Post Aql4ISiuHapbDEftUe by CleyFaye@mastodon.top
(DIR) More posts by CleyFaye@mastodon.top
(DIR) Post #AqhzpvaCvCdH1wrdS4 by baldur@toot.cafe
2025-02-02T11:03:24Z
1 likes, 3 repeats
I’d like to reiterate what I said a while back: integrating LLM-based tools into all corporate and personal workflows is outright dangerous. Even when run locally, most LLMs in use are trained and tuned by corporations that are now deeply in bed with a lawless authoritarian takeover of the USPeople who are removing all references to minorities, women, and equality from your public spheres will not hesitate to ask corporations to tune centrally-controlled LLMs to censor the same from your work
(DIR) Post #AqhzpwkAbeCCd8X6cS by baldur@toot.cafe
2025-02-02T11:23:27Z
2 likes, 1 repeats
“But I’d notice if the LLM started censoring my work!”Really? Did you notice this? https://github.com/orgs/community/discussions/72603The point of cognitive automation is NOT to enhance thinking. The point of it is to avoid thinking in the first place. That’s the job it does. You won’t notice when the censorship kicks in
(DIR) Post #AqhzpxPI8l5agflwiO by amerika@annihilation.social
2025-02-02T12:57:26.766602Z
0 likes, 1 repeats
@baldur The censorship has been here for decades already.AI will just make it easier and oddly, more accurate.
(DIR) Post #Aql4H2gNcGWQ9uodKy by nyx@im-in.space
2025-02-02T13:00:22Z
0 likes, 0 repeats
@baldur I always wonder how it is people forget that this "lawless authoritarian takeover" was actually a democratic election. More people think/thought this is all good than the other way around. They are probably more silent than those who are unhappy.
(DIR) Post #Aql4H3NH2mpiIwstCC by baldur@toot.cafe
2025-02-02T13:26:51Z
1 likes, 0 repeats
@nyx Considering how aggressive Republicans are about preventing entire subsections of US society from voting, coupled with endemic gerrymandering, I’m personally not convinced that “democratic” is a word you can genuinely apply to US elections. I don’t think they’ve been fully democratic since at least 2000Also, even if it had been a democratic election, Musk’s takeover of US institutions, at lest, is both unlawful and authoritarian
(DIR) Post #Aql4H3oZPIfRfbegk4 by 7073M@noauthority.social
2025-02-02T17:10:22Z
0 likes, 0 repeats
@baldur @nyx you say all this, but are you even doing your part for refugees?
(DIR) Post #Aql4H4TKxjHFi2jFHk by baldur@toot.cafe
2025-02-02T17:52:34Z
1 likes, 0 repeats
@7073M Might as well put a screenshot of your reply on the wikipedia page for Whataboutism. https://en.wikipedia.org/wiki/Whataboutism
(DIR) Post #Aql4I9Qq0Y5XNh4hBw by baldur@toot.cafe
2025-02-02T15:07:43Z
1 likes, 0 repeats
“This isn’t going to happen. These are profit-motivated companies who are trying to sell LLMs as a productivity miracle. They won’t compromise the LLMs because that would make them less productive.”Really?> I have trans_time all over my code and CoPilot refuses to talk about it.From https://github.com/orgs/community/discussions/72603#discussioncomment-10388112
(DIR) Post #Aql4I9uyCWBut9Al9s by gpshewan@mastodon.social
2025-02-02T13:29:18Z
1 likes, 0 repeats
@baldur I had that discussion late last year, just got a lot of tech bros saying correct prompting was the answer - so it obviously wasn’t an issue. It was my problem. They could not grasp you needed to have the awareness, in the first place, to recognise it
(DIR) Post #Aql4IAsAePXvqks2Qy by baldur@toot.cafe
2025-02-02T17:57:03Z
1 likes, 0 repeats
Going to mute this conversations because the replies have gotten quite unhelpful, but I'll close with a note that open weight models don't solve a single fucking thing. Regular users, both personal and corporate, won't be hand-tuning their own models and running them on their laptops. Actual adoption means buying these as services or products and those are de facto centrally controlled.
(DIR) Post #Aql4IBzIVOqDJ9DFBI by heiglandreas@phpc.social
2025-02-02T13:04:01Z
1 likes, 1 repeats
"The point of cognitive automation is NOT to enhance thinking."This might be the most important sentence I have read in the whole AI/LLM/ automation conundrum so far.And brings my gut feelings that I was never able to really put into words to the point so well.Let's wake up and take thinking back into our own hands... or heads...@baldur
(DIR) Post #Aql4ID2Wat16ZRjKqm by ahltorp@mastodon.nu
2025-02-02T18:58:59Z
1 likes, 0 repeats
@gpshewan Yes, and if “correct prompting” is a prerequisite for using LLMs, then why are people using them for anything? Then it’s just a shitty, undocumented programming language that gives unreliable answers.
(DIR) Post #Aql4IKxr882H9HbigS by emilis@fosstodon.org
2025-02-02T17:12:48Z
1 likes, 1 repeats
@baldur I'll start adding gender related keywords to my code, also Tienanmen square.I think this was prophetic: https://www.smbc-comics.com/comic/human-arts
(DIR) Post #Aql4IL2SqzipNZlOro by babble_endanger@freeradical.zone
2025-02-02T22:18:35Z
1 likes, 0 repeats
@baldur boosting with CW, Github Copilot won't talk about `trans_time` variableI wonder if this means naming all variables after LGBTQ+ concepts will hide it from training...
(DIR) Post #Aql4IODv0rfPFy95dY by max@mastodon.xolus.net
2025-02-02T17:36:56Z
1 likes, 0 repeats
@baldur In four days my students have a software engineering practical exam, i'm baking in references to gender and sex everywhere I can in the code of the exercise.Hopefully that will both reward people who did try to understand what they are required to do, and raise awareness about the dangers of outsourcing their thinking.
(DIR) Post #Aql4IP9hY1t69BBEhc by PythonLinks@mastodon.social
2025-02-02T19:40:57Z
1 likes, 0 repeats
@baldur Thank you. I was seriously considering using LLMs for curation of articles, but both the Chinese and US LLMs will not treat democratic links well.
(DIR) Post #Aql4ISiuHapbDEftUe by CleyFaye@mastodon.top
2025-02-02T19:11:05Z
1 likes, 0 repeats
@baldur The number of angles from which tightly integrating LLM into anything is a bad idea is astounding. Privacy, bias, ecology, efficiency, how they're trained, the actual training data…The upside of "but it can sometimes do a bad to average job quickly" is starting to look less and less attractive.
(DIR) Post #AqpGYuD1pp9lAqD8T2 by fred@mastodon.sujets-libres.fr
2025-02-03T07:08:38Z
0 likes, 0 repeats
@baldur The original post was in octobre 2023. It seems that it's not immediately related with the recent US actuality.
(DIR) Post #AqpGYv72TZxXyYPrlo by clacke@libranet.de
2025-02-06T01:07:38Z
0 likes, 0 repeats
@fred Regardless which year it is from, it's an example of bias that most people probably didn't notice, and a direct answer to "I’d notice if the LLM started censoring".@baldur
(DIR) Post #Av9KuqgdIcup0b7l1k by aral@mastodon.ar.al
2025-06-15T08:45:55Z
0 likes, 0 repeats
@baldur And that’s not the worst thing. These ineffably slappable tech bros defending even something like this is. Those fuckers get away with shit like this because of these fuckers.
(DIR) Post #Av9KurFNDShkkLNVB2 by nanook@friendica.eskimo.com
2025-06-15T08:58:49Z
0 likes, 0 repeats
@aral @baldur I think when they introduce such limits, they eliminate the ability to get at basic truths.
(DIR) Post #AvG2sFyGOjy7UPhjw8 by n_dimension@infosec.exchange
2025-06-17T13:26:49Z
1 likes, 0 repeats
@baldur It's already happening.Musk famously fucks with #Grok sending it into tirades abou South African Whites extinction.And military is de-wokifying LLMsWhat's worse than a murderbot?A dewokified murderbot!