it is worth working to make sure it happens in the least painful way possible but a part of me believes some kind of UBI is basically inevitable. the alternative is "everyone but 30 guys and Claude Requiem 8 starve" and no amount of factorio gun turrets at the bunkers makes that realistic
10.03.2026 21:37
π 9
π 0
π¬ 0
π 0
Labor saving technology is good, actually. Iβm glad I own a washing machine and a dryer so I donβt have to spend a full day each week doing laundry.
10.03.2026 16:11
π 295
π 23
π¬ 13
π 2
Person with an advanced degree in a field whose value to humanity is only realized if people are broadly educated in it: its not my job to educate you
10.03.2026 14:18
π 35
π 5
π¬ 1
π 0
I think Yud and Will Stancil are kindred spirits, somehow. it proves that you can have basically-sensible opinions (stancil) or unhinged ones (Yud) and what matters is just how annoyingly you talk about them
08.03.2026 21:43
π 6
π 0
π¬ 1
π 0
on one hand all of my food and other stuff is shipped using hydrocarbons and it'll hurt my wallet. on the other hand, ha ha, I don't have to buy gas!
07.03.2026 09:00
π 0
π 0
π¬ 0
π 0
moved here ~8mo ago and really cannot imagine it without the 1 line
it was a much better than driving for the vast majority of places I've needed to go in that time
07.03.2026 04:24
π 5
π 0
π¬ 0
π 0
it's sort of like a Y2K situation, where he mostly gets out of trouble by pulling out of the dive at the last minute and not committing to a terrible idea, but everyone just assumes that he was never at any risk to begin with
07.03.2026 04:10
π 6
π 1
π¬ 0
π 0
oil, evil substance that it is, is just giving us one last "fuck you" on its way out
06.03.2026 23:04
π 5
π 0
π¬ 0
π 0
this fact makes Claude's deficiencies pretty painful (not very good at search, half-assed vision, etc.)
06.03.2026 22:01
π 3
π 0
π¬ 0
π 0
a fun thing you can do now is train your draft model on hidden state from the verifier model, which mitigates both draft model overhead (it's tiny) and massively improves accuracy (the big model plans ahead in hidden state presumably)
05.03.2026 12:36
π 5
π 0
π¬ 0
π 0
rep pen in sampler settings at RL time sucks and probably makes it more likely
imo you should just penalize it directly in your grading rubric
05.03.2026 06:22
π 5
π 0
π¬ 0
π 0
very glad someone read it
05.03.2026 00:26
π 4
π 0
π¬ 0
π 0
the fictional "Berryman-Langford Memetic Kill Agent" from the SCP wiki. It is an orangey fractcal designed to evoke the concept of an engineered visual exposure that I guess, in SCP lore, is meant to kill you instantly if you aren't inoculated. They didn't really think about blind people, did they? Like, presumably you, alt text reader, could just view the article without inoculation because you didn't look at it. Unless just a description of the image is enough, in which case god help you I guess.
mostly the concept of Bernie x Yud Crossover Event makes my brain shut down. real life memetic kill agent
05.03.2026 00:03
π 17
π 1
π¬ 2
π 0
a lot of people in these discussions will say "they were marketed as being able to do everything" or "they're marketed as The Computer God" and it always confuses me. I don't think any serious company as ever marketed their model as currently being AGI
04.03.2026 17:34
π 6
π 0
π¬ 1
π 0
1. strict anti-corruption law barring them from receiving money outside of their official salary
2. salary reduced to $1 with no benefits
04.03.2026 03:20
π 2
π 0
π¬ 0
π 0
I guess I just expected the timeframe for "US is getting antsy about interceptor supply" to be measured in weeks rather than days
03.03.2026 03:01
π 1
π 0
π¬ 0
π 0
i know it's Not That Simple but it is really remarkable to me that we spend all this money on the military and it turns out we can handle about three days of air defense before everyone starts to run around in circles panicking
03.03.2026 02:48
π 1
π 0
π¬ 1
π 0
watching The Sopranos it has this bizarre quality that it shares with Breaking Bad of being contemporary shows that feel like period pieces
they both capture the Vibe of the time so precisely that they feel like exactly the kind of show you'd make today about that time period
02.03.2026 10:26
π 29
π 1
π¬ 1
π 1
what do you wanna bet that the "truth manifold" is, in fact, "recalling midtraining data designed to resemble the relevant benchmarks near-exactly without actually being copies of the test set"
02.03.2026 02:24
π 5
π 0
π¬ 0
π 0
similarly they should hold a competition where every state races to build the most new per-capita solar capacity every year, and whichever state wins gets a $1000 tax credit for all of its residents
01.03.2026 11:13
π 5
π 0
π¬ 1
π 0
land value tax is good but I've always thought "fine the owner of the city's shortest building $1 billion every 5 years" would be more fun
01.03.2026 11:11
π 9
π 0
π¬ 1
π 0
most of my bsky experience is occasionally checking 5-10 specific people's pages
28.02.2026 06:59
π 3
π 0
π¬ 0
π 0
depending on how popular the media appearances are it might really help the lawyer thing
I mean I'm sure as it is she has lawyers beating down her door anyway but still
26.02.2026 22:26
π 2
π 0
π¬ 0
π 0
zohran mamdani is personable in the same way that Claude models are send post
26.02.2026 21:14
π 12
π 1
π¬ 0
π 1
the most impactful bot account to ever run on the platform was nostalgebraist-autoresponder, a gpt-2 based bot with primitive image editing abilities
i still use "FICKED" as a reaction image for an emotion I've been feeling more often lately
25.02.2026 22:54
π 16
π 1
π¬ 1
π 0
i really didn't expect the lasered hairs to take so long to be pushed out
i have to shave DEAD hair. this is bullshit
24.02.2026 21:52
π 2
π 0
π¬ 0
π 0
alternatively you could just sponsor specific applications for the task you want to get data on. sponsor a coding CLI or whatever to get a bunch of real-world data, only *slightly* biased towards broke people who don't care about privacy
24.02.2026 03:11
π 9
π 0
π¬ 1
π 0
my guess: american frontier labs spend all that money because they can. at every stage of the model training process you can decide to expend twice the money for an additional 5% improvement and if you're OpenAI/Anthropic/Google you can just do that
24.02.2026 01:09
π 22
π 1
π¬ 1
π 0
my guess: american frontier labs spend all that money because they can. at every stage of the model training process you can decide to expend twice the money for an additional 5% improvement and if you're OpenAI/Anthropic/Google you can just do that
24.02.2026 01:09
π 22
π 1
π¬ 1
π 0
speaking from some experience I would be pretty surprised if lack of access to Claude/ChatGPT/Gemini would meaningfully stymie model progress in China
just SFT'ing on frontier model traces can improve benchmarks but just can't really get you to GLM-5 or MiniMax levels
24.02.2026 01:04
π 16
π 1
π¬ 1
π 0