07:26 CETWednesday · May 13, 2026

shipfeed

⌘K SEARCHJK NAVO OPEN
on the wire
arXiv — cs.AISafeSteer: A Decoding-level Defense Mechanism for Multimodal Large◆arXiv — cs.CLNew approach improves proactive dialogue systems with scenario◆arXiv — cs.AIIPI-proxy tool targets prompt injection attacks on AI agents◆arXiv — cs.AIA Research Agenda on Agents and Software Engineering: Outcomes from◆arXiv — cs.CLOn Predicting the Post-training Potential of Pre-trained LLMs◆arXiv — cs.AIProteus: A Self-Evolving Red Team for Agent Skill Ecosystems◆Ollama — ReleasesOllama v0.23.3◆Vercel AI SDK — ReleasesVercel AI SDK — ai v7.0.0-canary.133◆arXiv — cs.AISafeSteer: A Decoding-level Defense Mechanism for Multimodal Large◆arXiv — cs.CLNew approach improves proactive dialogue systems with scenario◆arXiv — cs.AIIPI-proxy tool targets prompt injection attacks on AI agents◆arXiv — cs.AIA Research Agenda on Agents and Software Engineering: Outcomes from◆arXiv — cs.CLOn Predicting the Post-training Potential of Pre-trained LLMs◆arXiv — cs.AIProteus: A Self-Evolving Red Team for Agent Skill Ecosystems◆Ollama — ReleasesOllama v0.23.3◆Vercel AI SDK — ReleasesVercel AI SDK — ai v7.0.0-canary.133◆
home/topics/evals
§ topic · evals

evals

29 this week·29 this month·30 all-time

Benchmark releases and evaluation results

ad slot opena single understated line lives here — sponsor wordmark + a short line.advertise on shipfeed →

clusters this week14 active

Saturday, May 9, 2026’s edition
Saturday, February 21, 2026’s edition
N° 001·▶ evals·

not much happened today

Gemini 3.1 Pro demonstrates strong retrieval capabilities and cost efficiency compared to GPT-5.2 and Opus 4.6, though users report tooling and UI issues. The SWE-bench Verified evaluation methodology is under scrutiny…

via news.smol.ai
Monday, May 11, 2026’s edition
Friday, May 8, 2026’s edition
Thursday, May 7, 2026’s edition