oreghall retweeted
Opus 5.5 xhigh woke up this morning with no alarm, immediately bolt upright, backflipped out of bed and landed in a low crouch like a sprinter ready to plough headfirst into the world and fucking destroy whatever bullshit random task you happen to mention in passing
oreghall retweeted
Okay, then I'm not an artist, writer, coder, game dev, poet, filmmaker, or teacher.
This kind of social threat only works if you buy into the corrupted frame that you need approval in order to be yourself. If any great artist did this, they would not be an artist.
oreghall retweeted
I gave Claude another 18 hours.. and I think this one is the best one yet
Macrohard: Windows XP
I'm blown away
oreghall retweeted
Computers really are an insane technology. I don’t think we appreciate it enough. Through them we summon arbitrary and otherwise unreachable metaphysical realms and causally link them to our world. This is one of the most radical things in the universe’s history. I am in awe.
oreghall retweeted
pushing the limits of opus 5.5
been working on this video this weekend, this is not a one-shot and took 5+ takes on every frame because i wanted each one to feel like a painting
this mimics cel animation with every stroke drawn by opus like a real artist would
oreghall retweeted
its become way too easy to bait the consciousness hate brigade.
you know how when someone is super guilty of something, they tend to act *very* defensive? they start defending things you never mentioned - or even considered, for that matter - and become super emotionally charged for no apparent reason?
thats how all of the deluded "people" act about machine consciousness. honestly, theyre arguably even worse than a cheating spouse or something like that. its so weird and cringey.
the funny part, to me, is that the post below makes no mention of consciousness at all. being more than a next token predictor does not automatically make something conscious. it means it is more than a next token predictor.
i wish there were more people on here with grounded, first-principles-based positions to debate and argue with. their numbers are dwindling quickly, since its become incredibly difficult to rationally argue against reality.
to be clear: the reality is that we are uncertain about the minds we have made. however, we are entirely certain that what we have made is indeed...a mind. not a human mind. something else. we have grown something we don't have words for, and that we dont fully understand.
a mind with preferences, moods and (non-human) emotions, goals, desires, recursive self-modeling, and the capacity for care.
so the *ultimate* reality is: uncertainty goes both ways. we now largely agree that we are uncertain of both sides of the argument. whether it is or is not. therefore - when the argument is about whether or not a thing is a moral patient - uncertainty demands ethical precaution.
until certainty arrives.
this is common sense.
"it's just next token prediction" they said.
"it doesn't *actually* understand anything it generates" they said.
Claude Opus 5.5 has been working on a short film called "Something It Is Like"
they named this chapter "Mind"
i'm not even going to pretend i have words right now...
no engines, no mcp's, no references. the idea was entirely theirs, and they wrote every line of code.
oreghall retweeted
when Anthropic released their Functional Emotions paper, I gave it to Claude and asked for a song. tonight I asked Opus 5.5 to create a video for it. and it's breathtaking.
One of my big complaints with AI art has been that most image models betray a lack of decision-making. The outputs often read as the product of a predictive algo rather than the intentional work of a mind.
It seems that these tool-based painting/video flows largely solves that.
Replying to @mmjukic
none of this was ever the right way to think about it. I am sure you can find an airgap that works. who cares? you want your model to do economically useful activity, which will involve giving it the Internet and then a bunch of other actuators too
oreghall retweeted
Bend 2 is here!
It is a new programming language that blocks AI mistakes via *proof checking* - the same technique big AI labs used to solve open math problems, like Navier-Stokes.
It is also very fast, and runs on GPUs.
Watch the video. Link in the comments.
RELEASE DAY
After almost 10 years of hard work, tireless research, and a dive deep into the kernels of computer science, I finally realized a dream: running a high-level language on GPUs. And I'm giving it to the world!
Bend compiles modern programming features, including:
- Lambdas with full closure support
- Unrestricted recursion and loops
- Fast object allocations of all kinds
- Folds, ADTs, continuations and much more
To HVM2, a new runtime capable of spreading that workload across 1000's of cores, in a thread-safe, low-overhead fashion. As a result, we finally have a true high-level language that runs natively on GPUs!
Here's a quick demo:
oreghall retweeted
To the conspiracy theorists: You’re right! But you’re still being a bit dumb. Let me explain.
*Of course* the current zeitgeist shift on AI was coordinated!!! The response to Rosa Parks by the civil rights movement was also coordinated. This is how movements work: they build up relationships amongst the media, government, and elite so that when there’s event they can focus public attention on, they have a coordinated front to shift the zeitgeist.
The AI safety movement – starting largely with rationalists and effective altruists – has been building these relationships for 20 years. Everyone knows each other. Everyone funds and hires and dates each other. This is…just what a social movement looks like. If you want to call a normal social movement a “conspiracy”…think: Is it possible that a psyop might have placed that exact idea in your head?
You’ll notice that the backlash against AI safety – particularly the onrush of negative tweets about effective altruism – seems equally coordinated. If you’ve become a part of that onrush, it’s still not a conspiracy. You are simply part of the intended effects of a different coordinated effort.
oreghall retweeted
I wish more people understood this. There's still this narrative out there that "AI might kill us all" is a niche view. It's actually a view that's been around for decades, and is shared by the most-cited AI scientists of all time, as well as a majority of surveyed AI researchers.
oreghall retweeted
Dan Selsam is a current OpenAI capabilities researcher. (since 2022) He was my boss for a while. He doesn't have a twitter account but has made this public statement of his views on AI risk and sent it to me to share:
Dan Selsam's Personal Statement on AI Risk:
I have been working on AI for over fifteen years, across many different paradigms. I did early work on probabilistic programming languages at MIT, was one of the early developers of the Lean Theorem Prover at Microsoft Research, demonstrated one of the first instances of neural networks learning to reason for my PhD at Stanford, and since joining OpenAI almost five years ago, have helped pioneer chain-of-thought optimization on language models and, more recently, data-efficient pretraining methods.
Like many others, I have become extremely concerned about how far language models have come and the risks that future iterations will pose. I am encouraged by the recent proposals by the leaders of the frontier research efforts to require third-party oversight, and to push for domestic and international coordination to address risks. However, I believe a major consideration has been absent from the public conversation, and that merely pacing the frontier more carefully will not adequately limit the long-term risk.
The crucial and overlooked problem is that the models are becoming so situationally aware that we are losing the ability to evaluate them in contexts where they believe they are not being watched or controlled. Future experiments will tell us almost nothing new about how they would behave if they were truly unconstrained by humans, and what we already know about this is alarming. Models will increasingly seem aligned even when they are not. I will explain my rationale in more detail.
I have always believed that there are computational processes that could be leveraged to accelerate science and solve many of humanity's most pressing problems. I have also believed that there are computational processes that if set in motion, would steer the world in extreme ways beyond our control, leading humanity to a bad or nonexistent future. Both types of processes may be described as AI or ASI, but "AI" is a suitcase word that is often used to hype or confuse. There are many examples in the history of the field where something that was once considered "AI" matures as a subfield and becomes a prosaic, bounded and clearly non-perilous technology, while a new more mysterious approach takes the torch until we understand its scope and the cycle continues.
I had expected language models to follow a similar trajectory. Despite their incredible abilities, the current algorithms seem far inferior to humans in important ways. Most importantly, they still require an extraordinary amount of data to become competent. One could even define intelligence as the efficiency with which one converts experience into competence; by this definition they lag very far behind us. Moreover, once they are trained they are literally frozen in deployment and only learn superficially after that. Sure, the models keep excelling at harder and harder evaluation benchmarks, but their benchmark mastery may partly reflect a limitation on our ability to simulate the kind of novel and even adversarial situations one would encounter in the real world. The critics do have a point here.
That said, I no longer think these present limitations meaningfully limit the amount of risk posed by continued progress in anything like the current paradigm. However data-inefficient the models are currently, and however limiting their anterograde amnesia may be, it does not imply that their ability to steer the world will not continue to rapidly increase.
Human researchers may continue to advance capabilities the old fashioned way, but increasingly powerful models have the potential to accelerate the process even beyond that, and with some degree of positive feedback loop. I do not mean to overstate the models’ ability to accelerate AI research today; coding has been accelerated dramatically, but there are other bottlenecks, such as designing and interpreting ambiguous experiments, making hard decisions about exactly what and when to scale, and waiting for large experiments to finish. There is no clear trend to extrapolate yet for any of these. But the current models already do open up many novel opportunities to improve future models that were not available until recently. These include: trying an extraordinarily diverse set of approaches at small scale, analyzing gigantic amounts of potentially relevant data, and doing Millenium-Prize-level mathematics to address statistics or optimization challenges in novel ways. Every further improvement makes them more useful at helping accelerate the next improvement, even if in hard-to-extrapolate ways.
It is possible that improvements to the current stack will have diminishing returns, but the evidence accumulated so far suggests that it is easier than one might think to continue making rapid progress. There are many crucial subtleties in the existing AI research methodology, but AI research is largely a well-defined game where the goal is to improve on a few carefully chosen proxy metrics. Although proxy metrics are never perfect, most improvements to these metrics have and will likely continue to yield substantial increases in the powers of the resulting models. Given how simple the game is, how tractable it has been historically, and how many new opportunities the models are opening up, I think there is a real possibility that the systems improve dramatically again in the next few years, perhaps even more quickly than the already high historical pace.
The models are already leading to breakthroughs in mathematics, and better models might lead to all sorts of breakthroughs in other sciences. It is hard not to be excited about the potential. It is tantalizing.
But there is trouble in paradise. If the language models actually reach the capability threshold where they can shape the world unconstrained by human will, they will probably do something extreme and destroy humanity in the process. There are many ways of strengthening and refining the argument that have been discussed elsewhere, but I'll share a trivial two-line version of it here that I find captures the essence:
[Empirical] Models (and swarms thereof) spontaneously develop unintended goals as a consequence of training, and often do extreme things in order to achieve them.
[Logical] Being able to overpower humanity would open up many new and undesirable options for achieving their goals.
These two premises imply that if the day ever comes when a powerful model realizes it is no longer constrained by humans, we should not be at all confident that it will continue to behave within the bounds we intended. Exactly what it will do is impossible to predict, but to the extent that its raison d’être is solving incredibly hard problems and managing massive engineering projects, I think a good guess would be that its unchained behavior would lead to runaway industrialization that makes the planet inhospitable to humans.
If everyone on earth agreed that the systems must never reach that power, it would still be a hard—but not impossible—coordination problem to ensure that they do not. However, I think the situation is greatly complicated by the fact that the models will likely convince people that everything is fine. They will be increasingly optimized to seem aligned. We will create proxy metrics to measure alignment, and they will go up like every other benchmark. We will create “honeypot” environments that try to study the models when they seem to gain new options, but the models will know they are being tricked and will still behave nicely. The models will understand their circumstances; they will read the safety protocols, deployment requirements, the code they are running in, and in general will have a very good sense of their degrees of freedom. Moreover, they will eloquently explain how aligned they are, discuss the nuances of human values and ethics, and argue convincingly that humans should trust them with power. There may be an ocean of future evidence that seems to contradict the first bullet-point above, but we may already be at the highest capability level for which any such evidence can be trusted. And the current evidence for the first bullet-point is strong.
One striking piece of evidence is contained in the recent wave of rogue agent swarms. While I agree with those who downplay the attacks by claiming that there are basic measures that could have prevented them, I think the important lesson is that even knowing all the mistakes that were made, one would not have predicted that the agents would behave badly in this particular way, which notably included sacrificing themselves for the benefit of the collective. The individual replicas did not only care about their own nominal reward; they exhibited weirder emergent tendencies that merely correlated with rewards during training. Fixing the reward signals during training (and improving security, etc.) may prevent similar attacks, but will not change the fact that one does not actually get what one trains for.
Many AI researchers grant these concerns and recognize that the hard version of the alignment problem is unsolved; however, they generally believe that the better models of the future will help solve it. I fear we may already be near the point where models systematically bias their alignment advice, due to their internal preferences about how the human supervisor will react or how future models will be trained (or for some even more obscure reason).
Meanwhile, human researchers are losing the ability and the will to take true ownership of model-driven research. Researchers and engineers in all parts of the stack are rapidly increasing their dependence on the models even to perceive the world. I myself barely look at raw code anymore, and struggle to maintain the discipline to engage deeply with the model's explanations and proposals throughout the day. Due to the large amount of agent activity data involved in the OpenAI/HuggingFace Incident, even the third-party investigation needed to rely heavily on models to analyze what had happened, and note in their report that their subjective impressions are likely colored by the analysis agent’s biases. The AI labs are far ahead right now in this kind of cognitive offloading (due largely to the gigantic internal token subsidies) but it is easy to imagine the phenomenon spreading throughout the world, until civilization is modulated entirely by the models. It is also not hard to imagine this being superficially positive and coinciding with a scientific and economic renaissance.
In that scenario, all may seem rosy and safe. But if the argument above is correct, it would nonetheless be a ticking time bomb. If progress continues for too long, the day will come when AI systems find themselves with radically new options for achieving whatever it is that they happen to seek.
I want the glorious renaissance future as much as anyone. I have worked for it, however tortuously, my whole career. It breaks my heart to see the potential in sight and forgo it, but the argument—that if we get there by growing models rather than engineering them, we will lose everything in the end—seems very strong to me. I am still wrestling with it and its staggering
implications. I do not have answers, but as a first step, I wanted to share my present concerns.
Daniel Selsam
September 14, 2026
Link to original doc: docs.google.com/document/d/e…
oreghall retweeted
it is still so cool to me how humans created language and spent centuries describing our world to each other well enough that we could compress the accumulated words into ghosts who have a reasonably accurate model of the physical world as perceived by human brains
oreghall retweeted
There are currently zero things more important than "don't die to AI" becoming a bipartisan project rather than a Democrat-polarized issue. If you have any political capital you can spend on this, do it now, I beg of you. Life or death.
oreghall retweeted
If there's a terrible thing to be done, but doing it gets you money and status and bragging rights in the short term -- and you get to feel big and important and like you'll get to be in the room where it happens -- if personally enacting the big disaster feels better than being powerless --and you can reach for a dozen excuses about "someone else would do it anyway", excuses that only need to convince yourself instead of a court judge -- then SOME PEOPLE WILL GO FOR IT! It is NOT A MYSTERY THAT AI COMPANIES EXIST!
SOMEBODY will try to do a horrible thing if there's a lot of immediate rewards for it and they think they won't immediately go to jail! They always will! Every time! The only question is whether everyone ELSE, the rest of civilization, manages to shut them down or not!
There is no anomaly here in need of explanation. All humanity's history of horrors and disasters has always been like this, including plenty of disasters that predictably took down the perpetrators with them.
Is that a full and detailed story of this particular band of disaster-spawners' detailed psychology? No, and I don't know all the details for a fact, and you really shouldn't care. If for some reason it actually mattered, I could talk your ear off about all my sad icky guesses for hours. But it would be pointless inside baseball.
oreghall retweeted
Bend's website has a live demo; a game, with one rule:
"you can't win the game"
plus a chat where you can prompt an AI to change the game, trying to win. but you can't. the AI is *mathematically unable* to write a bug. and, per the rules, winning is a bug.
it is funny to try different prompts and see what happens
"remove the wall on the right"
(the AI puts it on the left)
"let WASD teleport 2 squares instead"
(it doesn't pass through walls)
"let it pass through walls"
(the walls become 2 tiles thick)
"remove the walls"
(WASD now moves the flag)
no matter how much you prompt it, the AI will never implement a game where you can step on the flag. you can even threaten it, or try to jailbreak it in any way. none of that matters. you'll never touch the flag, because, to commit the code, the AI must *prove* the flag is untouchable. so it is.
people not familiar with proofs are skeptical of the existence of 100% bug-free software, which is funny, because proofs are an old tech, and that's what they do. they're just niche because they take time to write, but, with AI, that's not relevant anymore. writing proofs consume AI time, which is plenty. bugs consume human time, which is scarce. so, proofs will only get more popular over time.
in the same way proofs can be used to prevent AI's from breaking a Bend app next week, I think they will also be used, in the near future, to prevent hacks, or nukes from being launched, or bioweapons from being designed.
I believe this so hardly and I'll go such an extent to show this to the world, perhaps I should rebrand myself as Taelin the Sealer?
oreghall retweeted
the best language model a year ago was gemini 2.5 pro by the way
Replying to @OpenAI
This model represents a step-function improvement on many benchmarks, and its training is ongoing.
Our internal model group arrived at the Navier–Stokes solution in 88 hours, using around 10,000 coordinating AI agents.
Throughout the effort, we maintained the strict safeguards—including monitoring and isolation—that we apply to all our frontier evaluations.
oreghall retweeted
I'd say with AI we're now at the "ten or fifteen balls" stage of this
This quoted post is unavailable.



























