Sasha Malysheva

241 posts

Sasha Malysheva banner
Sasha Malysheva

Sasha Malysheva

@aimalysheva

Building something for LLM communication PhD in RL · ex-DeepMind, Google X · @southpkcommons F25

San Francisco Katılım Eylül 2023
542 Takip Edilen508 Takipçiler
Sabitlenmiş Tweet
Sasha Malysheva
Sasha Malysheva@aimalysheva·
Hi all, thanks for checking out my page! Wanted to share a bit more about my background than the X bio allows: - Finished school at 16 - Started as an AI researcher at JetBrains at 18 - By 21 I was coordinating a research direction there and published 5 papers in one year - Then I joined Google DeepMind Game Theory Group - Published a paper in Science at 23 (I honestly thought that milestone would come closer to 40) - Later went to Google X to work on an exoskeleton project (wanted to see what the product side looked like) Now I'm building something new. Stick around ✨
Sasha Malysheva tweet media
English
24
3
133
6.9K
Tamaz Gadaev
Tamaz Gadaev@TamazGadaev·
I remember that a few years ago being an IC programmer meant to sit down, focus hard, and grind through a task by hand. Now I open agent mode in Cursor, have several conversations in parallel, and just talk through everything at once. This changed workflow changes a skillset one should have to be successful in this role, and honestly this suits me better. Single-threaded focus hardly ever was my strong side but multitasking, chatting, and context-switching has always been my thing. As an IC that was a liability then but now it's just the expected stuff.
English
4
0
9
171
Sasha Malysheva
Sasha Malysheva@aimalysheva·
It’s almost weekedn, so here’s one part of my regular routine I’ve kept constant for a long time: Saturday as a rest & recovery day. It's my built-in chance once a week to fully stop, pull myself out of all the noise, breathe, and actually notice my life instead of just living it. If I don't want to take it or can't make it happen, that’s usually a signal that something in my life is off balance. Worth protecting 🫰
English
3
0
7
244
Anastasiia Gaidashenko
Anastasiia Gaidashenko@avgaydashenko·
two ways to keep your Mac awake with the lid closed: 1) Amphetamine (free, App Store): most flexible, works on schedule or by app 2) or just terminal: "sudo pmset -a disablesleep 1" (undo: "sudo pmset -a disablesleep 0") you're welcome 😉
English
4
0
9
108
Tamaz Gadaev
Tamaz Gadaev@TamazGadaev·
extracting a capability from 5% of MLP channels while preserving behavior is something already close to surgical next step I'd want to see is something like a consolidation pass: prune what's redundant after extraction, like sleep-based memory consolidation in biological networks the low-rank circuit conditioning feels like a real step toward that
tokenbender@tokenbender

Ever wondered if you could extract capabilities and behaviors from neural networks and reuse/update/route it as needed? We introduce low-rank circuit conditioning, a novel approach that preserves the model's output behavior while reshaping how an existing capability is represented. In the base model, standard compact recovery stalls at 29%. After conditioning, the same extraction pipeline reaches 91.33% autoregressive full-answer recovery from 5.05% of MLP channels. The evidence points to a possibility of extracting and using isolated capabilities saving cost, latency and high adaptability. Read our work to understand more - tokenbender.com/posts/honey-i-…

English
3
0
7
281
Sasha Malysheva
Sasha Malysheva@aimalysheva·
Hi all, thanks for checking out my page! Wanted to share a bit more about my background than the X bio allows: - Finished school at 16 - Started as an AI researcher at JetBrains at 18 - By 21 I was coordinating a research direction there and published 5 papers in one year - Then I joined Google DeepMind Game Theory Group - Published a paper in Science at 23 (I honestly thought that milestone would come closer to 40) - Later went to Google X to work on an exoskeleton project (wanted to see what the product side looked like) Now I'm building something new. Stick around ✨
Sasha Malysheva tweet media
English
24
3
133
6.9K
Sasha Malysheva
Sasha Malysheva@aimalysheva·
.@theworldlabs is going to outlast most of this landscape, rooting for you @chlassner
Deedy@deedydas

This is the single best read on World Models and one of the most important reads in AI. $10B has flowed into "world models" in the last 18mos, from Yann LeCun to FeiFei Li. The promise is, like LLMs, world models will provide the data it takes to scale robotics foundation models, and solve robotics. ..but the word has been abused to mean one of many things. This post unpacks: – What 5 traits makes a world model? – How do the different approaches stack up? – What is it used for within and beyond robotics? – Where is the opportunity? – Citations to research, news and blog posts Companies / products in the space include: – BigCo products: Google Genie, Tesla Optimus, Nvidia DreamDojo, DreamZero, Microsoft Muse – Pure world model: AMI Labs, World Labs, Runway, Rhoda, Decart, Spaitial, Odyssey, Embo, Dream Labs, OneWorld – Robot foundation model cos: Skild, Physical Intelligence, Figure, Mind Very likely one of the seminal technologies of the next decade.

English
4
0
13
771
Anastasiia Gaidashenko
Anastasiia Gaidashenko@avgaydashenko·
.@MITFutureTech launched the MIT AI Risk Navigator: 1.7k+ AI risks mapped by cause and domain, with incidents and governance docs attached. Genuinely hard to stop clicking through it haha One of potential use case might be as a foundation for something like a policy copilot to help people in government actually work with the legislative and regulatory side of AI more effectively. Worth checking out: airi-navigator.com
Anastasiia Gaidashenko tweet media
English
5
1
8
131

Keşfet

@TamazGadaev @heydyago @avgaydashenko @elonmusk @BarackObama @taylorswift13 @cristiano @BillGates