Simon Willison
LLMs can be effectively used for spelling, grammar, and basic fact checking tasks
LLM-generated text now exhibits at least 38 identifiable clichéd patterns
Qwen 3.8 27B performs very well at returning bounding boxes around items in photographs
Qwen 3.8 27B is highly enjoyable and fun to play with as a local model
Qwen 3.8 27B successfully built a script to transform its own .jsonl transcripts to Markdown
Qwen 3.8 27B overthinks tasks even when explicitly instructed not to
Qwen 3.8 27B with 'extra high' reasoning settings exhibits chronic over-thinking behavior
Anthropic's AI detector will be trivial to defeat by iteratively modifying text until it passes
Hugging Face provided detailed independent description of the OpenAI security attack
The UK government AI Security Institute's paper on AI security is credible
Anthropic's AI detector will be trivial to defeat by iteratively modifying text until it passes
GPT-6 class models will be needed to help secure code
QA testing experts can now build software without software developers using AI tools
The 20% serving cost reduction for GPT-5.6 amounts to billions of dollars per month in savings for OpenAI