Posts by TedUnderwood@sigmoid.social
(DIR) Post #AUibLxvG2x8wF4rZ56 by TedUnderwood@sigmoid.social
2023-04-16T18:50:55Z
0 likes, 0 repeats
@Jonathanglick I think people are imagining that current progress in AI will just extrapolate along a “technological marvel” axis, without thinking about issues like “who actually wants liability for that superintelligence,” etc. I think it’s more likely that we get GPT-4 level stuff & generative video on our smartphones. The disruptive consequences are less like Skynet and more like, um, hippies.
(DIR) Post #AUie46rUN7TscIHG3k by TedUnderwood@sigmoid.social
2023-04-16T19:21:59Z
0 likes, 0 repeats
@Jonathanglick @MattHodges @afamiglietti79 Yeah. I get this whole line of thinking partly from @ecourtem, who constantly reminds me that most ppl only wanted little computers in their pockets that they could use for selfies and cat memes.
(DIR) Post #AUieF2Cq1PotMYTMpM by TedUnderwood@sigmoid.social
2023-04-16T19:23:56Z
0 likes, 0 repeats
@afamiglietti79 @Jonathanglick @MattHodges Yep. I don’t know if all of this was consciously intended by me, but it all feels consonant & persuades me as I hear it.
(DIR) Post #AUip0fixuXVz4ur0C0 by TedUnderwood@sigmoid.social
2023-04-16T21:24:38Z
0 likes, 0 repeats
@jose_eduardo @afamiglietti79 @Jonathanglick @MattHodges I agree with the premise that cars reshaped society, and have no doubt that AI will as well. Lots of people don’t want that to happen — but then, lots of people in the 20c said they didn’t want the noise and pollution of the automobile. Individually, at least in the US, they did want the convenience of driving, so it happened.
(DIR) Post #AUugyh8nCniL9k033w by TedUnderwood@sigmoid.social
2023-04-22T14:50:58Z
0 likes, 0 repeats
I was skeptical about tenure as an asst prof, and even more skeptical after I got tenure and saw my cohort start to coast. But as I get to the stage of my career where I am really seriously annoying to Deans I have started to see the point of it.
(DIR) Post #AUukXWbYfag0TdkqBc by TedUnderwood@sigmoid.social
2023-04-22T15:31:05Z
0 likes, 1 repeats
Absolutely revelatory piece from Yoav Goldberg casting light on an overlooked puzzle about last year: why did we need *reinforcement* learning (RLHF) to unlock the potential of language models? Why wasn’t supervised learning enough? #LLM #AI https://gist.github.com/yoavg/6bff0fecd65950898eba1bb321cfbd81
(DIR) Post #AUulDm8PJw7InwGYUq by TedUnderwood@sigmoid.social
2023-04-22T15:38:42Z
0 likes, 0 repeats
@compthink I’ll give away a little of Yoav’s answer, because it’s kind of fascinating philosophically. What we needed to teach the models was, when to say “I don’t know.” And that’s not something we could teach by demonstration alone, because … we don’t know how much the model knows! Had to be dialogic.
(DIR) Post #AUuuyZaagHip5qq1oW by TedUnderwood@sigmoid.social
2023-04-22T17:28:01Z
0 likes, 0 repeats
@heuser you will pry default yellow from my cold dead not-actually yellow ✋ 🤚
(DIR) Post #AUwqOBuNEnZhisPL9c by TedUnderwood@sigmoid.social
2023-04-23T15:46:01Z
0 likes, 0 repeats
Essential article asking "Can #LLM transform #computational #socialscience": https://calebziems.com/assets/pdf/preprints/css_chatgpt.pdf (by @caleb_ziems with @Held, @omar, @diyiyang). Key excerpt below: zero-shot e.g. ChatGPT doesn't currently out-perform fine-tuned FLAN.
(DIR) Post #AUwxbO4e165utiS2Uq by TedUnderwood@sigmoid.social
2023-04-23T17:06:53Z
0 likes, 0 repeats
I found an article I really needed to know about through @adamghill 's fediview.com. It's a little clunky right now, like all the opt-in tools people are building to fix the lack of algorithmic discovery here. But somehow this needs to get built. https://sigmoid.social/@adamghill@indieweb.social/110242959974062768
(DIR) Post #AV1Y4FVNKQ81hZr7Qm by TedUnderwood@sigmoid.social
2023-04-25T22:14:18Z
0 likes, 0 repeats
I'm notoriously pro - #LLM, but I've been on two committees now where someone suggested "let's just ask ChatGPT to do our initial brainstorming for us," and I am getting intolerant of that gesture. I can see a model as a useful interlocutor for an individual's brainstorm, but "just paste ChatGPT output into our committee writing space" is LAZY.
(DIR) Post #AVOC9YuOstKQyOlegi by TedUnderwood@sigmoid.social
2023-05-06T20:27:05Z
0 likes, 0 repeats
Dispersal. #aiart
(DIR) Post #AVPdCUFHZPv5Ie78TI by TedUnderwood@sigmoid.social
2023-05-07T13:04:52Z
0 likes, 0 repeats
@mpe You may be coping by intellectualizing, but it’s great analysis and I’m bookmarking it mentally
(DIR) Post #AVTiVde4bhNyvMn3yq by TedUnderwood@sigmoid.social
2023-05-09T12:23:11Z
0 likes, 0 repeats
@brendel sure, absolutely — thanks!
(DIR) Post #AVe9eqCfvU22p1BedM by TedUnderwood@sigmoid.social
2023-05-14T13:14:31Z
0 likes, 0 repeats
My salt-of-the-earth grandparents used to warn me there was more to life than “book learning,” and now I want to repeat that advice to people who are worried #AI will take over. Sign manipulation is great. But there’s also a real world out there that doesn’t align w/ our sign systems.
(DIR) Post #AVeADbiPmgss1VbvnM by TedUnderwood@sigmoid.social
2023-05-14T13:20:48Z
0 likes, 0 repeats
Doing anything new requires time and experimental interaction with the real world. “Intelligence” is rarely the bottleneck.
(DIR) Post #AXDSpecz5y1KPaSDnk by TedUnderwood@sigmoid.social
2023-06-30T11:56:21Z
0 likes, 0 repeats
“A grad student who fell asleep in 1982 and woke up in 2022 might see large language models as a triumph for cultural theory.” My contribution to the debate this week in the CI blog. #LLM #Foucault https://critinq.wordpress.com/2023/06/29/the-empirical-triumph-of-theory/
(DIR) Post #AXG0Ewvxr8m1EKvAzQ by TedUnderwood@sigmoid.social
2023-07-01T17:20:07Z
0 likes, 0 repeats
Well, looks like I'm back on Mastodon.
(DIR) Post #AhrWiRxgbaJJVpHPu4 by TedUnderwood@sigmoid.social
2024-05-05T22:32:53Z
0 likes, 0 repeats
This is a test to see if a message posted in the Fediverse can automatically be forwarded to Bluesky via Bridgyfed: https://fed.brid.gy/
(DIR) Post #Ahxztxfn6fIEfTViCm by TedUnderwood@sigmoid.social
2024-05-06T01:12:50Z
0 likes, 0 repeats
If you’d like to make it possible for people to follow you from Bluesky — and you’re in a domain that permits it — all you have to do is follow @bsky.brid.gy.I think these connections have the potential to strengthen everyone. We need critical mass!