Sabitlenmiş Tweet
Clockwork Whale
2.4K posts
@St_Rev 🤜🤛
crazy how mentally exhausting it is to infer what someone said from context so you can pretend you heard them
English
@tszzl @teortaxesTex yeah ime it mostly manifests as pedantry, 5.5 likes to point out something kind of obvious that you already knew
I can tolerate it. pedantry is a common character flaw in human STEM people too so anyone who's spent time around them has practice tolerating it heh
English
@clockworkwhale @teortaxesTex this has clear downsides too. 5.5’s theory of the user’s mind is quite bad
English
After a long time with V4, talked to Opus 4.7 again and… the magic is gone, huh. It doesn't feel more sharp, more insightful, *bigger*. Just a well-trained code monkey.
5.5, however, is on another level.
English
@xwanyex did a bunch last year with my neighbour and didn't like the concrete mixing/pouring part
digging was very satisfying to the toddler area of my brain that feels a man digging holes is fundamentally what 'work' is, but the concrete ruined it by introducing too much technology
English
@0x49fa98 coffee shop macbook is performative, real work is being done in that situation maybe 10% of the time
English
@tszzl kinda boring result because they're just choosing the religious tradition that has the highest social status among western people who write the way they've been trained to write
this is what any smart person told that these are prediction engines would guess they would say
English
@nichessgame @robertlasagna1 life was created by God using magic spells
English
They shouldn't make up slurs for robots or ais, "clanker" is retarded, we should make slurs for people who USE ai
English
@moveincircles pretending to believe that akshully the voters are upset because we weren't radical enough is just what you do when you're left and there's an anti-left backlash
I think the message isn't for the public, they know that's futile, it's done to bolster morale for their own people
English
Keith keeps saying voters are punishing Labour for not delivering enough change fast enough
Has he considered that maybe voters are fed up with the kind of change Labour and the Tories like to deliver, and want less of it, not more
English
@woke8yearold it's real bad that so many CEOs are lying about their accounting-driven layoffs being about AI, because low info leftists are completely believing it and being radicalized by it
English
CEOs keep blaming AI for layoffs because “our productivity is surging and we’re cutting costs” sounds better than “we suck shit” but no one could believe it for Coinbase
Dr. Parik Patel, BA, CFA, ACCA Esq.@ParikPatelCFA
Imagine firing 14% of your staff and making the stock price go down not up
English
@teortaxesTex some pivot from "heh, fake" to "DESTROY"
others pivot from "heh, fake" to "full legal personhood rights for LLMs" activism
I am far more afraid of the second group and I expect it to be dominated by women, which will make it more politically dangerous in the west
English
I wonder how a normie AI bubble/stochastic parrot believer would react if you showed him a SoTA agent working on a nontrivial task, reasoning, experimenting, *orienting towards victory*.
They will see soon enough, though.
And I shudder to imagine the fallout.
English
@teortaxesTex 4.6 avoids this by being overfit and having much less variance in its answers. but tbh now that they don't allow you to control the temperature, everything I've described above could simply be sampling config issues
English
@teortaxesTex 4.7 is a weird weird uncanny feeling model, when writing it often chooses a totally incoherent/nonsensical token like a much smaller model, except because it's large it immediately realizes and corrects itself
English
Opus 4.7 has really bizarre behavior on several benchmarks
Jasper Dekoninck@j_dekoninck
The latest versions of ArXivMath and BrokenArxiv have been released! Nothing too surprising: GPT-5.5 remains at the top, and other models follow very similar patterns as the March version.
English
@teortaxesTex security people are hugely overestimating how scared normies are by the idea of models discovering new buffer overflow exploits or whatever
it's delusional to think politicians are going to take drastic measures over it, they were better off sticking with the biotech stuff
English
> Most likely, I expect that DeepSeek and other Chinese companies will move toward lobotomized open-source models
So far, OSS lobotomization has been the Western playbook, though mostly to preserve commercial edge.
let's see! But why is "Mythos" the point they should start at?
Dave Banerjee@DaveRBanerjee
In a post-Mythos world, the public will no longer have access to the cyber frontier. Some may argue "but China is only 8 months behind so they will open source mythos level models then." I disagree. It seems unlikely that the CCP will let one of their companies release a model that gives EVERYONE (including China's adversaries) access to a model that differentially increases EVERYBODY's offensive cyber capabilities. Most likely, I expect that DeepSeek and other Chinese companies will move toward lobotomized open-source models (models that have been specifically trained to have weak cyber capabilities). They would ofc have internally deployed cyber-capable models and I assume they will follow suit with Anthropic and give access to the Chinese gov and critical software maintainers (There are other possible scenarios but this is my modal guess)
English
@kendrictonn straight but it resonates, we're the same age so wondering how much is physiology
tendency for certain behaviours to continue after decay of the underlying hormonal substrate has ceased to render them titillating, like wile e coyote running off a cliff and not noticing for a bit
English
So you get, or at least I get, this incredibly monotonous stereotyped-language description of nominal hedonism. "Two guys asked you to pound them, eh. Everything about that sentence you just texted me gives me the same feeling as my taxes."
English
Man, I don't know if I'm more weary of the male gays for our inability to talk about anything but sex, or the rancid-banal tone in which we go about said, but I'm about ready to hang it up and go live alone in the bee-loud glade
English
"I've gotten so fat"
"You're not fat honey, how many times do we have to go over this. And I'm not just saying this to be boyfriend nice, I'm very sexually attracted to you"
"Yes, and that means I'm fat"
English
@teortaxesTex based
respectable scientists would never have added the sysprompt rp cot trick
English
I've been discussing DeepSeek V4 paper with Opus recently and it said that many things are explained by DS having a more competitive programmer than ML scientist culture – few ablations, sloppy citation discipline, trick stacking, "how" over "why" narrative
Made me think
Cosmin Negruseri@cosminnegruseri
to be fair here's the original paper also we were using the sqrt(n) trick in romania in coding contests in the 2000s to reduce memory in dynamic programming algos (backprop is a dynamic programming algo)
English
@clockworkwhale @hei_tetao virtually everyone who maintains this shithole of a planet will press blue
English
I really don't like this meme but I also don't understand the theory of victory for red buttoners. Do they think living in a world full of psychopath wannabe Chads will be nice? They won't get to mooch off anyone there. BOTH prosocial and antisocial rational response is blue.
MrBeast@MrBeast
Everyone on earth takes a private vote by pressing a red or blue button. If more than 50% of people press the blue button, everyone survives. If less than 50% of people press the blue button, only people who pressed the red button survive. Which button would you press? BE HONEST.
English
@hei_tetao @teortaxesTex the idea is that a nonzero number of people are going to press blue without full understanding of the consequences, so they are not making an informed choice
children, mentally incompetent, people who trip and fall on the blue button while trying to press red, etc
English
@teortaxesTex Well, if you accept that red is guaranteed survival, you also accept that blue is gambling your life when you along everyone else have a perfect choice not to, correct ?
If that's the case, then the danger is caused by people taking blue, not red
English