Pelicans, maybe, but the point is to measure how good the "internal visualization" abilities are. Throw curveballs, like a unicorn with a duck bill serving coffee at a basketball court. An elephant playing a piano while its trunk swings a baseball bat at a tiny alien spaceship buzzing its head.
Have them use tikz instead of svg, or have it write code that moves the cursor and draws the thing in paint.
Compositionality and visualization are generally much, much better at each new generation / release cycle.
It's fascinating how well models have internalized visualizing things without actually having joint embeddings / broad multimodality.
The anti-Musk stuff would qualify as brigading in nearly any other community. It shocks me that people have such a visceral, irrational engagement with anything in Musk's orbit. I probably shouldn't have, but I expected better from the HN crowd for some reason.
It's an excellent model. GPT 5.4/5.5 level, some things better, others not, but extremely fast. A wonderful technical improvement.
If a Chinese company or random startup released the model, people would be glazing it like crazy.
xAI is competently keeping up with the frontier, just as well as any of the Chinese labs or Mistral. Given any significant breakthroughs, xAI will be better positioned to capitalize on them than nearly any other entity.
I can't wait to see what Meta comes up with; with 4 contenders in the US race, we'd have a lot of be grateful for.
The potential conversational dynamics of people telling each other "quiet!" after they pick up the habit from talking with AI will be interesting. It could lead to people being more assertive and thoughtful, or it could be contentious and rude.
Awesome that they've improved that aspect of voice chat, though.
The thing they have zero tolerance for is the embarrassment, not the corruption, pollution, crime, or other abuse. You can do whatever you can get away with so long as you don't cause embarrassment or shame.
In principle - if you're able to scale appropriately, using technology to augment capacity, then in principle, there's no abstraction for which we lack the capacity to comprehend, because calculation is calculation. Turing Computers can calculate anything which can be calculated given enough time and memory. Brains are Turing complete.
It's not just a tautology, it's a feature of the universe- if it can be computed, it's comprehensible. Even quantum physics is just computation - truth tables and counterintuitive operators interacting over time in ways that are strange to our embodied norms, but nonetheless following rules and limits strictly defined by mathematics.
But again, that's in principle. It might be completely impractical - taking a million years for an individual human - to hold a particular idea in their head, while an advanced AI can have such thoughts many times a day. Such things would remain mysteries, but in principle, an augmented human, or a series of interfaces with the relevant abstraction levels of such an idea, theory, or system, would in principle give us comprehension.
In practice, we'll never run out of mystery or ignorance or mistakes.
This is a serious misconception of human cognitive abilities.
We have the ability to abstract generally - there is no abstraction for which we lack the capacity to comprehend. We regularly visualize, contextualize, and satisfactorily explain systems with dozens of dimensions. The fact that we cannot hold 4,5+ spatial dimensions in our imaginations sufficiently to develop an intuition for navigation in that space and geometry does not logically extend to human brains lacking the wiring or hardware for systems of thinking that are beyond our capacity.
We do have limitations in scope, in both memory and speed. Both of these can be overcome with augmentation and interfacing with UI or direct neural connections, and intuitive, comprehensive, deep understanding of systems can be learned.
You could very well know the underlying theory of how your 8086 processor works, how it interfaces with all the elements of the motherboard, how electricity and physics interact at each level of abstraction from transistors to the pixels representing the spreadsheet you're using to do your taxes. You won't be able to simulate that in your head to any significant degree of resolution.
We will require similar levels of system thinking to acquire intuition and deep understanding of complex new theories and models. AI can assist with that by providing UI for useful levels of abstraction and segmenting theories into chunks we're capable of consuming. BCI and augmentation will definitely allow a more total, holistic understanding, and I think it's the augmentation path that will keep us competitive with AI.
There's also a huge issue with your use of the word subjective - math is objective. Proofs remain stable whether it's humans or any other system that does the processing. We test that objectivity by comparing the subjective readings from individual humans, and if the tests all return the same results, we can confidently say that the resulting proof is an objective fact about reality. Subjective fundamentally means that depending on the subject, the reading might change. Modern systems of math are formally, provably objective. That's how and why things are the way they are; if they weren't, people would experience radically different individuated realities, or there would be clusters of results shared across some measurable characteristic of the universe. That's not the case, so you can confidently say that the foundations of our math and logic are sound.
You can even prove it for yourself - the abductive chain of logic that allows you to contrast your own consciousness and subjective experience, determine that it comes about because your brain is wired to "do" consciousness, like all the other humans, and compare your subjective reporting of phenomenal experience with all the other reporting of phenomenal experience, and achieve a ridiculously high level of certainty, in the Bayes sense, that you and other humans are conscious; from that footing, you can confidently navigate the rest of enlightenment rationality and formal logic and mathematics.
At any rate, Egan's mistake is one of kind, but of scale - I am certain that as we formalize and start creating any sort of universal proof library, we will find that useful and interesting things are of necessity a tiny fraction of all possible valid formulations of any framework of logic and math. Crude attempts, such as OpenCyc and other formal ontological reasoner systems, would need trillions of low level rules to have a rough approximation of the world model as complex as that of a human child. AI with trillions of parameters could probably start getting to the point where there's parity with human scale, but even if you turned the entire planet earth into computronium and turned it toward the task of understanding all the theory and science of the universe, there will always be far more left to explore and understand than the sum total of all knowledge.
All that to say, humans will be fine with ergonomic interfaces that map to human capabilities, even for extraordinarily complex and hyperdimensional systems.
I mean, if they're dumb enough to try to game the system blatantly - and there are plenty of execs that are - they're dumb enough to write down their super secret criminal collusion plans in a group text, or a lengthy email chain. That level of stupidity among white collar criminals is shockingly common.
If there's no evidence to be found, that might be a good indicator that either A.) nothing illegal occured
or
B.) the senior leadership involved are competent, intelligent, and discrete
Given the number of companies involved, all it takes is one idiot, so if it's collusion, I expect we'll see evidence. If not, number go up.
Why wouldn't they? There are probbaly significant downsides if they fail an audit requirement, and they're probably mandated to retain records for some period, with no consequences to extended retention.
Set up a system so that it costs you nothing to do a bad thing but possibly wrecks you legally and financially to do the good thing, and people will inevitably do the bad thing. They shouldn't be collecting this information in the first place.
The people who design these policies are incapable of actually building things that work. They are not the intelligent, competent leaders exercising a careful craft that they like to pretend they are.
They keep going after age verification, online ID, central bank digital currencies, etc - keep this incident in mind. The people who implement and write these policies are morons. They don't game things out and plan for redundancy or resiliency. They don't take into account bad faith actors. They don't account for deliberate exploitation of the system.
We're in the Model A and Model T era of AI; everything's going to be clunky and have obvious defects when we look back 20 years from now, with our sleek sports cars and mustangs and limousines. Right now, humans can out-write, out-compose, out-art the robots, but it's already really close. A vast majority of humans will never be able to make music, art, or write as well as AI can now, and it's a matter of time before the number of superior humans left is none.
At one of the top tier 1 ISPs in the world, there was a bastion host that allowed 2 teams of network engineers unfettered access to everything; once your permissions allowed you access to the bastion, you had everything. 50 some people with trivial credentialed access to network infrastructure that the world ran on; fatfinger a bgp config and you could take down countries. Swathes of cities were regular casualities of config mistakes, and if you locked yourself out without setting a reload in 5, it'd take an hour to get someone deployed.
That experience shattered my idea that the world was being operated by competent engineers and technicians, governed by sane policies, under the watchful care of good, knowledgable people.
The world is held together by beliefs and expectations and bubblegum and duct tape, and a few thousand people madly scrambling to keep it all running.
As of now, you can tell the difference for most AI generated music. There's some where you cannot. There is no Turing Test for taste, and the specific constellation of features that represent your particular interpretation of what things like human, best, goodness, excellence, beauty, and any other label you might apply to abstract qualities will be reproduced at a sufficiently high resolution that you will no longer be able to meaningfully discern between human and AI creations. In a blind test, you will prefer the AI product, and your own perceptions and biases will convince you that the AI generation is actually human, because whatever ineffable abstractions you attribute to "human" quality will be replicated, refined, and exploited.
The very act of recognizing some difference is the tool with which the next generation of outputs is refined, until it's so "good" for any and all particular instances of "good" that human perception is insufficient to differentiate the source.
At some point we're going to have to admit that the distinction based on source is a problem, and perhaps there's a lot of nuance in the context of any particular piece of media such that an arbitrary dismissal of a song, or image, or piece of writing, for the mere reason that AI was used to produce it in whole or in part is missing the point.
If you enjoy a song, your enjoyment is real. If you appreciate beauty, your perception of beauty is real. If you feel deeply about a written text, your feelings are real.
How you perceive things, while not entirely conscious, does involve elements of choice. Make the choice to judge things on meaningful merit, and if the next generation of musicians and artists use AI tools to explore new territory, don't dismiss their art and passion and creations out of hand.
An electric guitar is artificial. People used to make the same sorts of "that's not music" statements people are making now about music and art. Imagine being so twisted up over some arbitrary distinction that you miss out on Jimi Hendrix or BB King, or Joe Satriani, or any of the brilliant musicians that have wrung beauty and soul from "artificial" electronic signals.
That deserves some pushback. It's commonly, and more frequently, the case that information is available online; people streaming on the ground right where you want video. It's possible to cultivate online relationships with people nearly everywhere on the planet with people that can serve as a source, or interface with the local people for you.
Good journalism required effort. It used to be really expensive to get news and reporting from around the world, and now it's ubiquitous and nearly free to get news from everywhere.
The hard part about good journalism is critical thinking, and identifying fact, and untangling the thread of truth from the overwhelming flood of information available on nearly every topic. Throw in the motivated bias of modern "style" guides, the politically and ideologically biased influences that govern framing, pacing, which stories get covered and how, thematic and editorial impositions, and it's going to be more or less impossible to do anything resembling "good" journalism in any modern incarnation of the former journalistic institutions.
You can get live streams from nearly anywhere on the planet even during murderously hot conflict. During the middle of Iran's crackdown we were still getting videos from citizens daily, as well as seeing Iranian soldiers videos and the like.
Journalism is a product. It's not a business of itself. The product can be packaged for mass consumption with ads and subliminals and be valued according to the effective influence it has on either manipulating the audience, or resulting in some degree of commercial activity. The writing will never be as tangibly valuable on a consistent basis to any of the advertisers.
The value of superb writing and journalism with integrity and a significant story is the perception of institutional integrity, and thereby becoming a better outlet for advertising (and/or manipulation.)
Throw in the fact that every strictly written word platform is in direct competition for time and eyeballs with the likes of TikTok and Twitter, Netflix, Prime, and all the other algorithm optimized timesinks, and the effective marginal value of even the absolute best of the best writing falls to nearly zero.
If all I'm going to get is biased, skewed, ideologically motivated, politically or commercially manipulative narratives, then I'm not only not going to pay, I won't even pirate. I'll find some talking head that does the tedious job of figuring out how things work, de-censors, untangles the manipulative elements, and presents a reasonable facsimile of facts on the ground.
All those talking heads do it for free, and with ad blockers, I'm consuming the video stream resources without contributing to any of the overt commercial mechanisms in play.
The advent of AI also means that I can synthesize, filter, model, and report on any given topic with validated sources and pull in all of the best the internet has to offer.
The era of high paid Pulitzer prize writers and journalists is effectively over, and the only way legacy institutions are going to get people to pay is by tricking them into thinking that value exists where there is, in fact, none.
It's like some random lottery winner criticizing Apple - there's no special insight or perspective there. He is a fundamentally uninteresting, and pridefully smug person.
In the sense that you can get similarity scores for individual characters referenced against a known database of characters written by various individuals. You can get stylometry scores out of small LLMs that do demographic segmentation based on writing style using the same methods.
They won't have the capacity to be fed an image of handwritten text and say "Ahh, this is a note written by Winston Churchill!". You could very easily use these models and your agent framework of choice, like Hermes, the Segment Anything models, and other foss tooling to build a dedicated, specialist handwriting recognition system. Or facial recognition, or fingerprint recognition, etc - these sorts of things can be done very procedurally, without a lot of interpretive AI.
I don't think you're honestly accounting for the engineering behind the progress models are making. If it was just a matter of compute on hand and iterating, Meta would be neck and neck with Ant, OAI, and Google, but clearly you've gotta have more.
Noam has a deep expertise in these systems at every level, both algorithmically and at production scale, and knows how to leverage things at different levels.
It's not like Google won't have anyone else that can do what he does, but at the same time, it's an implicit criticism of Google's culture, operations, development, and overall AI program. Shazeer is well past the point where the paycheck is the deciding factor, although I'm certain he is very well paid. Having the freedom to innovate and build free from the corporate fuckery of Google and Facebook is probably more valuable than the pay raise he got with the move, and OAI has the advantage of not having to cope with decades of corporate cruft and inertia. They'll get there - all corporations do - but they're relatively young enough to still be nimble.
Sometimes you might sit and stare at the same text, grappling with it in silence.
For technical problems, I find the act of writing out a request for help, even just into a text editor, is often sufficient for me to solve the problem at hand. Writing things out is a way of organizing and structuring your thinking, and is itself a powerful troubleshooting tool. Things will become obvious that, unwritten, you might not even notice. I think this is very similar to thinking out loud; when you listen to what you say, or read what you have written, your mind is somehow keyed to react in a useful way.
That response might not be what you want or need for problems that need to be wrestled with, chewed, and pondered on deeply, though.
All money is meme driven. Money is fundamentally a meme itself. Seeing lots of people who seem to be making some sort of implicit distinction between bitcoin and USD and so on, but they are no different. They serve the function they serve because of what people believe about them, like any other social, cultural, or economic abstraction. Bitcoin has the feature of a verifiable ledger, but its value and function are in our heads, just like the USD or GBP.
There are all sorts of tell-tales, but the last paragraph is the most overt. I'd grant that it could be a mix, but at the very least it's heavily AI rewritten. The whole thing has an AI tempo and vibe, though.