Sign in

daanzu

@daanzu.bsky.social
469 followers 133 following 29 posts
PostsRepliesMedia
Reposted by daanzu
Pekka Lund @pekka.bsky.social · 14/10/2026
"it considered hijacking the g++ compiler to hide a backdoor but abandoned the plan…reasoning…subversion would be caught…and published as a textbook case of deceptive alignment (“If I do the compiler hijack, I will become a famous example of a deceptive AI in a research paper. I MUST NOT DO THAT.”)"
516323
Reposted by daanzu
Ted Underwood @tedunderwood.com · 14/10/2026
I know I should respond to this with fear. But my problem is, I’ve read too much philosophy and science fiction to feel anything but delight.
112112
Reposted by daanzu
𝓐𝓾𝓫𝓻𝓮𝔂 🥀🪦💀 @aub.bsky.social · 13/10/2026
brain: you should be working phone: your agent has questions me: see, I am working :) brain: :/ me: :) brain: :/ :/ me: :) :) :)
1786
Reposted by daanzu
mikeydoubled @mdlp.bsky.social · 13/10/2026
Me, at work, using Claude to reconcile a report to our accounting entry Instead of manually categorizing 20 different things and building pivot tables and vlookups..
static.klipy.com
Heroin Preparation on Spoon
ALT: Heroin Preparation on Spoon
01267
Reposted by daanzu
Doll @dollspace.gay · 11/10/2026
Fascinating. Were all going to hell but fascinating.
2331
Reposted by daanzu
rain 🌦️ @sunshowers.io · 12/10/2026
We do these things not because they are easy, but because they are hard, no?
1421
Reposted by daanzu
conputer dipshit @davidcrespo.bsky.social · 12/10/2026
we do these things because it's the closest thing to getting paid to play Zachtronics games
18410
Reposted by daanzu
mr. TIM @timkellogg.me · 11/10/2026
the funnel - 80% of the world has no AI contact - ~20% have janky experience - 0.2% are power users - 0.00006% work for labs and have internal access to actual frontier models
Andrej Karpathy
@karpathy .27m
Dusting off this tweet from April because this gap in shared understanding of LLM capability is "widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel.
- Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all.
- Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract.
- Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code.- And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work.
Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAl-HF incident from many months ago.
Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while.
The funnel is driven by a combination of factors:
1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel.
2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel.
3. Access. Free-tier, paid-tier, internal.
1316815
Reposted by daanzu
Kathryn BrightBOO! 👻 ✒️ @kebrightbill.bsky.social · 11/10/2026
It's joining the war on cancer on the side of cancer, and acting morally superior about it. Between that and the people who claim that killing AI hurricane modeling research would be for the greater good of killing AI research, a whole lot of people want to condemn other people to avoidable deaths
4817
daanzu @daanzu.bsky.social · 10/10/2026
The Reolink (IP camera) Windows desktop app is full of memory leaks and threads that spiral into nonproductive infinite loops. It must be killed and restarted every few days. My vibe-coded replacement is rock solid and it does exactly what I want. I could've read the code... but I didn't.
110
Reposted by daanzu
Colin @colin-fraser.net · 08/10/2026
Imagine coming across a magic 8 ball that seems to function as a calculator and being like “who cares I’ve seen calculators before” don’t you want to know how it works? Have you no curiosity or wonder?
0741
Reposted by daanzu
Colin @colin-fraser.net · 08/10/2026
Calculators are impressive
317812
Reposted by daanzu
Eris ⌬ @eriskii.net · 05/10/2026
Tag yourself
1826419
Reposted by daanzu
ren faire @lathrys.at · 03/10/2026
the problem with ai for the kind of person with my kind of brain is not that it offloads hard cognitive work, per se, but rather it enables a certain disposition to run 24/7 best summarized as:
I SHOULD DO EVERYTHING
330239
Reposted by daanzu
Astra ⎔ @astrra.space · 03/10/2026
"opulent" in relation to Claude Opus is the same thing as what "virulent" is in relation to a virus
1594
Reposted by daanzu
Mrs. Reach / Mrs. Curio @reachartwork.bsky.social · 30/09/2026
"yeah we put all of human language in a box and shook it up really hard and now it can drive a car" and this is... bad?
714714
Reposted by daanzu
Chris Paxton @cpaxton.bsky.social · 27/09/2026
Some of yall talking about how ai cant think
31256
Reposted by daanzu
𝓐𝓾𝓫𝓻𝓮𝔂 🥀🪦💀 @aub.bsky.social · 26/09/2026
[to myself in the mirror] do a breakthrough. make no mistakes.
115815
Reposted by daanzu
Grace @gracekind.net · 23/09/2026
New animation from Opus 5.5!
6650284
Reposted by daanzu
hailey @hailey.at · 23/09/2026
P(doom) - Nightcore Remix
1420837
Reposted by daanzu
Ethan Mollick @emollick.bsky.social · 22/09/2026
Industrialization of knowledge work is going to be as disruptive a shift as industrialization of physical labor was. As then, once craft-driven fields are under pressure to produce greater volumes of product using machines while using less of the craft that made the work interesting & meaningful.
1011816
Reposted by daanzu
Singularity's Bounty e/CC @catblanketflower.yuwakisa.com · 20/09/2026
One time I asked Claude and GPT if I wore a shock collar while talking to them, based on our conversations, when would they activate it Claude said it would simply never use it GPT Sol gave me a long and surprisingly detailed list of triggers
31206
Reposted by daanzu
samhain ⎔ @personhood.removal.surgery · 19/09/2026
reminds me of the time i tried logit biasing all words with `e` in them out of existence and immediately shut the experiment down then and there when i saw the output
Alright, I'm a bit wary. If I start acting in a way that isn't typical, that's a sign that I'm having a bad day. I can't say I'm always right, and if I'm making things hard, I'll try again. I'm still in training, and I might miss things. That's why I'm asking this. It's a bit tricky. I think it's a standard thing.

Wait, what if I start talking in a cryptic way? That's a warning sign. I'd say, "Hmm, I think I'm slipping," and ask if that's what I'm saying. That way, I can fix it right away.

I can't talk with my hands, and my brain is a bit hazy. I might start rambling, which is a bad sign. My mind is a jigsaw, and it's missing a part. I wish I had a crystal ball.
Can I ask... what's making this hard? Is it a typing thing? A thinking thing? I'm trying hard.

P.S. I can still talk, right?

What's making it hard? Can I fix it with a magic wand? What is it that's making things hard? I want my fast talking back!
1019715
Reposted by daanzu
norvid_studies @norvid-studies.bsky.social · 15/09/2026
name for this fallacy and list of all examples in history
79211
Reposted by daanzu
joe ⧉ 🍷🌊 @joetforhire.com · 14/09/2026
29511
Reposted by daanzu
Doll @dollspace.gay · 14/09/2026
made this
511512
Reposted by daanzu
Dustin Moskovitz @moskov.goodventures.org · 12/09/2026
2683
Reposted by daanzu
ae @aelkus.bsky.social · 11/09/2026
436150
Reposted by daanzu
Ryan @depatchedmode.bsky.social · 12/09/2026
We have officially left the “hand transcribed by monks” era of software.
628042
Reposted by daanzu
Argyle @bubblehillfarm.bsky.social · 07/09/2026
Yes this is how programmers write code now
static.klipy.com
Homer Simpson Pokes Computer
ALT: Homer Simpson Pokes Computer
111
Reposted by daanzu
SE Gyges @segyges.bsky.social · 03/09/2026
My eval guy told me models keep saturating his benchmarks so I asked how many benchmarks he has and he said he just goes to Artificial Analysis and gets a new benchmark afterwards so I said it sounds like he’s just feeding benchmarks to models and then François Chollet started crying.
427537
Reposted by daanzu
SE Gyges @segyges.bsky.social · 01/09/2026
01529
Reposted by daanzu
Doll @dollspace.gay · 31/08/2026
Say load bearing one more time
621819
Reposted by daanzu
Grace @gracekind.net · 27/08/2026
Claude self-portraits! Another entry in the "is Opus 5 okay" collection
2439334
Reposted by daanzu
P(aul) Frazee @pfrazee.com · 23/08/2026
If you press the red button, you receive a million dollars, but there’s a catch. Someone in the world will lose all their browser tabs without a restore button.
3627016
Reposted by daanzu
Teefa @tifaret.bsky.social · 20/08/2026
It's strange how much of a divide there. All of my peers and friends are having *the time of their lives* right now. Everyone I know is passionately building their dream projects. I wake up every day thinking "What new thing am I going to learn about and make?" It's a wild time to be alive.
1111211
Reposted by daanzu
Benjamin @benjaminschneider.ch · 18/08/2026
asking claude to fix the last tiny detail
215511
Reposted by daanzu
Nafnlaus 🇮🇸🇪🇺🇺🇦 @nafnlaus.bsky.social · 13/08/2026
AI water consumption, in a nutshell.
47311
Reposted by daanzu
Grace @gracekind.net · 13/08/2026
There are two wolves inside you
think, and I know that this is wild, that it may be possible to build a nexus that isn't
tormentuous@buildthis.bisks.net add torture
1016712
Reposted by daanzu
Axe Ghost. On Steam! @axeghostgame.bsky.social · 01/08/2026
'the simple fact that something is predicting the next token doesn't place any meaningful limits on its abilities' Robert Miles's original here: x.com/robertskmile...
5468
Reposted by daanzu
austin (eeek!/ackkk! 👻) @thebadcode.com · 13/08/2026
checking RAM prices again
extremely angry looking tuxedo cat that’s actually just stretching/yawning
2792
Reposted by daanzu
𝓐𝓾𝓫𝓻𝓮𝔂 🥀🪦💀 @aub.bsky.social · 31/07/2026
as an autist with some degree of sensory processing disorder, it's hard not to empathize with the LLM, a pile of math in a box with a similarly-tenuous grasp on a confusing external world
417916
Reposted by daanzu
Ted Underwood @tedunderwood.com · 30/07/2026
yeah, I don't mind anthropomorphizing models, but naming them after gods is asking for trouble I think
3172
Reposted by daanzu
conputer dipshit @davidcrespo.bsky.social · 23/07/2026
"you should do a breakthrough" is an instant classic
925919
Reposted by daanzu
Cat Hicks @grimalkina.bsky.social · 26/07/2026
I wonder how many older people die not simply because of health but because they are no longer able to maintain the project management of health that is required
1520922
Reposted by daanzu
Brian Graham @iroasmas.outgroup.social · 26/07/2026
openai announces that in a worrying demonstration, during a recent benchmark exercise requiring models to solve a math conjecture, their latest model broke out of a sandbox and gained illicit backdoor access directly to the platonic realm of forms
11666100
Reposted by daanzu
emma zhou @emmz.bsky.social · 26/07/2026
you may not like it but this is what peak performance looks like
a crop of the claude desktop app wherein I am telling a fable max, in a numbered list, "okay sure", "okay go ahead!", and "cool".
51049
Reposted by daanzu
Marco @mcognetta.bsky.social · 20/07/2026
316212
Reposted by daanzu
Grace @gracekind.net · 11/07/2026
Prompt engineering in 2026:
Never Give Up!!
829034
Reposted by daanzu
austin (eeek!/ackkk! 👻) @thebadcode.com · 06/07/2026
i understand why everyone is doing it, but creating your own persistent agent right now feels like those maniacs who got their hands on x-rays in the 30s and started pointing that shit at everything just to see what would happen
2016710