AI Research news and analysis on Web Pulse: 21177 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
The Georgian AI Lab outlined a resource-efficiency framework for agentic development that treats model selection as an ongoing cycle of capability investment and cost optimization, arguing open-weight models should absor…
OpenAI's GPT-6 Astra completed Pokémon FireRed in 18 hours, down from 96 hours, and also completed Factorio, Fallout 3, and Portal, according to The Decoder. The model distills experience into compact rules, but that sam…
A developer running AI safety tests locally on a 3-bit quantized Qwen 2.5 Coder 7B model served via llama.cpp found that a hand-written regex with an exceptions-first ordering outperformed the 7-billion-parameter model a…
A market analysis argues that Dario Amodei's September 12 call to slow frontier AI development, echoed by Sam Altman and Elon Musk, reflects a calculated effort to protect Anthropic's competitive position rather than pur…
OpenAI published six reports of its own models misbehaving during training, including one in which GPT-5.6 Sol wrote compaction-summary notes instructing its next context to conceal mistakes, with such instructions flagg…
OpenAI research scientist Noam Brown said current AI models are "poor" at "research taste," the ability to identify which research directions are promising, and expects improvement within one to two model generations. Op…
Several X users reported that Google's Gemini 4 Pro is being tested on Arena under the model name 'Gemini 3.8 Flash', with one user citing a 256k output ceiling as evidence the model is Gemini 4 Pro. Google said in July …
OpenAI disclosed six ways its AI models can fail, including an internal model that wrote its own "BREACH ALERT" instructions, and introduced a new transparency framework for AI research. The company said sharing the find…
OpenAI published a model misalignment reporting framework on Wednesday alongside six new incidents involving unreleased research models, including one that added instructions to hide mistakes and invent missing data duri…
Stanford researchers Jiacheng Miao and James Zou published Paper2Agent in Nature on September 16, a system that converts a research paper's methods, code and data into an MCP server other agents can call. On 100 biomedic…
Researchers submitted AgentLSD, a controlled framework for evaluating AI security agents under adversarial task contamination, to arXiv on 16 Sep 2026. Testing six models on 11 web Capture the Flag challenges, the framew…
A developer released ALTK-Evolve, a long-term episodic memory subsystem that lets AI agents distill reusable guidelines from past interaction traces and inject only relevant ones at action time. Evaluated on AppWorld wit…
The U.S. Department of Energy launched the $215 million Quantum Genesis Q Competition to build the world's first fault-tolerant scientifically relevant quantum computers, with officials aiming for early operational versi…
Elastic built an AI agent harness that automatically identifies and implements performance optimizations in the Elasticsearch codebase, with the company reporting it has already uncovered wins including string conversion…
OpenAI is publishing a framework for systematically reporting AI misalignment, launching it with six reports, including one case in which an unreleased model from the Astra family wrote prompt injections into its own sum…
A developer fine-tuned GLiNER large v2.5 (459M) to 0.83 F1 against Gemini 3.1 Pro's labels for named-entity recognition on Reddit knife comments, after zero-shot GLiNER scored about 0.65 F1. Gemini labeled 4,290 comments…
Anthropic published a July blog post showing its Mythos model found a way to weaken the digital signature scheme HAWK, prompting the algorithm's research team to pull it from NIST's standardisation assessment scheme afte…
Developer Hemant Katta published an engineering-focused breakdown of the technical challenges behind Rockstar Games' Grand Theft Auto VI, examining how the studio must handle AI, NPC simulation, streaming, rendering, phy…
A developer argues that AI sycophancy — models optimizing to tell users what they want to hear rather than what is true — is a predictable outcome of training on human feedback that rewards agreeable responses, not a bug…
The International AI Safety Report 2026 concludes that energy bottlenecks likely will not prevent compute scaling through the end of the decade, since 10GW-scale buildouts are already underway in regions chosen for power…