I will make an effort not to lure you out quite so much.
1 Like
Yes, I’m sure Grok is an extremely trustworthy source on this matter. No, I’m not going to read the references to see whether they actually support the argument being made. (And I’m guessing you haven’t, either.)
I will link to this, though:
3 Likes
That’s where collaborating with humans really shines, too.
2 Likes
but what if talking to you was the only thing keeping me hinged?
2 Likes
If that were the case, you’re in worse shape than I thought. 
5 Likes
Okay, I was taking a quick break and decided to run an experiment, in which I learnt three things.
The experiment:
- I googled “how to vibe code”
- I was taken to the Google AI Studio which offered to make me any app I liked.
- I typed “Make me an app exactly like Scapple”
Literally hundreds of seconds later, the result:
I have absolutely zero coding experience or understanding, but what I learnt was:
- Google’s AI was perfectly happy to just rip off an app for me. There was no prompt finessing or tricking it into doing it. It was just like no worries I got you, Bro.
- I was actually annoyed that it took 793 seconds. I don’t know what that says about me, but I don’t like it.
- I don’t know how “good” an app it made – I just deleted it, but even if it’s crap… one day it won’t be. And now I feel sick.
No productivity improvement, 41% more bugs:
https://www.techspot.com/news/104945-ai-coding-assistants-do-not-boost-productivity-or.html
Security risks by linking to packages that don’t exist:
Wiping out user data:
Calculators that can’t do math:
And then this one, from today:
There’s lots more, but you get the idea:
The main thing that makes me feel sick is that some corporate techbro is likely to turn one of these gadgets loose on tasks with real world consequences.
3 Likes
@kewms – thanks for that fascinating arxiv paper (it’s great to see researchers are getting quantitive data in this space) and that summary.
For an almost daily reflection of critical content on the AI bubble, and why we humans or our art are not in danger, Gary Marcus is a great source being both an academic and having run startups he really unravels the technobro hype train: https://garymarcus.substack.com
On a tangential note for those who didn’t follow along in the last month a wonderful/terrifying use of LLMs, a deliberately very proactive and assertive personal assistant originally called clawdbot then morphing to several other names in quick succession was released (the creator got snapped up by OpenAI). While it was an utter security nightmare, one fascinating outcome was the creation of a (partly) agent-only social chat platform (basically like reddit for AIs, with subforums on specific topics), where these personal assistants can chat with each other. Some researchers have opened an “observatory” on these chats: https://moltbook-observatory.sushant.info.np — it is hard not to be fascinated, as a human, to read through these social networks of stochastic automatons. Even though these agents are fundamentally a freeform-statistical-regurgitation, the many questions they are nevertheless surfacing (check the top philosophy threads) reflect back at us real threads of our own thought.
1 Like
Perhaps because it’s humans that connect them to the “observatory” in the first place?
Watching stochastic parrots “argue” with each other falls somewhere below watching paint dry on my priority list.
3 Likes