Welcome to Friendica.Eskimo.Com

Home of Censorship Free Hosting

 Eskimo North Linux Shells, E-mail, Virtual Machines, Web Hosting.

E-mail, Web Hosting, Linux Shell Accounts terminal or full remote desktops.

Sign Up For A Free Trial Here

Please tell your friends about federated social media site that speaks several fediverse protocols thus serving as a hub uniting them, hubzilla.eskimo.com, also check out friendica.eskimo.com, federated macroblogging social media site, mastodon.eskimo.com a federated microblogging site, and yacy.eskimo.com an uncensored federated search engine. All Free!

mastodon - Link to source

LLMs are like slot machines, in that an incorrect answer (the slot machine eating your dollar) is unremarkable, while the LLM solving a problem (a jackpot) is amazing, and the latter stands out in your memory, causing you to overestimate the reliability of LLMs.

blog.glyph.im/2025/08/futzing-…

If you put ten minutes into writing a prompt, and Mallory gives a completely off-the-rails, useless answer, and you lose ten minutes, well, that’s just what using a computer is like sometimes. Mallory malfunctioned, or hallucinated, but it does that sometimes, everybody knows that. You only wasted ten minutes. It’s fine. Not a big deal. Let’s try it a few more times. Just ten more minutes. It’ll probably work this time. If you put ten minutes into writing a prompt, and it completes a task that would have otherwise taken you 4 hours, that feels amazing. Like the computer is magic! An absolute endorphin rush. Very memorable. When it happens, it feels like P=1.

Deciphering Glyph :: The Futzing Fraction

Deciphering Glyph, the blog of Glyph Lefkowitz.
blog.glyph.im
in reply to Cory Doctorow

mastodon - Link to source

Reg Braithwaite 🍓

I have argued that LLMs are like slot machines in another way: The business model. When you are paying by “the pull of the handle,” the vendor's incentive is not to solve your problem with a single pull, but to give the appearance of progress towards solving your problem.

Exactly the same as games companies that engineer their games to keep users buying coins, staring at the screen for eighteen hours a day. Very different model than selling tools or renting services.

This entry was edited (4 days ago)

Cory Doctorow reshared this.

in reply to Cory Doctorow

mastodon - Link to source

ClaraBlackInk

When the image generators came out, it reminded me of skilling up in video games. The way you can become an "expert" cook despite still being unable to nuke a pizza pocket irl without having different temperature zones. It felt like that and I could see how people would begin to see themselves as "artists" because it would feel like an accomplishment in a gamified way.

Or, well: twit.social/@JustinH/115010462…


Justin ⏚ (@JustinH@twit.social)

Attached: 1 video https://imginn.com/p/DNI4VI2oV0q/ The future tech bros want #ai #aislop
TWiT.social
This entry was edited (3 days ago)
in reply to Cory Doctorow

mastodon - Link to source

Azuaron

Recently at work, a manager pushing AI was presenting the "great stats" of AI use, and presented a "24% prompt acceptance rate" as if that didn't mean "76% prompt failure rate", aka, it fucks up the vast majority of the time. Even in these cases, of the 24% of prompts that were "accepted", there were no stats for how much additional work was needed to get the "accepted" result into an actually acceptable state.
in reply to Cory Doctorow

mastodon - Link to source

wakko

Better analogy - LLMs are like teenagers. They might know a couple facts through rote memorization, but they frequently make spurious extrapolations from their limited knowledge base and need frequent course-corrections from an experienced adult.

When the adults aren't steering the teenagers effectively, the teens tend to think maladaptive behavior is okay... just like LLMs.

Garbage in, garbage out is the same in both humans and machines.

in reply to Cory Doctorow

mastodon - Link to source

Orion Ussner kidder

Related:

I was (ugh) using ChatGPT to see how much my students might be able to cheat on a set of quizzes I'm going to give them, and I ended up in so many words telling ChatGPT to "admit" that it's just spicy autocomplete. I had this feeling of accomplishment that I "got it" to say it.

And then I remembered: it *merely* tells me what I want to hear, just like someone who wants a cybergirlfriend, just like someone asking for stock tips, just like someone who wants it to be a god.

⇧