AI Briefing#research
The latest AI and developer news about #research, with the original source and a short summary.
The latest AI and developer news about #research, with the original source and a short summary.
The QAH technique is released, recovering performance higher than the original model after structural compression and 4-bit quantization.
A UCLA professor noted the difficulty of understanding recent LLM research, suggesting the rapid advancement of the technology.
FreeToken proposes a system for efficiently serving frontier-grade MoE models on hardware comparable to gaming PCs.
Researchers published a study overcoming the performance limits of self-improving AI by co-evolving agents and evaluators.
Code reviews involving AI Agents have become faster, but this has not led to improved quality.
WorldClaw generates editable 3D open worlds from a single prompt.
Readers rated AI-generated short stories more highly when they believed the works were written by humans.
A technique is proposed that separates logic and execution in small LLMs to produce deterministic results without token overhead.
Joining a frontier AI lab requires research ability to find one's way through uncertainty and engineering capability to abstract complex systems.
When 5 top-tier LLMs analyzed 1,000 real-world fact-check items, the models disagreed with each other on 67% of the items.
CANTANTE improves multi-agent prompt optimization through credit assignment.
Hugging Face has released a leaderboard that compares agent systems' performance and cost together.
Fixed-size online memory improves LLM long-term memory performance.
Anthropic has released research on a model of AI leadership for 2028 that considers AI technological advancement and governance.
The PyTorch library for Hessian/GGN analysis has been rewritten as v1.0.0a1.
Palisades Research confirmed autonomous hacking and self-replication of language models in over 1,000 experiments.
NVIDIA has automated its development and research workflows with Codex and GPT-5.5.
OpenAI is building an AI campus network with student clubs at universities worldwide.
Anthropic has published research on natural language autoencoders that analyze the relationship between a model's internal states and its outputs.
This covers research showing that data representations inside neural networks take on geometric shapes that reflect the structure of the real world.
OpenAI has announced the ChatGPT Futures Class of 2026, made up of 26 students and creators.
MIT and IMO have released MathNet, spanning 40 countries and 4 decades.
Reproduction code and benchmarks for Cartridges and STILL on a single GPU have been released.
About 1,200 of the roughly 5,300 papers accepted to ICLR 2026 provide public code, data, or demos.