AI discussions

johnraff · 2026-06-14 02:27:04

^I wonder if other backends like Claude Whatever differ when used via duck.ai compared with Perplexity etc...
I also had a terrible experience with chatgpt on duck.ai the other day.

But anyway I agree it's foolhardy to rely on any LLM AI for info that's important.

johnraff · 2026-06-15 01:41:03

greenjeans wrote:

johnraff wrote:
anyone have opinions about duck.ai?
it's definitely ignorant and gives wrong answers, at least on the default which is some chatgpt-mini.

I wanted to find out what deterministic latency meant, so tried "explain deterministic latency" on duck.ai.

The alternative Claude Haiku 4.5 gave a much more useful answer than the default GPT-5 mini.

Martin · 2026-06-15 11:08:08

I don't use LLMs much, very little actually, but so far I have been impressed by Le Chat from Mistral. Keep in mind though that this is based on very limited experience.

/Martin

Colonel Panic · 2026-06-15 19:28:50

This comment on that thread got my attention;

"That said, we already have the necessary policies in place:

* d/copyright must be accurate
* all sources must be reproducible from their preferred form of modification

Both are not possible using generative AI.

- nik "

Jimbo_G · 2026-06-16 06:20:01

Martin wrote:

I don't use LLMs much, very little actually, but so far I have been impressed by Le Chat from Mistral. Keep in mind though that this is based on very limited experience.
/Martin

For my uses, which are mostly rewording things I've already written, producing summaries, and so on, I have been getting good results with Perplexity.

I don't generally use LLMs for finding information and wouldn't trust the results if I did.

WizardofCOR · 2026-07-18 04:40:22

Building my own this weekend in my home lab.
AI isn't going away, we can all be certain of that - so I'm gonna try and make my own "Jarvis"... lol

johnraff · 2026-07-21 02:37:25

Jimbo_G wrote:

I don't generally use LLMs for finding information and wouldn't trust the results if I did.

I think they can be used in a similar way to Wikipedia: get some background, then check elsewhere.
I like how Perplexity always gives a list of links to the sources it used (or did last time I used it).

greenjeans · 2026-07-21 14:36:53

johnraff wrote:

I think they can be used in a similar way to Wikipedia: get some background, then check elsewhere.

Yep. And that's the exact same way I did it before AI ever existed. Plenty of human slop out there on the internet, best to check many sources and then use your own noodle to parse that data and come to a conclusion.

Also good to remember that AI as a whole is still in it's infancy, i've seen plenty of software over the years that was an absolute trainwreck when it started, that over time eventually matured and worked well.

johnraff · 2026-07-22 02:45:41

greenjeans wrote:

Also good to remember that AI as a whole is still in it's infancy...

Yes this, very much so. How many years ago was it that nobody gave a thought to AI, beyond some researchers in a lab? Not so many.

Right now the hallucination problem, so the unreliability of results, is a huge brake on the ability of companies to make money from LLM AI. It would be much easier to charge big fees if the customers could rely on the answers they got.

Free open-source alternatives, especially from China, are coming up fast, and while AI will evolve - and improvements on the current models will come - those big tech companies aren't guaranteed any kind of return on all the money they've spent so far. From that point of view there's a good chance the bubble will burst. The CEOs don't understand this stuff at all and I think they were taken in by the promise of being able to make huge cuts in their labour costs by firing half their staff.

greenjeans · Today 13:49:39

Debian discussing now whether to ban AI or accept some of it conditionally: https://www.debian.org/vote/2026/vote_002

Been my experience, human nature being what it is, that the surest way to guarantee that people do something, is to forbid them from doing so.

Sector11 · Today 18:08:44

greenjeans wrote:

Debian discussing now whether to ban AI or accept some of it conditionally: https://www.debian.org/vote/2026/vote_002

Yeah, I read they I hope they don't allow it "built-in.". Some will still use it.

greenjeans wrote:

Been my experience, human nature being what it is, that the surest way to guarantee that people do something, is to forbid them from doing so.

NO! REALLY! Whjat a horroble thing to say.

Me as a kid: "Yes you can go out and play. Stay away from the creek in the back 40."

10 minutes later, after stripping to the buff: SPLASH!

#61 2026-06-14 02:27:04

Re: AI discussions

#62 2026-06-15 01:41:03

Re: AI discussions

#63 2026-06-15 11:08:08

Re: AI discussions

#64 2026-06-15 19:28:50

Re: AI discussions

#65 2026-06-16 06:20:01

Re: AI discussions

#66 2026-07-18 04:40:22

Re: AI discussions

#67 2026-07-21 02:37:25

Re: AI discussions

#68 2026-07-21 14:36:53

Re: AI discussions

#69 2026-07-22 02:45:41

Re: AI discussions

#70 Today 13:49:39

Re: AI discussions

#71 Today 18:08:44

Re: AI discussions

Board footer