for the open reasoning models, response quality stayed the same not "slightly degraded." same. and here's the number that stops you: omitting assistant-side history reduced cumulative context lengths by up to 10x 10x less context. same quality. on real conversations.
RESEARCH
-
Real Chat Tests on Four AI Models
By
–
the experiment is clean they took real multi-turn conversations from WildChat and ShareLM. not synthetic benchmarks. actual human-ai chats then they ran every conversation two ways across four models (Qwen3-4B, DeepSeek-R1-8B, GPT-OSS-20B, and GPT-5.2): > full context: normal.
-
MIT finds LLM context pollution degrades performance
By
–
MIT researchers discovered a phenomenon called "context pollution" where llms get WORSE by reading their own prior responses errors, hallucinations, and stylistic artifacts from earlier turns propagate forward because the model treats its own output as ground truth and removing
-
Video Models Over Language Models for Robotics Manipulation Tasks
By
–
Taking inspiration from VideoJAM, would the physical consistency of the generated videos also improve? Accurate action prediction requires physically plausible imagination, conversely physically plausible imagination is best supported when it is consistent with feasible actions. Seonghyeon Ye (@SeonghyeonYe) VLAs (from VLMs) ❌ => WAMs (from Models) ✅ Why WAMs? 1️⃣ World Physics: VLMs know the internet, but Models implicitly model the physical laws essential for manipulation. 2️⃣ The "GPT Direction": VLAs are like BERT (rely heavily on task-specific post-training). WAMs are like GPT (pre-train & prompt), unlocking incredible zero-shot transfer! What I want to see in 2026: 📈 Scaling Laws: We will see much clearer scaling laws for robotics compared to VLAs. 🤝 Human-to-Robot Transfer: Unlocking massive transfer capabilities using video as a shared representation space. 🤖 Zero-Shot Mastery: Moving from short-horizon tasks to long-horizon, dexterous manipulation without task-specific demonstrations. We recently open-sourced the checkpoints, training and inference code. Dive into the research! 👇 📄 Paper: arxiv.org/abs/2602.15922 💻 Code: github.com/dreamzero0/dreamz… 🤗 HF: huggingface.co/GEAR-Dreams/D… — https://nitter.net/SeonghyeonYe/status/2024501978106061056#m
→ View original post on X — @shiqi_yang_147, 2026-03-03 10:53 UTC
-
AI Transformer Acts as a Functional CPU
By
–
A UW-Madison professor asked two AI agents to train a transformer that works as a literal CPU. Not a text predictor. A computer that takes programs as input and executes them. Running Fibonacci, division, sqrt on programs it never saw during training. 99.5% accuracy. Weights
-
Code generation efforts alongside parser and semantics development
By
–
Parser and semantics yes, but there were some serious code generation efforts.
-
AI’s Inevitable Reality: Beyond Normative Arguments
By
–
I wasn’t making a normative argument. Of course I think this is bad. I was pointing out what will inevitably happen with AI in reality
-
Computer Use moving from research to production with impressive demos
By
–
We're soon going to see Computer Use move from research to production use cases.
— m_ric (@AymericRoucher) 2 mars 2026
Standard Intelligence's demos are stunning, the model seems to have a very natural way of interacting with the computer. It can even drive a car through a webapp!
And Anthropic is working on it as… pic.twitter.com/FeGvZVFNeNWe're soon going to see Computer Use move from research to production use cases. Standard Intelligence's demos are stunning, the model seems to have a very natural way of interacting with the computer. It can even drive a car through a webapp! And Anthropic is working on it as
-
Context engineering is the new bottleneck
By
–
the broader observation matters more than this specific paper. context engineering is quietly becoming the real bottleneck. not model capability, not training data, not inference cost. the unglamorous plumbing of what information reaches the model, when, and how. Anthropic
-
Three Tiers of Persistent Context Explained
By
–
the paper defines three tiers of persistent context, each with different lifecycles. scratchpads (/context/pad/) are temporary working notes scoped to a task. think of them as the agent's rough draft space. episodic memory (/context/memory/episodic/) holds session-bounded