It's June 2026 and though I'm a big fan of analytical machine learning, I still see no use case for generative #LLM chatbots.

[mrundkvist@archaeo.social]

It's June 2026 and though I'm a big fan of analytical machine learning, I still see no use case for generative #LLM chatbots. This distinction should be front and centre in all discussions of the technology.

#aibubble #ai

[troed@masto.sangberg.se]

@mrundkvist They're great at software development and cybersecurity analysis.

Other areas I can't speak to.

[mrundkvist@archaeo.social]

@troed
For all my life, I have been happy to let software developers and cybersecurity analysts chose their own tools. I feel that I don't need to supervise that choice. I also believe that this choice is not important enough for anyone else to motivate an insane market bubble.

#aibubble

[troed@masto.sangberg.se]

@mrundkvist Of course there is an AI bubble. Local LLMs are good enough for companies and individuals to use instead of cloud AI which will need insane subscription fees to make back their investments.

I just oppose the sentiment that LLMs aren't _useful_ - because that they are. At least in all areas where there is a clear works/doesn't work condition that can be objectively evaluated.

[joeinwynnewood@mstdn.social]

@troed

There's a whole lot of senior software engineers here on Mastodon who will vigorously disagree on LLMs' software development capabilities.

I believe they refer to LLM code as slop.

I've seen articles on studies of LLM coded systems and they are not positive for anything larger than a single user scenario and their security sucks.

@mrundkvist