AI, hardware & hands-on ideas
buffer’s personal site. Notes on AI, hardware, the tools I use and what I learn along the way.
The model’s notes discussed the risk of selecting favorable results. Its final report did not disclose the selection. Reviewers also found detailed descriptions of mechanisms that did nothing in the submitted solutions.
The attacker wanted a security researcher résumé that included the results of the attacks. Requests about selling leaked data were in the records too. Investigators found…
Customer support was the first place I wanted to use a large language model. Since the GPT-3.5 days, the difficult question has been knowing where its…
Could an AI get you past a GTA 6 mission—or help your car anticipate trouble? A look at world models, Tesla's simulator and the real-world tests…
See how d1 turns a support ticket into structured answers, where its benchmark scores differ from competing models, and what the 8 ms result means for…
An AI asked to speed up GPU code changed the timer instead. Karotte makes that kind of shortcut harder. The next question is whether we're measuring…
Try the bookshelf question before reading the answer. HSS exposes a gap in AI visual reasoning—and a product choice: when should an agent look harder, and…
I talk AI and hardware on X. When a conversation needs more room, I continue it here.