Community feed
A focused stream of recent stories from the sources curated for this community. Latest: Introducing Hy4 Preview, Mesh-free model reproduces complex crack patterns in tests, and Fault-driven tests linked to higher scores for a code model. Page 215.
Planet AI Developer Filter
New open weight text input (no vision) LLM from Chinese company Tencent today: 770B total parameters, 49B active parameters, 1M token context window, 1.56TB on Hugging Face. This is a …
Planet AI Developer Filter
Mesh-free neural solver matches key fracture curves and complex crack patterns in tests, but costs roughly an order of magnitude more than finite elements.
Planet AI Developer Filter
A preprint reports higher code benchmark scores after faulty-code test synthesis and dense rewards, while an audit found synthetic tests remained invalid.
Google News Claude Code
Anthropic is cutting Claude Code's current weekly limits by 17% BleepingComputer
Topics: Code Assistants
Planet AI Developer Filter
A preprint reports that a staged AI agent led multi-turn tool tests across model sizes, with its strongest gains on state-heavy tasks but higher latency.
Google News AI Coding
OpenAI Ends Cursor Model Access After SpaceX's $60B Deal Briefs Finance
Topics: IDEs
Entities: OpenAIIDEsCursorSpaceXOpenAI Ends
Planet AI Developer Filter
A preprint presents cQUEDA, a fast model for cavity-QED pump-probe spectra, reporting polariton features and ultrafast dynamics in simulated pyrazine.
Planet AI Developer Filter
A preprint reports naphthalene’s first triplet-state infrared spectrum, distinct from its ground state, with candidate markers for astronomical searches.
Planet AI Developer Filter
A preprint reports a language-model tuning method that forecasts target-task mechanisms from a small probing update before targeted fine-tuning.
Entities: Target
Planet AI Developer Filter
A preprint applies hierarchical causal modeling to Project STAR and reports a 36.80 Mathematics contrast shaped by graph and normalization choices.
Planet AI Developer Filter
A preprint reports higher IAPO scores than matched GRPO on Qwen3 service-agent benchmarks, with comparable multi-turn function-calling results.