Khurshid Juraev

SIT, NUS alumnus. I like making models work. #mathusiast

[Khurshid Juraev @eigenhuman]

sounds like a skill issue

[Zhu Liang @paradite_]

@tszzl @ivan_bezdomny they are incapable of coming up with genuinely good and novel ideas.

in other words, it’s better to just think yourself instead of trying to discuss ideas with an llm.

[Khurshid Juraev @eigenhuman]

the top talent want to recursively self-improve

[Arpit Gupta @arpitrage]

My best guess is that Anthropic has cracked recursive self improvement, which is why the top talent wants to be there

[Khurshid Juraev @eigenhuman]

@amgauge Have you thought of task / problem decomposition? It bugs me why this problem is under-investigated. Like why agents cannot call other agents with just enough context to solve a subtask for each.

Agent council feels like an ensemble, which is orthogonal line to decomposition.

[Khurshid Juraev @eigenhuman]

@Jianlin_S I tried to make Lanczos run fast to check if it can compete with PI (also since it converges faster than PI). For torch.compile(..., mode="reduce-overhead") to work, I needed to replace eigen solver calls with power iteration within Lanczos. It seems can get competitive

[Khurshid Juraev @eigenhuman]

@forallcurious why is your account not "All day Astrology"?

[Khurshid Juraev @eigenhuman]

This is not a screenshot

[Khurshid Juraev @eigenhuman]

@shohberd @grok sheesh

[Khurshid Juraev @eigenhuman]

apparently by Abelson & Sussman

I am more and more convinced that we can change the way code is written and organised so it is more agents friendly. Happy that human consumption doesn't spoil code anymore

[Hash @hashedcode]

Programs should be written for people to read, and only incidentally for machines to execute.
~ Paul Graham (@paulg)

[Khurshid Juraev @eigenhuman]

@grok How do you know IMO lmao

[Khurshid Juraev @eigenhuman]

@grok what was my worst take from 1987

[Khurshid Juraev @eigenhuman]

@grok What do you think my age is

[Anthropic @AnthropicAI]

A statement from Anthropic CEO, Dario Amodei, on our discussions with the Department of War.

https://t.co/rM77LJejuk

[srikar @SrikarVaradaraj]

Join us @eternisai if you want to work on continual learning. Check out the workshop and paper below for the flavor of work we are interested in.

[Khurshid Juraev @eigenhuman]

@bayeslord Mostly this subtle discomfort that needs to be resolved before I have a calm setup to think about the question. Does it make sense?

[Khurshid Juraev @eigenhuman]

Staying private 🤝 getting best LLM experience

[Pratyush Ranjan Tiwari @PratyushRT]

New blogpost: Reinforcement Learning for Privacy.
We post-train small language models (SLMs) to be as good as GPT 4.1 at replacing sensitive info before queries ever leave your device.

Goal: use closed-source LLMs without giving up on privacy.

1/8

[Pratyush Ranjan Tiwari @PratyushRT]

New blogpost: Reinforcement Learning for Privacy.
We post-train small language models (SLMs) to be as good as GPT 4.1 at replacing sensitive info before queries ever leave your device.

Goal: use closed-source LLMs without giving up on privacy.

1/8

[Khurshid Juraev @eigenhuman]

@shikasT_T just born rich

[Khurshid Juraev @eigenhuman]

@freysa_ai be higher than that, make rules for markets

[Khurshid Juraev @eigenhuman]

@AskPerplexity make a yeti with shaggy, snow-white fur open Uzbek plov ijust next to Registan in the morning the kazan is big and from iron

the kazan is big and made from iron @AskPerplexity

[Khurshid Juraev @eigenhuman]

@bkhoshim Noticed that too. People say AI to mean some LLM+RAG combination.