So am I getting this right?

[tante@tldr.nettime.org]

So am I getting this right? The guy who vibecoded an insecure, expensive, worse version of IFTTT with delusions of grandeur ("OpenClaw") is now claiming that he'll solve the problem of "prompt injections" at OpenAI?

My man, you are you out of your depth, it's not even funny.

[endolexi@social.vivaldi.net]

@tante I suppose vibecoders need to believe they can vibecode themselves out of every trouble they've vibecoded themselves into

[koantig@mamot.fr]

@tante My bet is that he's going to add "Don't fall for prompt injections" in the prompt. That's it, that's the strategy.

(that'll go right after: "Don't hallucinate")

[tante@tldr.nettime.org]

@endolexi it really is delusional

[endolexi@social.vivaldi.net]

@koantig @tante yes, "communication is everything, and everything is communication, and that means you just have to say what you want and it will come true", say countless bUsiNess aNalySts who've never coded anything in their life, and are overjoyed to learn that now they won't ever have to talk to developers again.

[androcat@toot.cat]

@tante Isn't this what OpenAI is all about? Clueless jackasses who talk big and can't think?

[anderslund@expressional.social]

@tante somehow that sounds like a brilliant idea. Maybe he could fix some more ai stuff too.

[holgertdittmann@ruhr.social]

@tante It‘s like a proof of concept for a homebrew nuclear reactor. It works, should not be used, but he pinkyswears moderators and containment will come in later versions.

Meanwhile, use at your own risk.

The short interview on ct3003 gave me terrible „mad scientist“ vibes.

[semaphor@sunny.garden]

Without words.

@tante @endolexi

[semaphor@sunny.garden]

While riding on cannonballs captures the vibe more on the spot

@tante @endolexi

[themipper@mastodon.social]

@tante All of this is so completely bonkers to me.

Prompt injection is not the problem they say it is.

The LLM works exactly like it is supposed to be. It takes a text and outputs another text based on some weights.

Prompt injection just highlights the reality they don't want people to see:

LLMs cannot think!
They cannot really do what is advertised.

Any output that works "in the context of your input" is pure luck.

[joriki@infosec.exchange]

@tante

I read it as dedication to the bit: pretending that genAI is a normal technology with plausible cost/benefit following well-known patterns of innovation by smaller players and acquisition by larger players

the salary is probably a marketing expense

and right now some number of rubes are ramping up their u$age thinking they're going to be acqui-hired too