RWKV 7B teaser (not final) .... 🤞that we will pass mistral with only 1T tokens huggingface.co/BlinkDL/temp/b…

.Anthropic just published a WILD new AI jailbreaking technique Not only does it crack EVERY frontier model, but it's also super easy to do. ThIS iZ aLL iT TakE$ 🔥 Here's everything you need to know: 🧵

o1 was the first large reasoning model — as we outlined in the original “Learning to Reason” blog, it’s “just” an LLM trained with RL. o3 is powered by further scaling up RL beyond o1, and the strength of the resulting model the resulting model is very, very impressive. (2/n)

AGI ACHIEVED OpenAI just announced the o3 model that broke the ARC AGI benchmark 🔥 this is UNPRECEDENTED.... here's what you need to know 🧵:

When the name Q* was leaked, many thought Q* would be a combination of Q-Learning and A* search, or some other more advanced search powered by RL, over the generation space. I commented that is a dead end. Now my comment should be clear to them.

Raising visibility on this note we added to address ARC "tuned" confusion: > OpenAI shared they trained the o3 we tested on 75% of the Public Training set. This is the explicit purpose of the training set. It is designed to expose a system to the core knowledge priors needed to

Stanford has launched an incredible research AI tool. It’s called STORM, and basically you enter a topic and it will search hundreds of websites to write an article about its major findings. Available to everyone for free!

Microsoft’s new rStar-Math technique upgrades small models to outperform OpenAI’s o1-preview at math problems 🤯 rStar-Math technique enhances small language models (SLMs) using Monte Carlo Tree Search (MCTS) and self-evolution strategies. Applied to models like Qwen-7B and

Phi-4 bug fixes: 1. EOS should be <|im_end|> not <|endoftext|> 2. Pad token EOS should be <|dummy_87|> 3. Chat template shouldn't default add "assistant" & Llama-fied Phi-4 & split QKV to increase accuracy for fine-tuning & made dynamic 4bit quants! Details: 1. The EOS should

MiniMax-Text-01 maintains higher accuracy even with a 1M-token context window, outperforming others like Claude-3.5 and Gemini models, which show steep declines beyond 256K tokens.

1/ Google Research unveils new paper: "Titans: Learning to Memorize at Test Time" It introduces human-like memory structures to overcome the limits of Transformers, with one "SURPRISING" feature. Here's why this is huge for AI. 🧵👇

This blew my mind. PhD student Jiayi Pan reproduced the emergent “thinking” behavior in a 1.5b model using the DeepSeek R1 technique for just $30. This means we can give “thinking” to pretty much any model!! I broke down the findings in my YouTube video below 👇

New research paper shows how LLMs can "think" internally before outputting a single token! Unlike Chain of Thought, this "latent reasoning" happens in the model's hidden space. TONS of benefits from this approach. Let me break down this fascinating paper...

I love using AI so much.

The Command and Conquer source code was open sourced today and it's full of amazing comments 😂 Exhibit A: The "we will fix it later"

Introducing Qwen3! We release and open-weight Qwen3, our latest large language models, including 2 MoE models and 6 dense models, ranging from 0.6B to 235B. Our flagship model, Qwen3-235B-A22B, achieves competitive results in benchmark evaluations of coding, math, general

❄️Introducing Absolute Zero Reasoner: Our reasoner learns to both propose tasks that maximize learnability and improve reasoning by solving them, entirely through self-play—with no external data! It overall outperforms other "zero" models in math & coding domains. 🧵 1/

Absolute Zero Reasoner, the model with no human data. Breakdown below.

Windows Subsystem for Linux is now open source: github.com/microsoft/wsl

It's official: Perifractic's mission to buy Commodore Computers succeeded! Mind-blowing! After decades in limbo, Commodore has a new owner! I personally tried for years without success to buy Commodore, or at least convince wealthy business men to buy it, but I failed