Prateek Singhal

Tech Geek | Apple Intelligence | AGI | AR/VR

[Prateek Singhal @psinghal8]

Attending CVPR in Denver next week !

Over the last year, I have been working on Apple Intelligence. It has been a fun ride transitioning back to IC in the age of Claude Code and building on the current LLM stack. We are hiring across the board for strong ML scientists.

Looking forward to catching up with colleagues and friends. Hit me up if you are attending !

#CVPR

[Prateek Singhal @psinghal8]

@PKUWZP Congrats Zhipeng ! Excited about your next adventure.

[Prateek Singhal @psinghal8]

@roeiherzig IPA looks cool too (https://t.co/ZOrgveXI0F)

[Prateek Singhal @psinghal8]

@jon_barron Interesting to think about interactive media as separate, as they are still rendered pixels. Would be great to chat more !

[Prateek Singhal @psinghal8]

At CVPR this week and looking to meet people thinking about:

• social/contextual multimodal agents
• long-term memory and personalization in VLMs
• post-training/RL for multimodal reasoning
• mechanistic interpretability / internal representations
• interaction and social dynamics in video

I’ll be mentoring at LatinX in AI on Wednesday and at Apple Booth (#231) Saturday 2:30–4:30. Otherwise planning to spend time around IPA, MAR/CVinW, embodied-agent sessions, and social/memory posters. Happy to chat.

#CVPR2026

[Prateek Singhal @psinghal8]

@soumithchintala Super cool ! Excited to try it out .

[Prateek Singhal @psinghal8]

@Michael_J_Black Congrats Michael !

[Prateek Singhal @psinghal8]

@Michael_J_Black @CSProfKGD Metals research is cool name to be a director for 😁

[Prateek Singhal @psinghal8]

@LvZhaoyang @xuwu Why do you think it would be 5 years and is wearables not growing enough to keep the ecosystem repurposed for these glasses ?

[Prateek Singhal @psinghal8]

@henloitsjoyce So true ! Great you found your community !

[Prateek Singhal @psinghal8]

@minhsmind You are right that thinking is mostly shared / communicate through words , though for AI to build some form of intelligence, it also needs to experience the world ? That's how world models come in, I assume people who only read with no experience would be better examples ?

[Prateek Singhal @psinghal8]

@sarahookr Indian Accent is probably the best out there for Indian food

[Prateek Singhal @psinghal8]

@pitdesi Did not have a good experience with Kolapasi, not close to good South Indian . Luv the other two , happy to get good Jalebi in the city . Besharam is so good

[Prateek Singhal @psinghal8]

@PranavDixit @VeniGupta5 Hard disagree

[Prateek Singhal @psinghal8]

@DavidSHolz Meow !

[Prateek Singhal @psinghal8]

@RumiLabs_io @MattPoreda It was a great event ! Thanks for having us !

[Prateek Singhal @psinghal8]

@LvZhaoyang Yes, most robotics companies are sort of trying it and it falls within that boundary imo. Defining tasks and using them in a generic manner is what is missing. For example, if a FM knows how to open a jar, the models generalize to most shapes of the jar, but 3dness of it is not.

[Prateek Singhal @psinghal8]

Google hallucinating Alcaraz losing French open is poor from Google @Google