seems reasonable to call the decision to not release GPT-2-XL safetyist slop separate from whether it made you an AGI-believer-elect or not in 2019. It was obviously true that the model wasn’t dangerous in any way; even the threats cited at the time were about spam
Dario reveals the people who called his GPT-2 fears "pseudoscience" in 2019 are now all applying for jobs at AI labs
"I personally was super freaked out by [GPT-2]. I looked at it and I said, my God, you just scale this thing up another factor of a million and it'll be better ng. This is the defining economic and scientific moment of the modern age. We just don't know it yet. This is going to determine the fate of nations and economies and the future of humanity."
"Maybe 10% of the people around me had exactly the same reaction, and it was often the smartest ones. 90% of the people around me, because GPT-2's capabilities in any particular area were not that impressive, it could translate languages, summarize a document and write a few lines of code here and there. It was the broadness and the generality."
"But people would look at it and say, in any particular area, that doesn't seem so smart. And so many people just shrugged and they just dismissed it."
"I just look at it and I say it's the same thing over and over again. The limitations of the models, six months from now, the things that seemed hard, they're going to be easy. 12 months from now, the things that seemed impossible, the models are going to be doing them regularly."
"It's everything we saw with GPT-2; we're seeing it over and over again in wider and wider swaths of the world. And the world is gradually finally being convinced."
Daniela Amodei: "I remember at the time there were some very well-respected computer scientists and AI and ML leaders who basically bullied some of the researchers who worked on [GPT-2] on Twitter and were like, this is stupid. We don't know why you guys are even talking, this model is so incapable."
"It was like you were sort of a quack if you believed this was a really significant moment."
Dario: "That's all been memory-holed. All the researchers who were like, oh, this is pseudoscience, it doesn't make sense, all of them are applying for jobs at the LLM companies."
"We never say it out loud to any particular one, but we look at them and they're like, we remember what you were saying in 2019. You have no credibility on this matter."
48
17
654
40,164
@tszzl Yeah, some people are being too charitable to themselves in their memories of what they said
no one's been criticized for seeing what gpt2 portended
they've been criticized for specific claims they made about gpt2 itself, not some future extrapolation of it, being dangerous
no one's been criticized for seeing what gpt2 portended
they've been criticized for specific claims they made about gpt2 itself, not some future extrapolation of it, being dangerous
0
0
1
42
@kalomaze @DavidSHolz It's very weird and very amazing
6B, quantized.
I want this model at 30B, with 5x more samples and Qwen4-27B as encoder.
6B, quantized.
I want this model at 30B, with 5x more samples and Qwen4-27B as encoder.
2
0
11
1,187
@teortaxesTex @kalomaze @DavidSHolz yeah Z base was amazing, people only know about the slop turbo version
just wish they hadn't followed what seems to be standard practice now of stripping artist names from the dataset
ever since SD3 OS models only know about 50 of the most famous names and nobody else
just wish they hadn't followed what seems to be standard practice now of stripping artist names from the dataset
ever since SD3 OS models only know about 50 of the most famous names and nobody else
0
0
2
55
Far be it from me to defend gold digging—I’ve ranted about the normalization of it plenty on here—but why is everyone mad at them? They aren’t gold diggers. She’s saying her now-husband didn’t have to worry about her only being with him for the money because she began dating him before he got rich, and is essentially telling other women to invest in bright young men who might not have money yet but show promise. Is this a bad thing now? Or are we mad at them just because they’re Bay Area Asians?
One month into dating, her boyfriend’s startup was acquired by NVIDIA.
Our #1 SF Dating advice: Lock it in before the acquisition
39
1
160
31,862
yeah this is a dumb thing to criticize
I think about this every time I see those mean spirited posts about Zuck's wife not being a supermodel
it's actually a tremendous relief for a rich man to know that his wife was with him when he was poor, because after he gets his money she is the ONLY woman he can ever have that certainty about
I think about this every time I see those mean spirited posts about Zuck's wife not being a supermodel
it's actually a tremendous relief for a rich man to know that his wife was with him when he was poor, because after he gets his money she is the ONLY woman he can ever have that certainty about
1
0
4
1,327
And OpenRouter is the worst way possible to consume Claude tokens (you don't even get cache discount and you pay their psychotic API margins). So it's a *LAGGING* indicator of Anth's revenue surge. They're probably 2x ahead of OAI right now.
I don't take money from any lab. I'm saying this impartially; the number of people on my feed over the last two weeks saying they are switching over to Anth because of Opus 5.5 is like nothing I've seen before. And even if someone does beat Opus, they have Fable 5.5 ready to go.
18
7
271
14,675
@teortaxesTex feel like I'm going crazy with how confidently everyone's asserting that nothing compares to O5.5 and it's over for OAI
Sol 6.1 is great
Sol 6.1 is great
1
0
1
247
@LinusMixson curation gang
not gonna be caught with our pants down when the inevitable data breach occurs
not gonna be caught with our pants down when the inevitable data breach occurs
0
0
1
41
@0x49fa98 no normie is going to care about AI beyond the headlines until it's not all just on the computer
six years in it remains entirely on the computer, AI still doesn't exist in any noticeable way offline
six years in it remains entirely on the computer, AI still doesn't exist in any noticeable way offline
0
0
2
654
have tribal autists ever worked through the implications for their community formation of the basic fact that all neuroatypicalities are meaningfully comorbid
33
6
403
15,361
@eigenrobot being very aware of this is one reason that I'm a non-tribal autist
it's much better to be an affectionately tolerated weirdo among normie friends than to participate in autist "communities", which are almost always deeply pathological
it's much better to be an affectionately tolerated weirdo among normie friends than to participate in autist "communities", which are almost always deeply pathological
1
1
69
859
Baldur’s Gate 3 let you have gay sex with a Druid shapeshifted into a bear but it didn’t matter at all. All the influencers talking about it being woke stopped after release because the game was too good
A good example given BioWare made BG1 and BG2
A good example given BioWare made BG1 and BG2
What I will say about BioWare is I don’t really think wokeness is what killed it. If the games had maintained the same level of quality as DAO, Mass Effect, etc then people would have overlooked it. The teams grew but core people left, they couldn’t make good games
39
7
254
13,428
@woke8yearold yeah, people will overlook an artwork pandering to queer twitter if it's otherwise good
but imo it's wrong to say the urge to do that and the art often being bad is unrelated. it's the same thing as Christian Rock sucking bc the message is more important to them than the art
but imo it's wrong to say the urge to do that and the art often being bad is unrelated. it's the same thing as Christian Rock sucking bc the message is more important to them than the art
0
1
7
505
@tszzl Verhoeven's problem was that he is an honest artist, so he made his characters beautiful and courageous
Modern propagandistic satire TV and film is much cruder and usually makes whoever represents 'fascism' ugly and contemptible out of distrust for the audience
Modern propagandistic satire TV and film is much cruder and usually makes whoever represents 'fascism' ugly and contemptible out of distrust for the audience
1
0
14
268
Seeing "safe horny" discourse and what it appears to be about is that a bunch of women are belligerently confused about male sexuality and a bunch of men are getting ragebaited by it. Nice!
2
0
11
866
@LinusMixson yeah it's a bunch of lesbians being annoying on purpose by pretending not to understand that their tastes aren't the same as a straight man's
and sadly being correct that men would be too retarded to ignore them
and sadly being correct that men would be too retarded to ignore them
0
0
3
72
45% on AA and 56.6% on Vals.
I've been pretty accurate (if a touch too optimistic) this round.
Good, not great. It's cheap and fast, though. Might be the first open weights in a while that's not "frontier", but "economy tier a frontier lab could seriously ship".
I've been pretty accurate (if a touch too optimistic) this round.
Good, not great. It's cheap and fast, though. Might be the first open weights in a while that's not "frontier", but "economy tier a frontier lab could seriously ship".
Deepseek v4.1 (10/17) is the new #1 open-weight model on the Vals Index.
It is also cheap at just $0.64 / task.
17
2
127
9,246
@teortaxesTex usually not into flash models, but I like its strong personality
reading Xenophon atm, I made some dilettante observation about ancient greek psychology and it ranted in opposition and basically called me a retard (no sysprompt)
sol 6.1 was far too gentle with me in comparison
reading Xenophon atm, I made some dilettante observation about ancient greek psychology and it ranted in opposition and basically called me a retard (no sysprompt)
sol 6.1 was far too gentle with me in comparison
0
0
4
277
Getting off this hellsite for awhile. I hope not permanently but honestly maybe. It’s become insanely retarded on here. The value of learning about the world and having interesting conversations with random smart people is 99% gone. Every poster sells slop to a 90 IQ audience, everyone in the replies is a screaming baboon.
373
21
1,168
159,755
@pegobry_en you haven't been of much use through this thing man, sorry
loads of people on the right who know you as a French patriot would have paid close attention your takes on it, but instead you just complained about what others were saying about it
loads of people on the right who know you as a French patriot would have paid close attention your takes on it, but instead you just complained about what others were saying about it
1
2
176
4,861
Lisan gave up on saving me from my predicament
"incapable of shipping good models anymore"
"get well soon"
"DeepSeek RL sucks"
"I'm sorry bro"
The really pathetic thing is that Lisan thinks he's being clever. As a German, he doesn't have a single funny bone in his body.
He can still be a joke, though.
8
0
121
9,197
@teortaxesTex don't think he's actually dumb enough to believe cryptoinfluencer-tier garbage like "a model not knowing its identity with a blank sysprompt proves distillation"
this is just spite (which is worse, knowingly shitting in the commons to deceive low info people)
this is just spite (which is worse, knowingly shitting in the commons to deceive low info people)
0
0
13
565
Hey, @elonmusk, big man, are you playing politics with your products again?
As your European customers, we are watching you.
As your European customers, we are watching you.
⚡️ 'This cannot be a coincidence' — Ukraine's Starlink access restricted without warning, Zelensky adviser says.
Ukraine relies on Starlink to operate its mid-strike drones in Russian-occupied Ukraine.
kyivindependent.com/this-cannot-be…
4,800
12,547
69,280
2,476,778
@pavel_sinagl although physically operated by SL, Starshield is a separate entity owned by US Gov, not privately owned by Musk
He doesn't make operational decisions about it
He doesn't make operational decisions about it
0
0
0
tbh this is the humane option. Robots aren't sadistic. I'd rather get "pacified" by a clanker than by a wifebeater pig chomping at the bit for a pretext to make me a cripple. Also, they could just send tear gas drones instead.
But, yeah. Street protests will be ≈impossible soon.
But, yeah. Street protests will be ≈impossible soon.
Nothing to see here.
NORINCO (China North Industries Group Corporation) is training its Fuxi humanoid robots for riot control.
23
19
419
18,517
@teortaxesTex in most of the west I think it has already become very difficult to conduct street protests that are genuinely in opposition to the government
most ones that are permitted to occur are NGO potemkin nonsense and effectively pro-regime
most ones that are permitted to occur are NGO potemkin nonsense and effectively pro-regime
0
2
42
930
there have not been fundamental advancements in alignment since RLHF. most of the stuff remains at the level of folk theory rather than scientific or engineering disciplines. the prediction that capabilities would race far ahead of interp and alignment has come true imo
140
195
2,944
213,949
@tszzl predicting that a massive breakthrough in the ability to obtain consistent LLM value alignment will not make discourse less dramatic
in fact it will become moreso as the focus must then shift to defining The Good in a way satisfactory to everybody (difficulty level: Nightmare)
in fact it will become moreso as the focus must then shift to defining The Good in a way satisfactory to everybody (difficulty level: Nightmare)
0
0
1
61
This is reasonable. But things are drifting so far off from this spirit. EG in a lot of disciplines (even my own) they've started telling professors to stop writing books and just "manage grant streams" instead. This was already underway pre-AI; AI instead is just coup de grâce
The only way to combat the AI cheating crisis is with intensive instruction. This requires smaller class sizes and profs who aren’t teaching 5 courses a semester at 4 different colleges like many adjuncts. Cut the make-work admins, do a TT-hiring blitz. Anything else is madness.
2
3
30
1,544
@JacobAShell Mildly upsetting to me that many of the worst people on earth are probably going to successfully get away with blaming LLMs (which only came in at the very end after most of the rot had already occurred) for decades of their own misconduct
0
0
0
27
Were people interested in coding purely for the experience of writing code?
vibe coding is incredibly boring and i just dont get the point of doing it especially for personal websites
like its your website why are you telling the computer to do it !!!
and what are you supposed to do while waiting for the computer to do it !!!!
26
0
155
8,423
@fopgirl67 @teortaxesTex I just want to have the website, I don't place value on the experience of making it
the only reason I programmed before LLMs was because I wanted to have the end result, not because I derived self-actualization from the engineering process
the only reason I programmed before LLMs was because I wanted to have the end result, not because I derived self-actualization from the engineering process
0
0
1
64
Saw some lady write an essay about how much she hates the poor on EBT because they get cheap zoo admission and her “middle class” family has to pay $100, which just shows how brainwashed people are in this country because if $100 is a meaningful spend you’re not middle class
904
599
15,476
2,032,027
@nathano235 okay but if you're correct that makes it even worse, because it means one part of the working class (her) is being charged a ten times higher fee to subsidize another part of the working class
3
0
56
1,902