verify their work at scale. It’s not just what an agent can access or do, but who can teach it and who can update its behavior.

Our access controls and interfaces also need to evolve with this. The people teaching and updating the agent might not be the same ones verifying its work and signing off on it. The system needs to enforce that separation allow verifying the work efficiently.

This is a layer that should work across models, learn from demonstrations and corrections, and provide decision trails as it evolves: what changed, why, who taught it, who verified it. It should also capture how the agent made decisions and on what basis when it did the work. Over time, through learning, it should reduce the verification burden as well.

If you can verify it, you can solve it through AI. If you can verify it at scale, you can automate it through AI. Otherwise, every update to the agent becomes another thing we have to trust without being able to evaluate or verify.

[Satya Nadella @satyanadella]

https://t.co/MjwFy73LOF

[Amanpreet Singh @apsdehal]

@LeoTronchon @spc @finn_meeks @gopalkraman @adityaag Congratulations! Looking forward to what you build.