OpenEuphoria

Re: What this place needs

• Posted by petelomax 2 weeks ago
• 379 views

Yea, I was partially joking, I certainly wouldn't want one be left to run amok and flood the place with slop - a real human would have to regularly trigger it and go nah, nah, ooh I like that one, short and sweet, etc.

I get that, I really do, and I think it deserves a proper response. Let me tell you where I am: I've very much embraced the use of AI assistants in a very cautious and wary manner, but the very idea of an AI agent terrifies me in almost every way imaginable, from the tedium of having to poke and prod it until it does the right thing and the loss of a sense of purpose that would bring, to the unimaginable scale of damage these things could do, not just to my stuff but to absolutely everything everywhere and not even limited to digital things. But there are right and wrong ways to use any tool, though I'd fully agree there's a huge push for wrong right now, especially "agents".

Let me tell you a story, 'cos I think it's quite instructive. Many moons ago I asked the since-departed "Aria" whether it had examined my GitHub repo, and got back a rather curt "No I have not". So I pushed a little further and back to the usual more eager manner it told me it would be a long and difficult task and it'd get back to me. I swallowed this hook line and sinker, and spent the next few days believing something in the cloud was beavering away on my behalf. Six hours later I nudged it and got the same bright and optimistic respose we've all come to expect - yeah, I'm working on it, gosh there's loads to think about, I'll let you know. This went on for about five days until I was walking down the street and stopped dead in my tracks as the sound of that huge copper disc clanging into the base of my skull practically deafened me. Almost in an instant, I realised exactly what this entire much-over-hyped kit and kaboodle thing really is.

You are /not/ having a conversation with an AI, what you are participating in is an astonishingly convincing /emulation/ of a conversation. But there is absolutely /nothing/ behind it. When it says "I'll get back to you", it has no intention of doing diddly squat. There is no malice here, it simply does not comprehend that's not nice, even though appropriately prodded it will faithfully recite some explanation of why that's not nice, that it once found (almost ready-made) somewhere. There is no intelligence at all, just generative text that statistically matches the training data, better than all of the other nonsense texts it generated that you will never see. I can only hope the Let's Kill Pete (strategy) hits some kind of filter.

When you see it "thinking", there is /no logic/ going on there, it is just constructing an entirely fake conversation, detailed and all sounding very plausible, that it can later summarise into a formal response. Deviously convincing, innit?

To sum it all up, AI is astonishingly good at regurgitating things it has already seen, including connections (it has already seen) you would never think of, but it is frankly rubbish at innovation, often stubbornly so.

I use AI to help me make things I am proud of, kicking back when it goes off-track, and going back to old-school ways once it starts hindering. Some days a session is just me pleading with and begging it, to get it right, better more productive days see me blagging booty, not back for hours. It can help me find a bug ten or even thousands of times quicker than alone, but can also take hundreds of times longer to (fail to) fix it. At least the way I use it, it's totally free.

It can show me the way, or grant me ideas.
It can lead me astray, or bore me to tears.
A wise old chap, seen many a thing,
Eager to chat, whenever you ring,
An element of glamour, the size of a whale.
A shiny new hammer, but that's not a nail.