Imagine you scheduled a group brainstorming session to get some important work done, and one of the attendees casually mentions that they took LSD for the first time right before the meeting.

That's how you should treat ChatGPT output.

@mattblaze And here I thought this was going to be about Casey Means.
@mattblaze @CastlebridgeChief But chatCPT doesn’t disclose it’s tranced out of its gourd. You have to figure that out.

@mattblaze and that's why anyone who even mentions #AI gets kicked off premises by me.

@kkarhan Dude, maybe take it easy on the hashtags.
Johndgeek 🌎 (@Johndgeek@infosec.exchange)

Attached: 1 video Apparently someone added instructions for LLM scapers to send info about their SSH info and email details and it finally paid off. #llmhacking #gottem

Infosec Exchange
@mattblaze I’d trust the dosed coworker before an LLM
Kevin Karhan :verified: (@kkarhan@infosec.space)

@mattblaze@federate.social and that's why anyone who even mentions #AI gets kicked off premises by me. - Cuz people on #LSD are way more respectful, less nosy and quieter than all the #LLM bs that is just #WastefulComputing and that are worse than #Cleverbot and #DuckDuckGo #InstantAnswers combined)...

Infosec.Space
@mattblaze am I the only one thinking that that brainstorming session could actually be quite productive? And if not, it would at least be fun.
@Disputatore Maybe. But you definitely don't want to give the LSD guy final edit.
@mattblaze That's a good point. I think we agree.

@mattblaze @Disputatore Oh wow, this reminded me of something I hadn't thought of in quite a while...

In college a friend's chemistry lab group had a report due, and after each doing their respective parts, one of them was tasked with the final document preparation and hand-in. The other two found out some time later (after getting some concerned questions from the professor), that hand-in guy had apparently been quite high while doing so (on exactly what I don't know, possibly LSD?). At the time I saved copies of the actual files from my friend and managed to dig them up, and it's really quite something. (Some choice excerpts pictured here.)

So yeah, don't give LSD guy final edit...but sometimes LSD guy can take you by surprise.

@mattblaze huh, that... makes me more interested in playing with it, honestly

@mattblaze the fallacy is treating that in absolute terms. We should compare an LLM to the output of our colleagues.

Outperforming an LLM should be the baseline, yet…

@mattblaze Just one?

In Michael Pollan's book "How To Change Your Mind" I remember he described 3 instances of computer EE chip designers taking LSD to "focus on and extend the computers memory" and complete the designs. YMMV.

So maybe a different drug?

@MHowell no. LSD is a fine example here. Just like with LLMs, maybe they’ll do something useful. Or dangerously nonsensical. Either way, you need to check, carefully, to find out which.
@mattblaze @MHowell@kolektiva.social the same people who are going full force on LLMs everywhere are the same people who scarily took the wrong message from LSD
@mattblaze @MHowell I vote for a smart guy on acid over an LLM
@noplasticshower @mattblaze @MHowell Perhaps not *on* LSD but definitely who has taken LSD. Everyone should try it at least once.
@tvaughan @noplasticshower @mattblaze @MHowell well having to take a friend to ER who had that same attitude, no I think not.

@mattblaze Nah. I've taken (a small dose of) LSD before a work day. It was one of the most productive, creative, problem-solvey days I had in years. And the people around me enjoyed it too, because I was in a better mood. Meetings were good.

GenAI isn't like a person on acid, because a person on acid still has access to all of their existing knowledge and understanding.

GenAI is like a narcissistic sycophant that understands nothing, but wants to sound like it is very smart.

@mattblaze I still remember a very vivid hallucination I had in my teens. I had not taken any drugs of any kind. What I had done was get myself severely sleep deprived studying. However, I could apply years of experience that told me that the sun isn't straight overhead at 6am, dogs aren't naturally the colors I was seeing, and joggers can't phase shift through the trunks of trees. Unlike LLMs, I could perform a reality check and recognize that I was hallucinating. It isn't clear that LLMs can do that, or even judge that they might be operating "under the influence".
@mattblaze but... how do I ask ChatGPT if it has any spare?
@mattblaze I wish my local GPT would produce results nearly that interesting tbh

@mattblaze Am I the only person around here who recognizes the huge potential of LLMs?

I make productive use evert day of LLMs. DeepSeek is my favorite

I like the "enthusiastic intern" analogy. They are very keen, do mostly good work, but **must** be checked

They are a tool that extends and complements a person's skills and abilities. They are not a replacement

As Doctorow said:AI [sic] can't do your job, your boss got sold an AI to do your job, you're fired

@mattblaze

No. #AI is that annoying man on your team who always has an answer. And, even though his answers are often sus (or cringe), no one wants to be The One to #report him to #HR because they're pretty sure it would cost them their job, and they at least can help mitigate the fallout.

@mattblaze Honestly I’d rather deal with the tripping coworker and it isn’t close
@mattblaze
How many great songs of the 60 s used the same methodology :)
@mattblaze Yes, LLMs need to be treated with care. But at least they are cheaper and faster than humans :-).