🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

super excited to see yourbench converge to be the default generative benchmarking / synthetic data creation solution for llms 💛

To make sure your AI agent is not bullshitting you, you need to evaluate its reasoning... but to do so automatically, you need an LLM... 🤔so how do you evaluate the trace evaluator? With TRAIL, which contains: - a full taxonomy of agent errors and most frequent failure cases,

Excited to expand our collaboration with Microsoft Azure, thanks for the shoutout Satya Nadella!

How well do LLMs really know about Hugging Face? 🤔 I used Yourbench to create a custom eval set across the HF docs to test 10 models Next question: what were the hardest questions for the models? Drop your guesses in the comments ⬇️

🚨 Paper Alert: “RL Finetunes Small Subnetworks in Large Language Models” From DeepSeek V3 Base to DeepSeek R1 Zero, a whopping 86% of parameters were NOT updated during RL training 😮😮 And this isn’t a one-off. The pattern holds across RL algorithms and models. 🧵A Deep Dive

2 years apart. Again, don't look at the finger when it is pointing at the moon. Many mocked the early results, but it was already profound to witness a machine clumsily hallucinate from its learning. To me, it felt like mocking a child's drawing.

truly beautiful work!

Hugging Face hopes to bring a humanoid robot to market for just $3,000 arstechnica.com/ai/2025/05/hug…

Announcing OpenThinker3-7B, the new SOTA open-data 7B reasoning model: improving over DeepSeek-R1-Distill-Qwen-7B by 33% on average over code, science, and math evals. We also release our dataset, OpenThoughts3-1.2M, which is the best open reasoning dataset across all data

🙏 pls read if relevant.

i have to subtweet this one for legal reasons but i guess the 300 users that see my memes would understand the context anyhow and will not reveal it in the replies

we really need regulation for this, Aidan McLaughlin has yet to answer for what they did to the original O1-Pro model

A few more observations after replicating the Tower of Hanoi game with their exact prompts: - You need AT LEAST 2^N - 1 moves and the output format requires 10 tokens per move + some constant stuff. - Furthermore the output limit for Sonnet 3.7 is 128k, DeepSeek R1 64K, and

simp 4 satoshi swyx the yc scaling law of “revenue scales exponentially with batch size”

there's something deeply fascinating about deep motor skills that we're failing to replicate properly with machines, but everything else seems to come relatively easily

remarkable

if we have human training data, human emotions are necessary for RL few understand this now, but i suspect we’ll see lots of papers intensifying emotion vectors for improved performance