Hacker News new | past | comments | ask | show | jobs | submit
jxcole | on: I'm not convinced the AI bubble exists
If you don't talk about competition from DeepSeek you haven't really thought it through all the way. You might argue it's like Windows vs Linux, but operating systems have huge natural motes. Any person can just switch to a provider of an open weight model instantly. Most people don't know about it but we may very well see people learn about this broadly in the next few years. Changing models is not super complicated. Anthropic is unlikely to make $2 to the dollar when competitors are taking only pennies of profit.
mindwok | parent | next
The assumption here is that having "the best" model is not worth that much of a premium over a "good enough" model at a lower price. This remains to be seen, though, and people's behaviour paying for Claude Code and Codex suggests at least for now people are willing to pay.
zahlman | root | parent | next
> This remains to be seen, though, and people's behaviour paying for Claude Code and Codex suggests at least for now people are willing to pay.

I feel like this will depend on the user.

mrheosuper | root | parent | next
A lot of people use Opus more than Fable, so yeah, a "good enough model" definitely exist.
wyager | root | parent
That's just because opus 5.5 is arguably better than fable 5.1. I'm sure once a better model comes out, people will want to use it
mrheosuper | root | parent
Even before opus 5.5(which is opus 5), most people i know still used it more than Fable. Fable is kind of big gun that you use when opus fail.
etempleton | root | parent | next
It also assumes Anthropic can maintain the lead. If anything the competition both in the USA and in China seems to be getting closer and closer to parity with Claude.
QuercusMax | root | parent | next
I use Claude a lot at work and Deepseek for personal stuff, and they feel very comparable to me. I haven't found a task using Deepseek that makes me wish I had Claude - especially given how expensive Claude is.
simianwords | root | parent
This is important. I think people would pay a large premium for a marginal increase in intelligence. Primarily because intelligence is correlated with exponential utility which is often ignored.

For example, I'd much rather pay for the highest intelligence model when designing something or developing something because the problems it can preempt will save me more money in the long run.

The whole issue is that intelligence (as measured in AA benchmarks for example) is not linear utility

zahlman | parent | next
> Changing models is not super complicated.

Don't the harnesses try to lock you in?

matteoraso | parent | next
>Any person can just switch to a provider of an open weight model instantly. Most people don't know about it but we may very well see people learn about this broadly in the next few years.

I mention this in the article...

TacticalCoder | parent
> You might argue it's like Windows vs Linux ...

Sorry what? There are 10x if not more Linux installation vs Windows installation in the world. Linux powers everything, including billions of smartphones, where Windows is nowhere to be seen. Same for routers. And actual proper servers.

If it's like Windows vs Linux it means open-weight models are going to be incredibly more successful and touch just about everything. All the while being cheaper and working not a bit but much better.