Anyone else find deepseek v4 flash swears a lot? I keep getting things like “here is a no bullshit answer..” or “mother fucker” a lot .. 💀 my sys prompt is pretty conservative..
This is part of working with the EU AI Act, other labs are adding similar watermarking.
It’s hard to identify AI-generated text, and this gives people better tools to do that. We’ll also be a shipping text detection API that you can use yourself.
"That last call was a mistake — I accidentally launched a duplicate agent with a placeholder prompt instead of waiting. Let me stop it."
Little quirks like this from Claude get a bit annoying. Always needless token burning
Deploying an SPA on cloudflare is so nice. The S3+Cloudfront setup in AWS is a bit annoying. just push a commit, “npm run build” and it’s accessible. I like.
Everything is about screaming how shiny a product is with no regard to the actual user. Don’t really get it, if you want your company to never get in situation like this and sleep easy at night.. just do the basics right.. proof you’ve done the best you can is better than not even thought about it.