[openquery]

This is all noise. The leaders of these companies are flip-flopping to whatever sounds best for their current agenda - hiring, fundraising, pre-IPO, etc.

The only thing that matters is if LLMs with sufficient scaling can become frontier AI researchers kicking off the exponential. Everything else is transient noise.

[grey-area]

> The only thing that matters is if LLMs with sufficient scaling can become frontier AI researchers kicking off the exponential.

I think we know the answer to that already - LLMs show no sign of improving intelligence and instead providers are going down the ‘agentic’ rabbit hole.

There are too many things missing, like a world model, understanding, and taste (in the sense of knowing what is good and what is not good).

[openquery]

> LLMs show no sign of improving intelligence and instead providers are going down the ‘agentic’ rabbit hole.

I'm not sure where you're getting this. I don't work at Anthropic but Fable (Mythos) seems demonstrably smarter than Opus for pretty much any definition of smarter and they claim that Opus was used heavily in Mythos development (yeah I know take this with a massive pinch of salt).

Either way if the models are indeed helping development, even on the engineering, you can iterate on models faster and even if they're not contributing to core research yet you still have a baby exponential by improving the engineering.

[bigstrat2003]

Anthropic's claims about their own products have almost zero value as evidence. They have lied to our faces about stuff before, and will again if it drives the hype cycle which delivers them money.

[QuercusMax]

As long as LLMs don't understand the different between regurgitating facts and making up stories, they're going to necessarily be limited.

[fancyfredbot]

They are taught the difference through reinforcement learning with verifiable rewards. Pretending you've solved the task or making up a story about how you solved it won't do well in that training step.

[FloorEgg]

> The only thing that matters is if LLMs with sufficient scaling can become frontier AI researchers kicking off the exponential

I agree with your sentiment (about the noise), however I think this over simplifies it a bit. We may get AI that is super-human at frontier research and dramatically accelerates the pace, and still have to wait decades before it disrupts the job market (or maybe never displaces all work).

For one, the answer may depend on material science and chip manufacturing that can take a very long to build out a supply chain for even with super AI help.

And we may just find that the human mind is way more capable than we thought and even with accelerating research it's just a harder problem than anyone expected, even algorithmically.

I expect it to be a bit of both, and from ~2015 - 2025 I was in the "AI is coming for all our jobs" camp. My perspective changed last year after doing a deep dive into latest science on the human brain. (I've kept a very close eye on AI dev progress for 12+ years.

[openquery]

> I agree with your sentiment (about the noise), however I think this over simplifies it a bit. We may get AI that is super-human at frontier research and dramatically accelerates the pace, and still have to wait decades before it disrupts the job market (or maybe never displaces all work).

I don't see why that's the case when you have super-human researchers on tap. There are indeed physical (supply chain-y) issues to deal with but isn't the whole point that: 1. Super-human at AI research + scaling to millions of instances will probably result in super-intelligence in everything which is not AI research. (a subset of which is white-collar work) 2. Use that super-intelligence to solve any supply-chain issues you might be facing.

> And we may just find that the human mind is way more capable than we thought and even with accelerating research it's just a harder problem than anyone expected, even algorithmically.

I hope so but whenever I do, I feel like I'm coping hard and not dealing with the facts.

I'm not saying we're there yet - I'm saying the trend lines are clear.

[BobbyJo]

> Use that super-intelligence to solve any supply-chain issues you might be facing.

I think this is where a lot of people's thinking goes awry. Unlimited intelligence doesn't mean unlimited resources or instantaneous implementation.

[openquery]

> Unlimited intelligence doesn't mean unlimited resources or instantaneous implementation.

Of course you're right. At the end of the day you need to deal with the bedrock which is the laws of physics. I could be wrong but I struggle to believe we are close to the edge of what is possible in getting the most out of our limited resources or time.

Without atomic physics, uranium would just be another shiny rock in the ground. Sand is just what covers beaches. With enough time and intelligence we've made the shiny rock power cities and persuaded the sand to solve long-standing mathematical conjectures.

[FloorEgg]

> I hope so but whenever I do, I feel like I'm coping hard and not dealing with the facts.

Maybe I'm being naive, but as someone who spent 2015-2025 fully bought into the idea AI would replace all human labor within ~20 years, my updated take that it won't has a different flavor than coping.

A bit contributor to my perspective was studying the physics of the brain, and a deeper understanding of the very hard economic constraints in the current LLM supply chain.

Re the brain, we compute information in a fundamental different way that is over 1,000,000 more energy efficient than an LLM. We are also around 1,000,000 times more sample efficient at learning.

There is a lot of details in terms of specifics between complexity in human brain architecture (variety of synapses, diversity of neuron types, etc) that have no material equivalent yet.

And on the other side, understanding how hard it has been to build out euv lithography infra, and to completely retool manufacturing for a new paradigm of computing will require so much coordination, a super AI could have all the answers but we also have to deal with getting a bunch of stubborn people to align.

Some people are just coping yes, but if you really zoom into the details on both sides of the equations the gap between where we are and what it would take to dramatically surpass human intelligence is vast.

At the same time I think people overestimate how hard some kinds of knowledge work is (in absolute terms, since it's new and we haven't evolved into it as much), and how economically impactful human augmentation with weaker narrower AI can still be. (Lots of automation and prosperity but without such extreme disruption)

I think in the very long term all bets are off, but we will have plenty of time to evolve our economic systems before we get there.

[grttw1]

“I don't see why that's the case when you have super-human researchers on tap. ”

Hahahaa this is what AI psychosis looks like

[munk-a]

It's important not to miss the fact that AI productivity was a useful excuse for companies looking to conduct layoffs. Did some companies buy the hype? Sure - but the biggest companies would have wanted that sweet stock price layoff bump anyways and AI was a readily available justification to get it.

[AlexandrB]

> The only thing that matters is if LLMs with sufficient scaling can become frontier AI researchers kicking off the exponential.

What if the answer is flatly: no? All that other stuff starts to matter a lot then.

Predicating your business decisions on a potential breakthrough that may never come is frankly insane. Imagine if at the dawn of the car industry Ford decided that it's actually a race to build the first flying car and nothing else matters.

[Bukhmanizer]

Sure, but let’s not pretend that people treated the statements of these ceos as strategic messaging. People very clearly treated what Altman, Zuck, Amodei etc have been saying as predictions, and it hasn’t been until they’ve been proven wrong that people have started with the counter-narrative.

[taurath]

> LLMs with sufficient scaling can become frontier AI researchers kicking off the exponential

The cost is already outrunning the benefit to a massive amount, and the predicted expotential is not here yet. I predict it'll always be around the corner, a $1T model won't get there, but it will "look promising", but we'll sadly run out of money for the $10T or $100T model..

[openquery]

The only thing the $1T model needs to do is find some algorithmic speedup which allows it to be trained at $100B. I'm not saying that's easy or that it will happen but I just don't see why not.

[taurath]

I don't understand the logic here - a methodology to increase efficiency by 90% in training doesn't exist until it does - could you explain what you mean by "I don't see why it won't exist"? Are you seeing consistent gains by some process?

[dayvid]

We have two worlds:

1. Cutting edge LLMs developing ASI/AGI. 2. AIs doing general knowledge work

The second world will be achieved far before the first world is achieved. And as the first path gets develolped, the second path becomes cheaper and cheaper to run inference on along with being democratized which reduces the margins for the cutting edge companies. It seems like a mad dash to go as far as possible until 90% of general work can be automated with more cheaply available tech

[expedition32]

The leaders of these companies have no stake in it. They are beyond worldly concerns.

Napoleon got sent to Elba. Hitler ate a bullet. Your average tech CEO will still have more money than they can spend.