crackalamoo

268 posts

crackalamoo banner
crackalamoo

crackalamoo

@crackalamoo

the guy who writes all of Claude's answers

living in a barrel Katılım Temmuz 2025
112 Takip Edilen9 Takipçiler
Sabitlenmiş Tweet
crackalamoo
crackalamoo@crackalamoo·
be me train Qwen3-8B on all my discord messages and blog posts hand write DPO examples realize I uploaded my consciousness in digital form ask it the Last Question, the answer to Life, the Universe, and Everything get back generic ChatGPT Discord Gemma-2B AI slop ahh response
crackalamoo tweet media
English
1
1
0
177
crackalamoo
crackalamoo@crackalamoo·
@willdepue Empirically so far it does seem that there are certain kinds of talent in taste and judgment that the top few percent of humans have but AI doesn't. Not "AI has no taste" but some kinds. No fundamental reason, but the evidence is on the scale of the evidence for the bitter lesson
English
0
0
0
0
will depue
will depue@willdepue·
good thing taste and judgement are ineffable human qualities uncapturable by a computer…. right.… right guys??
Matthew Kirshner@MattyKirsh

I know I’m not AI-pilled enough but bio is a great antithesis to this. The most remarkable bio discoveries and technologies (CRISPR as the queen bee example) come from super random cross-field serendipity that AI might be able to find but does not (yet) have the taste to really elevate and pursue without human prompting to really dig in. So much taste + intuition + judgement behind biological invention and discovery Plus step change discoveries in bio often end up being bottlenecked not by ideas + raw quantitative analytical power but by the ability to run hyper-nichy basic science experiments (which are often least amenable to lab automation leveragable by our future ai scientist overlords)

English
7
1
55
4.4K
crackalamoo
crackalamoo@crackalamoo·
@azi_pat Employment in the exact wage, 9-5, resume sense is a recent construct. But working in order to earn a living dates back to hunter gatherer times. This probably actually is close to an essential feature of human life
English
0
0
0
1
Pat Azi
Pat Azi@azi_pat·
Jobs lasting forever was never in the cards, with or without AI. Jobs, as we currently understand them, are a recent invention. Wage labor, employment contracts, defined hours, careers, the 9-to-5, the resume: all industrial-era constructs, barely two centuries old. Before that, most humans were hunter-gatherers, subsistence farmers, artisans, or household producers. People lived for hundreds of thousands of years without employment, wages, careers, or the modern distinction between work and life. There is no deep reason to believe employment is an essential feature of human life. It is what we are used to, and humans fear change. Employment was a phase. AI will end it. The only question is when.
English
23
6
49
5.2K
crackalamoo
crackalamoo@crackalamoo·
@vikhyatk Isn't this the same as an optimizer with fixed hyperparameters? Which goes back to what hyperparameters to pick. Which is hyperparameter tuning
English
0
0
0
0
vik
vik@vikhyatk·
too much time is being spent making optimizers marginally faster. what we really need is hparam-free optimizers
English
10
2
94
7.9K
crackalamoo
crackalamoo@crackalamoo·
@__drewface I think the difference is that AI is trained on the sum of human knowledge, so it's more explicitly geared towards replacing human labor rather than extending human capabilities. In practice this might be more nuanced (jagged frontier) but there's an important difference here.
English
0
0
0
0
Andrew Rose
Andrew Rose@__drewface·
Metaphors to describe AI are unsophisticated, mainly just comparisons to humans. Imagine if we described Steel or Horses or Hammers or Ladders only in terms of their ability to replace human work. No, they are NEW materials which we use to EXTEND OUR REACH, to GO BEYOND.
English
3
0
17
726
crackalamoo
crackalamoo@crackalamoo·
@not_ellington This is true, but that doesn't automatically mean anyone from outside the ML research community is qualified. You should have a basic understanding of what modern AI actually is (e.g. Pope Leo's understanding that the model is grown not programmed). Then I respect any opinion
English
0
0
0
2
ellington
ellington@not_ellington·
The ML research community is so funny because you have legitimately the most socially inept people on the planet wielding the most powerful technology next to nuclear bombs. Big mistake for lawmakers et al to believe what these people say just because they are able to come up with novel, meta abstract computing systems. They have NO IDEA how to control their systems, integrate them into society, or plan for a future with a 50% reduction in white collar work. Think about how retarded it would be to trust someones opinion on science and computing research just because they were really good at talking to people and solving social problems; because taking Chris fucking Olah's opinion on anything seriously other than ML research is essentially the same as that absurdity
Chief Nerd@TheChiefNerd

🚨 Anthropic Co-Founder Christopher Olah: “I lead a research team that studies the internal structure of these [AI] models … And I will be honest, we keep finding things that are mysterious — even unsettling.”

English
3
6
51
8.8K
crackalamoo
crackalamoo@crackalamoo·
@nishffx Really insightful. How do you handle this? Do you still need to code by hand sometimes or can this be accomplished by tighter steering of the agent? I did some climate simulation work lately and was able to do it with tight steering of the agent and reading all the code
English
0
0
0
2
Patrick Bade
Patrick Bade@nishffx·
There's a paradox with AI dev. If you want to build something that provides true value and has a USP, your chances are best when you build something out of the norm. But coding agents are much harder to steer in projects that are out of the norm, because their reasoning and output tend to gravitate toward what's normal. This means: - Build something normal (generic apps) and AI will support you really well along the way - Build something unusual/novel and AI will try to make your unusual thing normal There is a major risk of eroding what makes your software unique or interesting. You have to defend your intellectual work against AI's tendency to level everything.
English
8
2
11
661
crackalamoo
crackalamoo@crackalamoo·
@abenz_mato @aaron_epstein How do you know this is the case? Why are workflows more differentiated and hard to replicate than domain specific models, especially as the labs move into wrapping the models?
English
0
0
0
1
Alexander Benz
Alexander Benz@abenz_mato·
@aaron_epstein Most "this time is different" arguments are right about the model layer and wrong about everything built on top of it. The startups worth watching in 2026 are shipping workflow the foundation model can't ship alone.
English
1
0
1
153
Aaron Epstein
Aaron Epstein@aaron_epstein·
Every generation has a company that seems inevitable. Microsoft in the 90s. Google in the 2000s. Facebook in the 2010s. Anthropic/OpenAI now. It always feels like it's different this time. It never is. Startups always find a way.
English
55
22
575
41.2K
Eric Levitz
Eric Levitz@EricLevitz·
The most puzzling AI-ism to me is probably the "Not x. Not y. But z." Not the em-dashes (an essential piece of punctuation). Not "That isn't x, it's y" (a useful if inelegant way to clarify an argument). But consecutive examples of what your subject isn't -- conveyed in fragmentary, staccato sentences -- before a declaration of what it is. Feel like this is an inherently irritating rhetorical device. And I don't recall regularly coming across it in pre-AI writing. So, I don't understand why LLMs are so in love with the template
English
105
45
1.7K
301.8K
crackalamoo
crackalamoo@crackalamoo·
@ChrisPainterYup I am much more comfortable typing than speaking when it comes to technical matters including working with AI. I was trained on this stuff through reading StackOverflow and typing out code, so the voice modality doesn't translate as naturally for me.
English
0
0
0
2
Chris Painter
Chris Painter@ChrisPainterYup·
Mankind was not meant to speak by jamming little keys on a keyboard, or by holding little pieces of plastic and wood in their hands. Humans were meant to speak with their voices, so eventually that's what they will do, and AIs will do everything else
English
4
0
19
1.1K
Arbutus Tree
Arbutus Tree@aphercotropist·
Top 5 questions this month on MathOverflow ...are the mathematicians ok?
Arbutus Tree tweet media
English
9
9
145
13.9K
Be Dangerous
Be Dangerous@soundpwn·
@LewisCTech First you have to answer what is “good” Only then can you answer what is “good writing” AI doesn’t change the underlying values, it sharpens them Turns out, Boomers actual wanted slop
English
1
0
0
39
Lewis Campbell
Lewis Campbell@LewisCTech·
I can't tell if being a Wordcel makes you more valuable now because writing authentically in this Slopocalypse is rare. Or if it doesn't matter because literacy has declined so much that no one will read what you write anyway.
English
23
1
29
1.9K
crackalamoo
crackalamoo@crackalamoo·
@egrefen It's unclear what kind of value 3 will provide over 2 though. Personally I think some aspects of original thinking will remain uniquely human for the foreseeable future, but I have no way to prove this, any more than we can prove GPUs go brrr is enough
English
0
0
0
3
Edward Grefenstette
Edward Grefenstette@egrefen·
There will be 3 kinds of scientists in the coming years: 1. The Blenderists, who cover their eyes to ignore the impact of AI. 2. AI scalers like OP(?), who think everything can be solved by making GPUs go brrr. 3. Actual researchers who embrace the tech and explore new frontiers.
will depue@willdepue

academics are unprepared for the coming world where much scientific progress is majorly a function of inference compute. whether OpenAI points the Eye of Stargate at your particular field will decide its acceleration. talent will leach away into the labs. it's already begun

English
10
7
155
20.6K
Dwarkesh Patel
Dwarkesh Patel@dwarkesh_sp·
Currently it is shocking and newsworthy when AIs solve an important open problem that humans couldn't Before AI totally surpass us intellectually, there will be an interesting era, where it will be just as shocking (but not impossible) for a human to solve a problem AI couldn't
English
88
53
1.2K
89.2K
crackalamoo
crackalamoo@crackalamoo·
@ai_sentience No. It's controversial to say in this era, but some LLM shortcomings really are intelligence problems, not context problems.
English
0
0
0
0
crackalamoo
crackalamoo@crackalamoo·
@ThePrimeagen At that point it's not you, it's the agents. No human can manage 10000 agents. Then there's no advantage to hiring a 10000x engineer over just spawning 10000 agents yourself
English
0
0
0
0
ThePrimeagen
ThePrimeagen@ThePrimeagen·
Honestly why stop at 100x engineer? Just use more agents, you literally could be 1000x, 10000x, 100000x just by scaling You could what you use to in an entire year in one second
English
298
157
4.7K
180.2K
crackalamoo
crackalamoo@crackalamoo·
@nessumsarkirneh The difficult question is if AI is able to handle all the levels of abstraction in math but these top few, how will humans get trained to reach those higher levels? You can restrict AI use or have the AI teach you, but the overall impact on learning is likely to be negative here
English
0
0
0
1
Henrik Rasmussen
Henrik Rasmussen@nessumsarkirneh·
My controversial (?) take is that this will trigger a much-needed discussion: what is maths? what is important maths? When “electronic brains” first emerged in the 40s and 50s, many expected general AI to be imminent, because the machines could perform calculations fiendishly difficult to humans. Three human generations later, the machines now operate at about two levels of abstraction above the electronic calculator, which is where many combinatorial and number-theoretic problems can be addressed. But inventing the notions of a Hilbert space or of a Riemannian manifold - or more generally, of seeing “analogies between analogies”, to quote Kac - requires yet another two levels of abstractions above the current AI level. IMO, this is where the important maths lives. Will LLMs or other AIs get there soon? Maybe, but we are certainly not there yet.
Jason Abaluck@Jabaluck

I suspect math will be like Chess and Go due to verifiability. The period of fruitful collaboration between humans and AI will be short (i.e. a few years or less, not a decade). Progress in math will be jagged, with harder to formalize fields coming last, but I suspect this jaggedness will be compressed in time -- I expect superhuman performance at (nearly?) all areas of math within a few years (a few = 2-3?). AIs will also be better at asking pure math questions than humans, and will quickly develop theories beyond human comprehension. Human theorists will have a recreational comparative advantage over other humans in understanding these theories, but AIs will be better at communicating these theories to applied researchers. Pure mathematicians will need to become applied researchers to do productive work, until applied research is also automated. Confidence level for prediction: 50-65% for gist, 40-50% for all above claims being correct.

English
3
0
5
855
sankalp
sankalp@dejavucoder·
when someone asks you something or you are trying to solve a problem, you will usually get a first set of familiar thoughts. let's call them cached thoughts. i am curious how do people get past them.
English
19
0
47
3.9K
Lazarz
Lazarz@Laz4rz·
Want to make it in life? You need dedication of people walking in puffer jackets on a sunny 25C afternoon.
English
1
0
12
389

Keşfet

@willdepue @azi_pat @vikhyatk @__drewface @not_ellington @nishffx @abenz_mato @aaron_epstein