GPT-5 still can’t say how many B’s are in blueberry on its first try.
ChatGPT’s User-Agent… Obfuscation

ChatGPT’s User-Agent… Obfuscation

If you ask ChatGPT to “please fetch https://whatmyuseragent.com/” in regular mode, it gives an answer like Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; ChatGPT‑User/1.0; +https://openai.com/bot Notice it clearly labels itself ChatGPT....
AI Small Wins; Real World Difference

AI Small Wins; Real World Difference

I’m deeply skeptical of most commercial claims about AI “personalizing” education—at least in the short term. My doubt isn’t about the underlying technology, but about how poorly it’s likey to be executed. (I ought to write another post on this, because it also shows...
OpenAI Has No Moat (And Sam Knows It)

OpenAI Has No Moat (And Sam Knows It)

This might surprise many, but the state of OpenAI large language models are not far enough ahead of their competitors to be a serious barrier to competition. I use models from OpenAI, Anthropic, and Google every day. When any one of them goes down, I can swap over to...
Give Claude Code Access to Gemini

Give Claude Code Access to Gemini

One of the best things I’ve done recently is to install Simon Willison’s llm tool in my development environments and give Claude Code instructions on how to use it. I set the model to gemini-2.5-pro-preview-05-06 by default, but I plan to try the latest...
AI: It’s Upcoming Impacts On Small Businesses

AI: It’s Upcoming Impacts On Small Businesses

This essay is the result of months of thoughts and now about three years of work with AI (namely, LLMs), and 15 years in the MSP industry. It is an attempt to sketch out the future as I see it coming. This will dive into some of the technical, but as high of a level...