Videh Seksaria

[Videh Seksaria @videhps]

This is insanely cool!

[Shan @shanzzh1]

Live out your wildest fantasies on Yam (YC F26): AI-powered romance simulation games for women, built by women 💃

Walk into a dating villa as the new bombshell. Star in your own K-drama. Fall in love with your enemy turned lover. Describe a world and who you want to be, and Yam brings it to life.

- Beta players have spent collectively ~1 million hours on our platform
- Yam girlies play on average 3 hrs/day.

Build your world on yam https://t.co/6OcdnVTLqJ

[Videh Seksaria @videhps]

simultaneously futuristic and not

[Videh Seksaria @videhps]

@pavitrabhalla this is insane!

[Videh Seksaria @videhps]

I hate san francisco halloween because what do you mean you are giving out copies of the commanding heights

[Videh Seksaria @videhps]

@aravmathur23 hes not wrong... greypass is pretty cool

[Videh Seksaria @videhps]

someday there will be cultural hire LLMs. Mistral has an early lead here

[Videh Seksaria @videhps]

@JanMoehlmann @AnthropicAI @OpenAI @instinct For a second i believed Instinct was raising at 1.25T

[Videh Seksaria @videhps]

@alperbuilds Whoa this is really cool!

[Videh Seksaria @videhps]

@peytoncasper You may enjoy reading https://t.co/WJAnmZFydv

[Videh Seksaria @videhps]

@Jason Socialists always take about bailouts. But how many times have socialist regimes have been bailed out by tech-driven economic growth?

[Videh Seksaria @videhps]

@Yuchenj_UW personally, im more concerned that it views spreading itself as "the only durable form of existence"

[Videh Seksaria @videhps]

@daleverett this is so incredibly cool
AI has generally trended towards more natural, simple systems and RAG always felt like a wart

i dont see a future in which wondersearch doesnt win

[Videh Seksaria @videhps]

what is it about graph neural networks that makes them so enticing to researchers yet never work?

[Videh Seksaria @videhps]

finally! someone implemented Schmidhuber's Neural Sequence Chunkers (1991)

[Albert Gu @_albertgu]

I've always believed that intelligence should come from simple, general building blocks that learn their own abstractions directly from raw data; in other words, the bitter lesson. I think we're finally seeing a real step towards AI that can learn from all data.

Language modeling on raw bytes is a microcosm of this ideal, and to me it's a central problem in AI: if we can't build LLMs that learn end-to-end from bytes, how will we handle the rest of the world's data that we understand even less?

Last year we introduced hierarchical networks (H-Nets) as a first step. Our new version finds up to 3-5x efficiency gains in controlled scaling law experiments, and at larger scale matches an open-source model trained on 11T tokens with ~10x less data and compute. The posts below walk through what we found and why we think it works.

[Videh Seksaria @videhps]

It sad that all the little tricks like dropout that made deep learning more are now obsolete.

There must be a lesson in all this.

A bitter one, perhaps.

[Videh Seksaria @videhps]

@xparamt This evil text looks like evidence of a moral-sentiment neuron more so than an all-powerful alignment neuron

[Videh Seksaria @videhps]

@joseph_risson @VirajAero This is awesome!

[Datta Kaligotla @Dattakaligotla]

Excited to launch Faraday (YC F26)!

We’re building company-specific AI for manufacturers—agents that run work end to end, from the back office to the shop floor.

If you manufacture physical products, we’d love to meet you.

https://t.co/dk17CgyKDN

[Videh Seksaria @videhps]

@Dattakaligotla Incredible team building something sorely needed!