Picture

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ @_rdm_8

I retweet content I think is important. 😊 Singapore Joined May 2014
  • Tweets
    772
  • Followers
    27
  • Following
    105
  • Likes
    1K
v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
thinkymachines Profile Picture

Thinking Machines @thinkymachines

7 days ago

LoRA makes fine-tuning more accessible, but it's unclear how it compares to full fine-tuning. We find that the performance often matches closely---more often than you might expect. In our latest Connectionism post, we share our experimental results and recommendations for LoRA.…

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
ahatamiz1 Profile Picture

Ali Hatamizadeh @ahatamiz1

6 days ago

Are you ready for web-scale pre-training with RL ? 🚀 🔥 New paper: RLP : Reinforcement Learning Pre‑training We flip the usual recipe for reasoning LLMs: instead of saving RL for post‑training, we bring exploration into pretraining. Core idea: treat chain‑of‑thought as an…

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
cloneofsimo Profile Picture

Simo Ryu @cloneofsimo

2 weeks ago

Damn, very interesting paper. after rapid loss reduction, we see deceleration and follow "scaling law": this is because at these steps, gradients start to conflict each other. Updates are 'fightining for modal capacity' in some sense, and larger the model less fighting there…

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
Tim_Dettmers Profile Picture

Tim Dettmers @Tim_Dettmers

2 weeks ago

Looking closer, PyTorch also uses FP32, but here's the real reason why bnb Adam is better: we optimized for float numerics, order does matter! Computing sqrt(v) + eps*c2 then dividing avoids amplifying errors vs PyTorch's sqrt(v)/c2 + eps. Same math, better stability!

Tim_Dettmers Profile Picture

Tim Dettmers @Tim_Dettmers

2 weeks ago

Looking closer, PyTorch also uses FP32, but here's the real reason why bnb Adam is better: we optimized for float numerics, order does matter! Computing sqrt(v) + eps*c2 then dividing avoids amplifying errors vs PyTorch's sqrt(v)/c2 + eps. Same math, better stability!

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
HKydlicek Profile Picture

Hynek Kydlíček @HKydlicek

4 weeks ago

We are releasing 📄 FinePDFs: the largest PDF dataset spanning over half a billion documents! - Long context: Documents are 2x longer than web text - 3T tokens from high-demand domains like legal and science. - Heavily improves over SoTA when mixed with FW-EDU&DCLM web copora.

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
rohanpaul_ai Profile Picture

Rohan Paul @rohanpaul_ai

a month ago

jax-ml.github.io/scaling-book/g…

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
edwinarbus Profile Picture

edwin @edwinarbus

2 months ago

You can learn more about it here: cookbook.openai.com/examples/gpt-5…

_rdm_8 Profile Picture

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ @_rdm_8

2 months ago

GPT-5 Thinking is incredible! I asked algo interview questions that are asked to SSEs. these are not available on the internet, made up by adding more constraints or twisting familiar scenarios. More than solving the questions, the reasoning it shows gives me goose bumps!

_rdm_8 Profile Picture

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ @_rdm_8

2 months ago

What it's like to Vibe Code youtube.com/shorts/ql56K3s…

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
GoogleDeepMind Profile Picture

Google DeepMind @GoogleDeepMind

2 months ago

What if you could not only watch a generated video, but explore it too? 🌐 Genie 3 is our groundbreaking world model that creates interactive, playable environments from a single text prompt. From photorealistic landscapes to fantasy realms, the possibilities are endless. 🧵

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
Eric_Wallace_ Profile Picture

Eric Wallace @Eric_Wallace_

2 months ago

In releasing this paper and model, we hope that it can aid safety research and serve as useful guidance for other groups looking to release open-weight models. Paper: cdn.openai.com/pdf/231bf018-6… w/ @OliviaGWatkins2 @MilesKWang @kaicathyc @chrisk99999 and many others!

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
AnthropicAI Profile Picture

Anthropic @AnthropicAI

2 months ago

New Anthropic research: Persona vectors. Language models sometimes go haywire and slip into weird and unsettling personas. Why? In a new paper, we find “persona vectors"—neural activity patterns controlling traits like evil, sycophancy, or hallucination.

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
tooliense Profile Picture

Tooliense @tooliense

3 months ago

Now you can just use an agent than can solve olympiad level problems with completely FREE. Also this intelligence can be utilized at coding, science, ...etc any domain you want. We just opensourced our agent system Crux. We don't require you subscribe or any payments. Just…

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
OwainEvans_UK Profile Picture

Owain Evans @OwainEvans_UK

3 months ago

New paper & surprising result. LLMs transmit traits to other models via hidden signals in data. Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies. 🧵

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
bobabowen Profile Picture

Bowen Baker @bobabowen

3 months ago

Modern reasoning models think in plain English. Monitoring their thoughts could be a powerful, yet fragile, tool for overseeing future AI systems. I and researchers across many organizations think we should work to evaluate, preserve, and even improve CoT monitorability.

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
Kimi_Moonshot Profile Picture

Kimi.ai @Kimi_Moonshot

3 months ago

🚀 Hello, Kimi K2! Open-Source Agentic Model! 🔹 1T total / 32B active MoE model 🔹 SOTA on SWE Bench Verified, Tau2 & AceBench among open models 🔹Strong in coding and agentic tasks 🐤 Multimodal & thought-mode not supported for now With Kimi K2, advanced agentic intelligence…

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
buitengebieden Profile Picture

Buitengebieden @buitengebieden

3 months ago

Sharing is caring.. 😊

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
LoubnaBenAllal1 Profile Picture

Loubna Ben Allal @ COLM @LoubnaBenAllal1

3 months ago

Introducing SmolLM3: a strong, smol reasoner! > SoTA 3B model > dual mode reasoning (think/no_think) > long context, up to 128k > multilingual: en, fr, es, de, it, pt > fully open source (data, code, recipes) huggingface.co/blog/smollm3

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
_akhaliq Profile Picture

AK @_akhaliq

3 months ago

discuss with author: huggingface.co/papers/2506.19…

v̴̝̐i̖̍x̘̍t̵̙̖̆̅ retweeted
_AndrewZhao Profile Picture

Andrew Zhao @_AndrewZhao

4 months ago

RL scaling is here arxiv.org/pdf/2505.24864

Trends for United States

$AMD
27,5 B posts
$ZOOZ
#IDontWantToOverreactBUT
#zoozstock
1.011 posts
#Klink
1.372 posts
#MondayMotivation
29,2 B posts
OpenAI
41,7 B posts
Good Monday
41,8 B posts
Victory Monday
2.243 posts
MIU MIU PARIS FW
38,3 B posts
#NationalCoachesDay
The Free Press
6.470 posts
Bari Weiss
4.476 posts
Ghislaine Maxwell
2.216 posts
Cornyn
CBS News
10,7 B posts
Paxton
1.972 posts
Wesley Hunt
StandX
10,2 B posts
National Guard
218 B posts

You might like

Technology Incubation Hub (TIH) @ IIIT Hyderabad on Data Driven Technology | National Mission on Interdisciplinary Cyber Physical Systems (NM-ICPS)
IHub-Data
@IHUB_Data
263 Followers
scaling agentic rl @openblocklabs. prev. @a16z
Daljeet
@daljeet_v
1K Followers
SUSE es líder mundial en innovación #opensource segura, especializada en Business-critical #Linux, Enterprise #ContainerManagement y soluciones #Edge.
SUSE España
@SUSE_Spain
898 Followers
AI will eat your soul if you let it. Thinking is a thought. perpetual motion metamathemagician impartially partisan
even constant
@pfitzart
1K Followers
The translation quality prediction API - ML to translate content more efficiently, at high quality
ModelFront
@modelfront
133 Followers
Alan N
@snalnolan
182 Followers
Tech fangirl since the mid-60's. Even fonder of fair play. Slot machines have better security than our elections. #unhackthevote #NoElectronicsInElections
Joan Gipson Fredin
@JoanFredin
1K Followers
Financial Tech, Sec & symbiotic automation systems. Torontonian, Londoner, NewYorker, Bournemouthian, Ohioan, Bangalorean. #GlobalCitizen & I love our forests.
Chandan Maddanna 🌏
@Chan_Maddanna
1K Followers
Eric Ososki
@eososki17
212 Followers
Tired of all this fascist shit!
Melanie
@mel4uofa
602 Followers
Engineer, Musician, Finance and Investment Savvy! Stock views are personal and not recommendation! Apolitical!
venkateswaran
@hareaswar
250 Followers
mi mamá me enseñó a dividir y ahora soy Físico
El Físico Bizcocho
@RodolfoAntonio9
513 Followers
trapped in a scientific matrix grid in the early phases of being inserted into an extermination system
J
@JayyBerd
354 Followers