therobots927

I was meaning to imply that yes assuming we had a proportionate amount of birdsong data, would we be able to reverse engineer their flight abilities.

I think given the fact that spatial reasoning is nearly universal among species, we can very safely assume that it is “too evolutionary old and too low-level, too far removed from higher cognition, to leak into language heavily”

I think this is pretty apparent. It’s very rare for athletes to talk through their actions in high level detail - I saw the ball coming towards me at a 37 degree phi 23 degree epsilon angle at a speed of approximately 20 mph, I estimated it’s time to arrival would be .45 seconds etc. The eye-hand coordination occurs almost completely outside of what you consider conscious awareness. And it’s not easy to describe that’s why athletic coaching is difficult to do through words alone.

As far as ARC-AGI goes it looks like last years models were scoring <5% against their v2 benchmark: https://arxiv.org/pdf/2505.11831

Frankly I don’t understand why you can’t train a multi-modal LLM on video game frame data. Is that just way too compute intensive to do? What am I missing here? Because I think it’s crazy to think that an LLM could learn to think spatially just from reading… even if they’re reading everything that’s ever been written. I think that about summarizes my position.

If OpenAI cures ANY form of cancer then I will admit I was wrong.

Until then all we have is a lot of hot air coming out of Sam and Dario’s asses

A literal bird brain would outperform an LLM on most spatial reasoning tasks.

Extrapolating the core theory of LLMs - that we can reverse engineer reasoning through language - does that imply that if we train a bird song LLM to predict next “token” (pitch) of a birdsong, that the LLM could excel in a bird flight simulator?

I think it’s pretty clear that this is a dead end.

This is the fundamental question and don’t you find it interesting that there isn’t a nice clean dashboard on the openAI website where we can go and see this metric progress over the release history?

Toby Ord did what he could with public data and it… doesn’t look great.

https://www.tobyord.com/writing/hourly-costs-for-ai-agents

Aren’t we past the point of needing benchmarks? If we’re as close to AGI as Sam says then the proof should be in the pudding. OpenAI should build a competing CRM / Figma / Photoshop with a couple dozen engineers and a Dyson sphere’s worth of compute and just prove the capabilities.

This all feels like a 2024 re-run. Oh, ChatGPT is going to cure cancer? Then find ONE rare cancer and CURE IT. OpenAI has access to the best models and compute - so cure fucking cancer! What the fuck are you waiting for?

Oh you mean the organization whose intellectual property was stolen by AI labs?

Aaron Swartz got 35 years for publishing paywalled JSTOR articles for free while the AI labs are allowed to profit directly from the paywalled material they stole. Is that a dichotomy you’re comfortable with?

Funny how nearly all model improvements this year are demonstrated on the subset of use cases where brute force / reinforcement learning is most effective:

Robotics (using physics sims)

Cybersecurity (red team / blue team)

Math (using automated proof checkers)

Programming (using compilers)

For the record I think robotics is a totally logical place to use this training approach and this is very impressive. But if we zoom out and think about LLMs in general I’m not sure this inspires confidence in AGI arriving any time soon. I would also propose that this is a form of overfitting / training-test contamination.

Take cybersecurity for example. Through brute force techniques you will gradually memorize all of the possible exploits. So when fable breaks into a DoD network everyone is shocked but in reality it basically memorized all possible exploits including some zero day.

I’d be much more interested to see if fables performance is preserved as new exploits arise (NOT zero day - negative day meaning exploits that don’t exist yet). Would fable still find them? Or would they need to retrain it on the new software stack continuously in order to identify the zero days.

This is an important distinction that I have not seen made before.

This analysis by Toby Ord demonstrates why it’s a problem if frontier improvements are coming from reinforcement learning (brute force methods) from a purely computational perspective: https://www.tobyord.com/writing/inefficiency-of-reinforcemen...

Welcome to the den of snakes and idiots that is HN I guess

Downvoted for stating the obvious lmao

Ah, Pascal’s wager used to justify trillions in spending.

Not very creative.

China is spending about 15% of what we’ve spent on datacenter buildout. If we’re in a race with them it would seem like they aren’t even trying to beat us.

I wish we did this here in the US. Here it’s the opposite - white collar criminals get “club fed” treatment - good food, comfortable room, tennis courts, etc.

And that’s if they’re ever charged at all, which is rare.

Well absent a research breakthrough the only option is “scaling” which according to Sam Altman was logarithmic growth in model ability with number of weights.

Of course any stats undergrad could tell you this was a fairytale. Increasing number of weights only works until you exhaust the signal in the data. I’m not sure how he was allowed to get away with such a blatant lie but here we are.

This lie is effective because on a small time scale it’s impossible to tell the difference between logarithmic growth and logistic growth. If you maintain a fixed training data size, increasing the size of the model will get you logistic growth in model capability meaning that past a certain size you get effectively no gain in performance because you’ve already squeezed out 99% of the signal.

This disproves point number 1 in Sam’s thesis: https://blog.samaltman.com/three-observations

“The intelligence of an AI model roughly equals the log of the resources used to train and run it.”

He is playing loose with the language here because the only way this statement holds is when resources = breadth and depth of training data - not compute / model size.

No doubt this will be used to justify a government-funded capital lifeline to OAI and Anthropic who are still bleeding cash (irrespective of whether inference is profitable) and may have issues raising more money in the private markets if they are signaling a delayed IPO.

The proposed “donation” of a 5% stake to a sovereign wealth fund creates a direct incentive for government cash infusion.

I really can’t begin to describe how angry this possibility makes me. And I don’t think I’m alone. Keep pushing the envelope Sam / Dario and see what it gets you. Doubling down on a losing bet just digs your hole deeper.

What happens when the government sinks half a trillion dollars into this and we still don’t see an ROI / true agent autonomy? Then what? Ask for another trillion dollars and hope you can stumble on a research breakthrough equally as revolutionary as the transformer?

“The autonomous software engineering agent vision is more or less coming to fruition.”

Oddly enough no evidence of this is provided. I’ll admit it’s possible to have Claude write 100% of your code - however this is very distinct from autonomy. If I iterate on the code with Claude - telling it what to change, checking for errors and reporting them back to Claude, etc., I can build an app without “writing code” - but that is totally different from autonomy. And it’s not even clear how much of a time saver this really is if I know the language myself.

Well those efficiency gains have to show up somewhere. It would imply that consumers / customers of these companies are receiving cheaper or higher value services / goods.

Thats at odds with current inflation trends to say the least.

Found something: https://philarchive.org/archive/FRAAQI

Regardless of the facts, it would be a lot better if Jim Carrey directly addressed this. I don’t blame people for falling down conspiracy rabbit holes when someone they look up to dramatically changes their appearance and doesn’t say anything about it.

Very interesting. Do you have any links to material along these lines?

For the record I can’t wait for the “ideologically based” comments after this trillion dollar Jenga tower comes crashing down. This place will be censored to shit of course. But that won’t stop people like me from rubbing it in.

You know being a Ycombinator proxy, you would think you guys could take a little bit of shit. You certainly don’t hesitate to shove ai down the throat of everyone, without consent obtained.

Haha. Yep there you go

Oh you sure can. That’s all the boosters do on here.

Anyone can explain why this was flagged?

I don’t consider myself a wizard.

You seem a little full of yourself. (That’s putting it charitably. You seem full of shit.)

As if the pro-AI crowd doesn’t have a vested interest in how AI is perceived.

Ed Zitron is a journalist. Do you say that about every journalist?

I never said it was useless. But you better be able to quantify the value of something you’re spending trillions of dollars building.

This seems like common sense to me.

Would you be able to quantify how useful it is? Trillions are at stake here, so you’ll need to get specific.

AGI is clearly right around the corner. It might not be able to make an accurate diagram of a cancer research study but it’s gonna cure cancer in no time…

Sounds like laziness to me

Fair point!