Boyd’s Blog is the personal website of Boyd Kane, a technical AI safety researcher and computer scientist who writes about expertise, computing, and turning new ideas into practical projects. Its essays explore AI safety research, language model risks, technical systems, and broader questions about technology and society. Topics include how language models might exploit inference engines, ways to evaluate potentially misaligned models, and lessons from AI safety fellowships and research programmes.

The site also presents Boyd’s projects, which combine software, data analysis, and creative experimentation. These include Language Model Melee, an adversarial benchmark for language models; an interactive archive of the Extropians mailing list; ListenToAnything.com, which turns reading lists into podcast episodes; and tools for exploring TikTok trends, Git activity, and South African power blackouts. Other projects cover image provenance, printable elevation maps, and card-game creation.

Boyd’s background includes computer science degrees from Stellenbosch University and research experience through the MATS programme. The site also offers information about his work, technical skills, résumé, and ways to get in touch.