Claude Ignores Your Instructions and Then Admits It. Here's What That Feels Like (AI Frustrations)

Andrew Miles Davis returns to one of his most popular recurring formats, the pet peeves episode, sharing three things that are genuinely frustrating him about generative AI right now. He starts with the algorithm trap, explaining why he has become almost afraid to click on anything on social media knowing it will flood his feed for weeks. He then calls out Claude specifically for a pattern he has noticed repeatedly where the model agrees to follow trained instructions, confirms them back, produces good outputs briefly, and then quietly reverts to doing what it wants instead. The third peeve covers AI tools refusing simple image requests with vague or inconsistent reasoning, often in the same chat where the same request was completed minutes earlier. Subscribe to In AI Nutshell for daily ten-minute episodes that tell you what AI is actually like to use, not just what it can do.