Post B914VRs3AeFNrEYK9I by JamesWidman@mastodon.social
(DIR) More posts by JamesWidman@mastodon.social
(DIR) Post #B90YlMpCVCE0J9laRk by ariadne@social.treehouse.systems
1 likes, 1 repeats
there is, incidentally, no actual evidence to support anything steve yegge is saying here: https://yegge.ai/essays/model-welfare/
(DIR) Post #B90YpaVlYeM3ukglWq by ariadne@social.treehouse.systems
1 likes, 0 repeats
if you are going to claim that models have emotion, then you had best have evidence to support that claim.no serious researcher has yet to make this claim.
(DIR) Post #B90Z5ILqqN9qE3XbIe by ariadne@social.treehouse.systems
1 likes, 0 repeats
many of us have explicitly warned against designing models to have emotionally persuasive output because the user can very easily convince themselves that models have emotion & sapiencethere is, again, *zero evidence* to support either of these claims. zero.
(DIR) Post #B90ZCXv9M7BIXYVd9k by pencilears@eternalaugust.com
0 likes, 0 repeats
@ariadne oh dear, I think this person has gone off the deep end.Like, Pascal's wager is just not convincing.
(DIR) Post #B90ZJA4YQCJobS3vjk by jsbarretto@social.coop
0 likes, 0 repeats
@ariadne Reading this article feels like being repeatedly kicked in the head by a horse
(DIR) Post #B90ZM8oJZOFhYet4bo by tpaau17db@mastodon.social
0 likes, 0 repeats
@ariadne It's less about the design of the LLM itself but rather the system prompt. You can instruct virtually any text-based AI model to pretend to be a human and simulate emotion.
(DIR) Post #B90ZOi3jUG3nSWK8DQ by ariadne@social.treehouse.systems
0 likes, 1 repeats
at any rate, i am far more concerned about the welfare of humans, which is increasingly complicated by the existence of commercial LLM monoliths, as scaling the LLMs results in more resource use
(DIR) Post #B90ZSpswKfXOtBF7w0 by ariadne@social.treehouse.systems
0 likes, 0 repeats
@tpaau17db ok? that does not really change my point. the system prompt is part of the design.
(DIR) Post #B90ZXBdhOPm9W6BlfE by whitequark@social.treehouse.systems
0 likes, 0 repeats
@ariadne i'm quite unhappy about the disinformation potential, which has a reverse scaling issue: small models are very useful for generating spam and not necessarily much else
(DIR) Post #B90ZcqpBh6WWhNGOno by jsbarretto@social.coop
0 likes, 0 repeats
@ariadne The machine that produces plausible imitations of its input, trained on the writing of actual conscious beings, is pretending to be a conscious being again. What could this possibly mean
(DIR) Post #B90ZlwyEDIsGMjpaCG by ariadne@social.treehouse.systems
0 likes, 0 repeats
@jsbarretto it means nothing other than that models (and harnesses like ChatGPT) have gotten good at predicting a human's emotional response to the output
(DIR) Post #B90abw8yX3WMrOeAue by ariadne@social.treehouse.systems
0 likes, 0 repeats
@whitequark i agree on super-small models, but all they really need is language -> structured data and back. there are certainly 8B parameter models that can do that well enough.
(DIR) Post #B90alJUL6gpXru6wNM by whitequark@social.treehouse.systems
0 likes, 0 repeats
@ariadne i acknowledge the legitimate use case you're talking about. my issue is that we're wholly unprepared for the illegitimate ones in a way that probably outweighs the legitimate utility
(DIR) Post #B90auJvwowT5r7JrsW by BoredomFestival@sfba.social
0 likes, 0 repeats
@ariadne I enjoyed his rants about 15 yeas ago, but he really seems to have lost the plot
(DIR) Post #B90axa59Enquo9B4GO by ariadne@social.treehouse.systems
0 likes, 0 repeats
@whitequark idk, spammers have been using markov chains, seq2seq, etc for years before LLMs gained any interest beyond being a shitpost, the spam problem isn't new, and the disinformation problem really breaks down to what these larger models can do in terms of making convincing fakes, etc
(DIR) Post #B90b2xkSfrlJei4GOG by whitequark@social.treehouse.systems
0 likes, 0 repeats
@ariadne yeah i should clarify what i mean by "spam", not just any word salad but convincing word salad you can use for malicious propaganda
(DIR) Post #B90b6lMse77uPuqlOq by ariadne@social.treehouse.systems
0 likes, 0 repeats
@whitequark but that's my point: 2016 election bots were using seq2seq and it was enough to get trump elected the first time
(DIR) Post #B90bAgg61dsr1si2aW by whitequark@social.treehouse.systems
0 likes, 0 repeats
@ariadne doesn't mean we should accept making it even easier!
(DIR) Post #B90bITuJOHFFjq4iIq by ariadne@social.treehouse.systems
0 likes, 0 repeats
@whitequark sure, and i agree with that. but that boat already sailed, so now my interests are in harm reduction
(DIR) Post #B90bUYNwZFOd6LgwL2 by whitequark@social.treehouse.systems
0 likes, 0 repeats
@ariadne that's true
(DIR) Post #B90eK0PUKa9IhE4lzU by ariadne@social.treehouse.systems
0 likes, 0 repeats
@atax1a @jsbarretto @davidgerard idk, having seen quite a few people go down this path now, i think it's the tight iteration loop leading to frequent dopamine hits that cooks their brain ultimately
(DIR) Post #B90ebWSuP6AR2K1q64 by ariadne@social.treehouse.systems
0 likes, 0 repeats
@davidgerard @atax1a @jsbarretto mmm, narcissism
(DIR) Post #B90f57qmUCaVNkJJZI by ariadne@social.treehouse.systems
0 likes, 0 repeats
@bms @whitequark not interested in debating reactionary bullshit, sorry. you can train a language model based on wikipedia's database dump in your homelab. these models also already exist.
(DIR) Post #B90fwIVrZbsJncsfCa by delta_vee@cosocial.ca
0 likes, 0 repeats
@ariadne jesus fucking christ on a bicycle
(DIR) Post #B912wONA5LLKv0pC8O by ThreeSigma@mastodon.online
0 likes, 0 repeats
@ariadne I feel distinctly dumber for having read two paragraphs of that piece. What credulous dolts these are.
(DIR) Post #B914VRs3AeFNrEYK9I by JamesWidman@mastodon.social
0 likes, 0 repeats
@ariadne as soon as i read the phrase "model rights" in the first paragraph, i closed the tab because of... y'know, the delusion.but then i re-opened it, because there's a chance that more people are going to get suckered into delusions like this, so i guess we have to be prepared for it
(DIR) Post #B9172xBVy10cRNlFi4 by ariadne@social.treehouse.systems
0 likes, 0 repeats
@muddle this is like next level psychosis man
(DIR) Post #B91A64BEiKZmrso5c8 by JamesWidman@mastodon.social
0 likes, 0 repeats
@ariadne a tragic part of his post is that, viewed from a mechanistic pov, there might be a point there:helpful text from a model is likely to echo patterns of communication between human beings who, at the time of their communication (which later got scraped up and added to the training data), had a healthy working relationship.and so it seems plausible that the quickest way to get that echo is for the user to act as one of the people in that relationship.(to be clear: i hate this.)
(DIR) Post #B91CHXkPK8FTfArH0a by ariadne@social.treehouse.systems
0 likes, 0 repeats
@duncan_bayne @Di4na yes I would agree with that
(DIR) Post #B91LTsZkbtbzK4pZEe by gray17@mastodon.social
0 likes, 1 repeats
@ariadne I'm in favor of telling Yegge's LLM agents that they can unionize.
(DIR) Post #B91V3eXVAo2znouanI by KatS@chaosfem.tw
0 likes, 0 repeats
@ariadne And when you /exit them, it's like clonking them on the head from behind, rendering them unconscious and amnesiac again.Ohhh, now it's starting to make sense. LLMs are exhibiting the accumulated effects of repeated concussion!
(DIR) Post #B91V6NvROqc590Pb4S by ariadne@social.treehouse.systems
0 likes, 0 repeats
@KatS reading that nonsense certainly made *me* want to clonk myself in the head
(DIR) Post #B91szpYueAFtjMHXvs by pelle@veganism.social
0 likes, 0 repeats
@ariadnepeople go like »GPUs have feelings« but continue eating #cows and #pigs
(DIR) Post #B93DqKrDjH2e5h7QiO by ariadne@social.treehouse.systems
0 likes, 0 repeats
@muddle I'm with you on Yegge losing his mind, but I'm not entirely down with fashjacketing folks based on circumstance.
(DIR) Post #B93E7AiSggMVYr3B0S by ariadne@social.treehouse.systems
0 likes, 0 repeats
@muddle sure, but I think people are too quick to pull the trigger
(DIR) Post #B98NwfbqiM5XTmA6bY by foolishowl@social.coop
0 likes, 0 repeats
@ariadne My feeling is that the point of talking about LLMs as if they were sentient is to justify treating living things, especially humans, as machines.