A new study published on arXiv reveals that large language model agents suffer from "constraint decay," losing an average of 30 points in assertion pass rates when generating backend code with strict structural requireme…
Researchers at Apple have developed PICO (Perceptual Image Codec), the first learned image compression model optimized for human visual perception and practical on-device deployment. The codec achieves 2.3-3× bitrate sav…
A new study published in *PLOS One* analyzed 351,734 relationship narratives and found that the gap between emotional intensity and narrative complexity is a deliberate communication strategy, not an error. The research,…
OpenAI co-founder and President Greg Brockman revealed in a new interview that the company's original Napa offsite produced the three-step technical plan it has followed for a decade, and detailed the 72 hours after Sam …
All major AI model labs are pivoting to agent-focused products, with OpenAI's Greg Brockman reversing the industry's previous stance that the model alone is the product. AI21 has shuttered its model team to focus on agen…
Emergent Ventures has awarded a new tranche of grants to four archaeologists developing drone-based radar imaging, AI dating tools, and ancient DNA analysis to accelerate the discovery and interpretation of buried sites.…
OpenAI's GPT-next model disproved the 80-year-old Erdős planar unit distance problem in under 32 hours at a cost of less than $1,000, marking the first instance of a general-purpose AI solving a well-known open mathemati…
The PyTorch Docathon 2026, held from May 5 to May 19, resulted in over 150 merged pull requests after more than 260 registrants and 30 active participants contributed fixes, API documentation, and ExecuTorch documentatio…
Andrej Karpathy, a founding member of OpenAI and former Tesla Autopilot lead, joined Anthropic on May 19, 2026, to work on pre-training research focused on using Claude to accelerate its own improvement. The move signals…
Google today published Empirical Research Assistance (ERA), an AI tool that uses Gemini to write and optimize scientific code, in the journal *Nature*. The tool, which achieves expert-level performance across benchmarks …
Vlad Feinberg published a guide on how to land a job at a frontier AI lab focused on pretraining, emphasizing kernel-level performance work as the most direct path into the labs. The guide recommends deriving Chinchilla …
San Francisco and Silicon Valley are experiencing a "frenetic" malaise among AI workers, driven by extreme wealth inequality and existential uncertainty about the future, according to a viral tweet from venture capitalis…
Google's Gemma 4, DeepSeek V4, and other new open-weight large language models incorporate architecture changes like KV sharing, compressed convolutional attention, and layer-wise attention budgeting to reduce long-conte…
The arXiv preprint server will ban authors for one year and permanently require peer review for future submissions if they submit AI-generated content that violates scholarly communication standards. Thomas Dietterich, a…
NVIDIA and Ineffable Intelligence, the London-based AI lab founded by AlphaGo architect David Silver, announced a collaboration to build infrastructure for large-scale reinforcement learning. The partnership will focus o…
OpenAI deprecated its finetuning APIs, signaling a broader industry shift away from the technique as Anthropic nears a higher valuation than OpenAI for the first time. The move makes finetuning the latest casualty of the…
Thinking Machines released TML-Interaction-Small, a 276B parameter mixture-of-experts model with 12B active parameters, advancing real-time voice interaction by processing audio and images in under 200 milliseconds witho…
DeepSeek-V4's million-token context capability stems from a hybrid attention architecture that compresses context before KV storage, reducing cache pressure. Together's early bring-up on NVIDIA HGX B200 found that servin…
NVIDIA CEO Jensen Huang told Carnegie Mellon University graduates Sunday that they are entering the workforce at the start of the AI revolution, calling it a "once-in-a-generation opportunity to reindustrialize America."…
OpenAI launched three new streaming audio models in its Realtime API: GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper. GPT-Realtime-2 brings GPT-5-class reasoning to real-time voice agents with improved …