It probably shouldn’t surprise me, but I am often taken aback by people’s misunderstanding of the Burden of Proof. The normal failure mode is “I am presenting this new thing and you need to prove it doesn’t work (is not great, etc.)”. The correct use of Burden of Proof is a requirement that the new… Read More »
MIT Review published an article that covered some of the security issues involved in these tools. Security Intelligence article on OpenAI security breach. Unfortunately, almost all new technologies end up being used first without a lot of thought going into security. For some reason, every new technology I’ve seen in my career starts wide open… Read More »
Survivorship bias is a subtle cognitive bias that appears more often than you might think. Places where I’ve seen it: None of these are wrong, and any of them might work out. But they are only focusing on the survivors. Something like 95% of all startups fail, and you can’t tell which ones will succeed… Read More »
I’m seeing more people becoming enamored with the LLMs, in particular ChatGPT and Copilot. While I can certainly see some interesting results, it seems that people underestimate the fact that the models don’t really know anything. My thought is that these models have actually automated the Dunning-Kruger Effect. The tools can generate confident text (or… Read More »
One approach for design trade-off is between: Until recently, I tended to treat the second and third as the same. With some more thought, I’ve convinced myself they are not.