Robert KingettYou all should read this fantastic post by @onepict about how, largely, AI people just can't leave us the fuck alone. But I'm noticing more and more and more that they get super mad when they wanna shove their #AI creation at you and you simply say, no thanks. That's it. No extra bashing, just, no thanks, and they get super offended. https://dotart.blog/cobbles/ai-and-that-guy-at-the-bar #AIHype #LLM
Bogdan Buduroiu@WeirdWriter @onepict there's literally been a witch-hunt by "never AI" types on Mastodon because Cory Doctorow dared to use ollama to spell check his posts. Idk who is the group getting offended.
@budududuroiu @onepict Using an LLM to spell check is all sorts of cringe because there are better applications of machine learning but moreover how do you know they never use an LLM?
@WeirdWriter @onepict you're literally proving my point.
Someone: I find small LMs useful for spell checking
Anti-AI types: Nooo, how could you? You're so cringe, what's wrong with you.
MarjorieR@budududuroiu @WeirdWriter @onepict we had and still have spellcheckers, which might be labeled AI (machine learning), long before someone had the idea of using LLMs to spell check.
Using a LLM to spellcheck is amazing overuse of resources to do something badly. A spell checker should be trained on your own practice and good quality sources such as dictionaries not on what MAGA idiots write on Facebook, X or Truth Social.
LLMs are general purpose stocastic parrots and have very limited context, they are not trained on how you write, unless supplemented they will not remember how you write.
Using a LLM to spellcheck is amazing overuse of resources to do something badly. A spell checker should be trained on your own practice and good quality sources such as dictionaries not on what MAGA idiots write on Facebook, X or Truth Social.
LLMs are general purpose stocastic parrots and have very limited context, they are not trained on how you write, unless supplemented they will not remember how you write.
Tiota Sram@marjolica @budududuroiu @WeirdWriter @onepict also Doctorow didn't just use an LLM for spell-checking, he pivoted to defending LLM use in general and disparaging those against it because he rightly predicted we'd be unhappy about his use of LLMs for spell-checking.
Also: we're unhappy because the technology being pushed on us is harmful in numerous ways. The AI boosters are unhappy because someone won't use the thing they're trying to sell (and/or because they keep getting called out on the harms). These are very different reasons to be unhappy with different levels of validity.
@tiotasram @marjolica @budududuroiu @onepict These pro LLM types always miss the larger point, don’t they?
@WeirdWriter @tiotasram @marjolica @onepict What's the bigger point? That you take a moral stance against LLMs but use transformers on the regular. So what are you really against? Instruction fine tuning? Reinforcement learning of pre-trained transformers?
@budududuroiu @tiotasram @marjolica @onepict You are that Tech Bro at the bar. https://dotart.blog/cobbles/ai-and-that-guy-at-the-bar
@WeirdWriter @tiotasram @marjolica @onepict Idk man, you seem like more of the obsessed person that has to convince me of your viewpoint, calling me technically illiterate while not knowing most spell checkers today use transformers
David Phd@budududuroiu @WeirdWriter @tiotasram @onepict there are some spellcheckers/autocomplete these days that are absolutely demented, such as the one in Google's Messages. Maybe that explains why they are so poor.