Post ArpEHfIf4IwlA9fmd6 by feliks@chaos.social
(DIR) More posts by feliks@chaos.social
(DIR) Post #AroYPcMhfSXGIofnH6 by feliks@chaos.social
2025-03-07T14:45:16Z
0 likes, 0 repeats
do you believe the fediverse to be a good source of information regarding #machinelearning?
(DIR) Post #ArofX5H5HUPTfZYI40 by jaymoore@genomic.social
2025-03-07T16:04:57Z
0 likes, 0 repeats
@feliks more trustworthy than most other sources I would say
(DIR) Post #Arok0g6BqdAe5np74K by feliks@chaos.social
2025-03-07T16:55:13Z
0 likes, 0 repeats
@jaymoore what do you mean by "other sources"? and how do you establish trust?
(DIR) Post #ArolmTkz8dP3fWGJCi by lizzy@social.vlhl.dev
2025-03-07T16:59:57.199076Z
0 likes, 0 repeats
@feliks Technology and its societal impact and consequences are two very different pairs of shoes. If I complain about advertising online I don't really care about the video or image encoding algorithms used to deliver and display the ads - I care about the impact.I think fedi is a good source of information regarding the societal impact of the corporate LLM bubble. To learn about the technology itself I think wikipedia would be the right place to go. or someone's DMs ;)
(DIR) Post #ArolmUUiObyzxLepU0 by feliks@chaos.social
2025-03-07T17:14:59Z
0 likes, 0 repeats
@lizzy i second your opinion that fedi is spot on in criticizing capitalism and everything it connects towhat i meant though is not it's societal impact though but more the technical side of it. this is something that i see lacking here. mostly because the machine learning bubble seems unwilling to move away from twitter (lecun, chollet, etc). while i see that this would lead to disrupting those peoples experiences on the platforms i regret they're not willing to put in the extra effort
(DIR) Post #AromFbfmmkL0cNieps by feliks@chaos.social
2025-03-07T17:20:20Z
0 likes, 0 repeats
@lizzy then again i see this also rooted in the chaos community. before publicly discussed successes of machine learning people in the spotlight here were mischaracterizing its achievements and helped create the toxic breeding ground of "AI bad"almost nobody talks about the technical achievements on how e.g. #mumble is technically superior to most audio communication software only through its use of machine learning, namely #RNNoise. honestly this makes me sad
(DIR) Post #ArondV2FDRKAw4BHdY by jaymoore@genomic.social
2025-03-07T17:35:47Z
0 likes, 0 repeats
@feliks I guess I'm comparing with other social media principally. I quantify trustworthiness by gut feeling, the perceived motivation of the poster, and the expertise I judge them to have based on their and my experience. I guess I feel quite strongly there's a lower bullshit quotient and a higher expertise level here generally. It's just a gig feeling though.
(DIR) Post #Arp1Yw04voxoTbsx0K by lizzy@social.vlhl.dev
2025-03-07T17:25:32.702913Z
0 likes, 0 repeats
@feliks I don't think disagree with the "AI bad" bubble. It's a bit like "crypto bad" is a criticism of blockchain and shitcoin scams and not of legitimate applications of cryptography. AI is the term that all the bullshit is marketed under, so if you want to attack it in a way that people understand it, "AI" is the right combat term to use. I don't think that "there are legitimate uses of this technology / a similar technology" really helps the debate, I think people who aren't fully dense understand that, but it's a bit besides the point they're trying to make.
(DIR) Post #Arp1YwjoBnXklRHTHc by feliks@chaos.social
2025-03-07T20:07:41Z
0 likes, 0 repeats
@lizzy that might be true but at the same time it feels like getting thrown under the bus by this criticism because "AI" and "genAI" also includes some technology that really advances human civilization. LLMs are not only used to generate texts for everyday use but also in different fields of science
(DIR) Post #Arp1hQm8OVwaEyPkw4 by feliks@chaos.social
2025-03-07T20:13:26Z
0 likes, 0 repeats
@jaymoore i agree to some degree. i also see less competetive minds that are more concerned with the actual quality of the work and less the presentation of it. i see this to be a good thing. however quality-wise i have to admit that i've rarely if ever seen something relevant on fedi ever. it's lacking the hype train with all the tech bros but also all the experts they are latching on. while twitter has high diversity in quality it also still is home to most top minds in the field
(DIR) Post #Arp4ECDpivxaZL7bwe by jaymoore@genomic.social
2025-03-07T20:41:41Z
0 likes, 0 repeats
@feliks I left twitter a year ago so can't comment what it's like these days
(DIR) Post #Arp4N24Cqls5gApf1M by jaymoore@genomic.social
2025-03-07T20:43:20Z
0 likes, 0 repeats
@feliks I always found the best minds on kaggle tbh
(DIR) Post #Arp4qIUUrgkT115GqW by feliks@chaos.social
2025-03-07T20:48:37Z
0 likes, 0 repeats
@jaymoore true. didn't consider kaggle a social medium and therefore didn't include it
(DIR) Post #Arp4v2mkgY2Zklt07M by jaymoore@genomic.social
2025-03-07T20:49:29Z
0 likes, 0 repeats
@feliks I used to enjoy the chats there
(DIR) Post #Arp4yX5KULFCawEt3w by feliks@chaos.social
2025-03-07T20:50:08Z
0 likes, 0 repeats
@jaymoore why did you leave?
(DIR) Post #Arp5HRb57H1b4o6nGi by jaymoore@genomic.social
2025-03-07T20:53:31Z
0 likes, 0 repeats
@feliks I kind of moved on from the brute force competition to squeeze out an extra tiny margin by flexing on techniques and went more deep into the specifics of a new application area (genome editing) where it's less relevant because we don't have large enough training sets yet
(DIR) Post #ArpEHfIf4IwlA9fmd6 by feliks@chaos.social
2025-03-07T22:34:24Z
0 likes, 0 repeats
@jaymoore seems reasonable. also that competetive environments can be fruitful for innovation but coming with a set of risks on its own. interesting. so this is all in a small data regime? how do you then induce bias and on what levels? are there simulations or models?
(DIR) Post #ArpJflQKS4yasMZOPA by jaymoore@genomic.social
2025-03-07T23:34:47Z
0 likes, 0 repeats
@feliks we have some data from experiments and some models but not such large data sets for machine learning,
(DIR) Post #ArpTxVa35ujAUdDT5E by feliks@chaos.social
2025-03-08T01:30:05Z
0 likes, 0 repeats
@jaymoore what modalities?
(DIR) Post #ArvH98mzWZxapJFJK4 by jaymoore@genomic.social
2025-03-10T20:34:43Z
0 likes, 0 repeats
@feliks I can't talk about too many details but our tech uses gene editing to redirect microRNAs to silence target genes in pathogens. Data on microRNA expression and accumulation is not super abundant so we use what we can get hold of.