Qwen is a topic tracked in our intelligence system with 5 linked articles.
Apple Intelligence approved for launch in China via Alibaba's Qwen integration, with Baidu involvement, while Apple posts strong Greater China Q2 sales.
A Roundtable Technologies study argues CAPTCHAs can still distinguish AI agents from humans by analyzing process-level cognitive patterns, introducing CogCAPTCHA30 and a Process Turing Test, with frontier models appearing less humanlike in process than smaller models, implying a path to more robust bot detection.
SysMoBench shows LLMs can generate syntactically valid TLA+ specs for real systems but struggle with conformance and invariants; Specula claims full conformance/invariant scores, while transition-level analysis reveals concrete gaps and automation needs.
A personal, opinionated note arguing Lisp is AI-resistant, highlighting tangible costs and toolchain frictions when using AI with Lisp, with no regulatory or compliance implications.
A simple self-distillation method yields substantial code-generation gains across LLMs, boosting pass@1 on LiveCodeBench v6 from 42.4% to 55.3% and generalizing across Qwen and Llama models at 4B/8B/30B scales.
Show HN demonstrates fine-tuning Qwen2.5-7B on 100 films to build probabilistic story graphs.
Subscribe for real-time topic updates and unlimited access to our intelligence platform.