This post by @algernon is a surprising confirmation of one of my theories about why many of "us" aren't really impressed by #LLM's:

https://chronicles.mad-scientist.club/tales/conversations-with-an-artificial-intelligence/

Thing is, we've seen this before, played with it, found its limits and got bored. Of course LLM's provide much better results, but I still don't think the underlying principle is that much different.

Same with shitcoins: we designed a proof-of-work system as teenagers for password cracking, so the principle is not magical to us and this goes against the marketing.

@buherator @algernon indeed, that's a good point. We've also had such a bot in the past in one of our channels, and I even (much to the disliking of some others) trained that bot in a private query.

I must admit: it was super fun, though. That bot would come up with quotes like "greg is not terminated" or "phones are fluffy :D

@buherator @algernon and maybe we shouldn't have given the bot a human-sounding name. We were once asked if "that guy" was doing drugs or something. I just brushed it off with "naah, he's just not a German native speaker and is trying to learn the language".

@greg @algernon Thanks that's another data point then! I also remember having fun, and also being highly impressed by what one of the Markov-based tools could produce. In fact, I remember our reactions were really similar to those of early articles about ChatGPT (e.g. "it's arguing with me about X and making kind of good points").

@buherator @algernon is that a real post with content for humans? i only get the llm garbage maze in my browser...

ah, confirmed, it misidentifies my browser as an llm. :/

@stf @buherator It is entirely possible that some of my... more ambitious rules accidentally caught your user agent by mistake.

Can you tell me yours? I'll go and adjust the ruleset to make sure it passes.