EpeeGnome

@EpeeGnome@feddit.online

• ⁨Comment⁩ on ⁨ChatGPT blindly trusts browser content, turning the page into a payload⁩ ⁨⁨1⁩ ⁨week⁩ ago⁩:

  Turned off my Google AI summary blocker and searched for this article by title. It didn’t load an AI summary. I search for random other stuff and get an AI summary at the top. Not sure what’s happening there. Some sort of attempt at security?

  Prompt injection is a hilarious and, as far as I can tell, somewhat unavoidable consequence of using LLMs.

• ⁨Comment⁩ on ⁨Lasagna dreams, spaghetti reality⁩ ⁨⁨2⁩ ⁨weeks⁩ ago⁩:

  I actually have a reasonable number of hours in Dwarf Fortress and Satisfactory. Factorio though, yeah…

• ⁨Comment⁩ on ⁨Vibe coding service Replit deleted production database⁩ ⁨⁨10⁩ ⁨months⁩ ago⁩:

  It wasn't the user's infrastructure, it was the LLM company's. The selling point is that it's all integrated together for you. You explain what you want, the LLM not only codes it, but launches it too. Yes, his screen shots of the LLM "taking responsibility" are idiotic, but so many people don't understand that LLMs don't actually understand anything.

• ⁨Comment⁩ on ⁨AI agents get office tasks wrong around 70% of the time, and a lot of them aren't AI at all⁩ ⁨⁨11⁩ ⁨months⁩ ago⁩:

  "Basically their methodology was that they asked ChatGPT whether the job could be automated," he explained. "They also asked people whether the job could be automated and then they said ChatGPT and people agreed some portion of the time."

  lol. This is such an idiotic thing to do. "Hey, you know that linguistic pattern matcher that doesn't actually reason or introspect? Since it can talk, why don't we just ask it what it can and can't do?" Seeing this, published by an AI research institute no less, is what inspired the creators of the actually riggourus test the article is about. It inspired me with a desire to smack those idiots upside their empty heads.