simsa0

🚨 BREAKING: NVIDIA researchers wiped an AI agent's memory after every action. It still scored 70.2 on SWE-bench Verified. And the cost math behind it is brutal. The paper is called "You don't need long context." They ran Claude Opus 4.5 and GPT 5.6 Sol on 13 benchmarks of long coding, agent, and reading tasks. Here is how it works. Instead of keeping the whole history, the agent keeps a short note on where it is, then restarts with a clean slate. Long documents sit in files, and the agent searches them with simple tools like grep instead of reading everything at once. Why bother? Because later tokens cost more. At 256K context instead of 16K, each output token costs 3.6x more on GLM 5.2, 5.6x more on Kimi K2.6, and 7x more on Qwen 3 Coder Next. Here is what happened: » Long tasks: accuracy moved between -3.7 and +5.6 points for Opus, and -3.3 and +4.4 for GPT, across 22 setups » Estimated cost: 2.2x to 8.8x lower for Opus, 2.4x to 5.0x lower for GPT » Needle in a haystack: Opus went from 60 to 100, GPT from 80 to 100 » RULER recall: Opus went from 62.5% to 95.7% » AA-LCR: Opus went from 65.0 to 84.0 Less memory made the models better at finding things. But here is the twist. They also tried one action per context. The agent forgets everything after each step and only carries a short note. On SWE-bench Verified, Opus scored 70.2 against 70.0 for the full-history baseline. GPT 5.6 Sol scored 82.4 against 84.4. The biggest context in those runs shrank from up to 200K tokens to between 15K and 37K. To be fair, this is two closed models. Qwen 3 Coder Next fell from 53.0% to 3.6% on SWE-bench Verified at the tightest budget. The cost numbers are estimates, and the authors say they cannot rule out tasks where long context helps. Their proposal is a model trained to keep its own written summary. They have not built one yet.

🚨An AI just looked at another AI's brain and rewrote it in a few seconds. This is the scariest research paper of the year, and almost nobody is talking about it. Here is what happened in plain English. For the last 10 years, every AI model got built the same way. Billions of training examples. Weeks of compute. Thousands of engineers. The entire AI industry runs on that process. A paper that just dropped shows you may never need to do any of it again. Researchers took a small AI model, let a bigger AI look inside its brain, and asked it to make the small one smarter at math. The bigger AI changed 0.016% of the small one's brain. No training. No examples. No learning. The small AI's math score jumped from 7.9% to 48.1%. Then they tried it on a broken self-driving car system. The bigger AI rewrote 3.5% of its brain and the car started driving properly again. Then they asked it to build small AI brains from scratch. The ones it built beat models that were trained the normal way for weeks. Eight months ago we were worried about AI running its own research experiments. That alone caused OpenAI and Anthropic to panic for an entire summer. This paper skips the research entirely. The AI just rewrites other AI directly, like a kid editing an essay. Independent researchers already reproduced it within days. For the first time, AI can make AI better without humans or training anywhere in the loop. Nobody has a plan for what happens next. Link in the comments 👇🚨

This system uses 17 public skills to automatically generate optimized LinkedIn posts, video scripts, and performance analytics for Codex and Claude. I do ghostwriting and posting for people. This repo is bad for people like me. Someone crammed a whole social media team into 17 free skills for Claude and Codex. It's called Social Media Skills, built by Charlie Hills. 900+ developers rated it already. Here's how it works. Most AI posts are written like "AI slop" because the model knows nothing about you and your ideas. The first thing that I got impressed by is this: the voice builder interviews you and sees how your old posts and your writing style by analyzing five to six of your old posts, and then notes it down into a file and makes it into a structure. Every skill after that goes through that file before writing even a single word. Here's what's inside: 1. Postwriter, generate hooks, plan content. 2. Do research on your basis. In the specific niche, I'm searching 30-40 stories from the last seven days. 3. Generate thumbnails and carousels and infographics. The post-scorer is what got me. It gives you a rating against your old posts before you publish the newer one. Seriously, don't share it. I'd like to keep my clients. 4K stars. MIT License. 100% Opensource.