Hacker News new | past | comments | ask | show | jobs | submit
Models are popping from left and right. Is that couple months of lead trillion dollars more valuable?
They sell trust, and they sell "the best one".

The specific models and capabilities are a sideshow.

And hence, the money flows all to the top despite viable alternatives...

(a separate market of workloads is contracted out to Fireworks, Baseten et al. and a third segment DIY it with rented or owned GPUs. But those customers are different buyer archetypes little overlap with those buying trust.)

Well then they have failed miserably. They cannot be trusted, with all that anti-user behavior, fearmongering and all that stuff.
Why trust them though? They're a cult.
Models are useless if you don't have the infrastructure deals to run them at scale. I know people that use cheap Chinese models and get good results from them. But they are paying a lot for self hosting those models. The models are cheap but the infrastructure isn't.

The reason I think Anthropic's position isn't that strong is that most of the money they raised will go into making other companies very rich. Because they don't actually own a lot of the infrastructure they need. OpenAI has similar issues. The companies that actually provide the data centers (i.e. hyperscalers), chips, gas turbines, etc. Will be on the receiving end. And of course some of those hyperscalers have their own ambitions on tapping into the same markets. The SpaceX pivot to data centers in recent years is particularly interesting. Anthropic is already paying them a lot.

The reason I'm fairly optimistic overall about the economy is all the spending that will result from these investments in all that physical stuff represents an enormous cash injection in the economy. The investment bubble may or may not burst but it won't wipe out the infrastructure. If Anthropic collapses, that infrastructure will just be gobbled up by whomever takes over.

I wonder if China will get their head out of their asses and start manufacturing chips to construct electronics and infrastructure to run AI.

Oh, you mean they have already started on it??

They also has the most energy production. They will go past anyone else at this rate.
Are you willing to bet they can't stay ahead?
For actual science done by scientists they're far behind open models now.

The people I know using this model are perpetually hamstrung by slowness, pedantry, misplaced assumptions and appeals to authority. You can't move forward if you're just trying to norm your training data.

They've created this extremely slow, overwrought system with lots of guardrails and token limits designed only to create addiction. Why would you use that when you can have a marginally less "smart" system that's 50x faster and 100x cheaper? Who cares if the model one shots something. It needs to be checked anyway, by test in a machine or a lab.

What models are scientists using that they gain so much more marginal value from than Claude-class models?

Who are the pedantic, authority-appealing users who are using this model?

What makes you believe they’re norming their training data to that group?

Cloud models have zero value if the findings of scientist are absorbed by these same cloud models and "by coincidence" then published before them. It is impossible to get ZDR as individual that you can trust.
Statistically how many companies that were first to any market the eventual dominating leader of that industry? To me it would still be like picking a single number off a roulette wheel because it is a very slightly larger slot to bet on. Maybe it will pay off, it might even be statistically the best choice, but it is still a slim chance overall that it will be the winner.
I am willing to bet 1:1 that a model from any other lab will beat Fable 5.2 within 3 months in any metric of your choice.

I am also willing to bet for their next frontier model that a model will beat it within 3 months of its release.