[johnQdeveloper]

They already do both.

The real competition is coming out of China right now and I doubt the Chinese government is going to let them buy out their "fast follower" AI companies that are consistently 6-12 months behind in terms of quality. That said, I'm factoring quality as in Opus 4.5/Sonnet 4.5/GPT-5.5 as break points since I haven't really seen an improvement since that point when using AI.

[satvikpendem]

They'll just lobby to ban Chinese models as they're already doing.

[Robdel12]

> Compounding the problem, labs in China often release dual-use capable models as open-weight. Once a model is open-weight, safeguards that do exist can be removed, making the model available to any state or non-state actor to use for malicious purposes, including the cyber and CBRN misuse those safeguards were built to prevent.

https://www.anthropic.com/research/2028-ai-leadership

They are already starting that now.

[potsandpans]

I have to hope they won't succeed. Maybe for a short time, but eventually open weights will prevail.

[johnQdeveloper]

Probably but the reality is I doubt that actually works outside of US government contracts in practice simply because Europe/et al aren't going to follow their lead.

[WarmWash]

Europe has been slowly committing economic suicide for the last 30 years by outsourcing everything to the US and China (and Russia), and it looks like maybe Europeans are finally starting to wake up to this.

I wouldn't be one bit surprised if a rash of digital sovereignty movements in the near future hamper Chinese model adoption.

[ponector]

And yet average European citizen have a longer life with higher quality of living than US.

[WarmWash]

Burning furniture will in fact keep you warm, and if you can do it long enough, your kids will wonder why others toil with trees and axes so much.

The downside of course, is that it is much much harder to go back to chopping wood once that furniture is all gone. Especially if all you ever knew was burning furniture.

[satvikpendem]

For now. Without industry of their own Europe won't have a high quality of life much longer, as recent unemployment statistics already show.

[JacobAsmuth]

Opus 4.8 is a significant coding quality gain over 4.5 - I'm not sure there's any third party testing which indicates otherwise.

[johnQdeveloper]

My experience isn't consistent with it being significant (or really any) quality gains on actual real world usage for me or the team I'm on.

[JacobAsmuth]

The plural of anecdote is not data. What are your evals telling you?

[johnQdeveloper]

The % of accepted, actionable prompts is not up if I use Opus 4.7/4.6/4.8 if that is what you are asking.

[fieldcny]

You speak so authoritatively about quality and performance of these models, yet there are no quantitative metrics that correlate to real world outcomes that indicate that the quality and performance of these models is anything but subjective noise and classic benchmark nonsense.

A company consumed half a billion dollars worth of tokens in a month and nobody noticed anything until the bill came due.

Tha $500m dollars is roughly equivalent to 2000 people working for a year or 500 people working for four years, they can and would accomplish a lot if they worked in companies that add value to the economy by solving real problems.

[4ffdd]

Indeed Its irrelevant. Each firm will make its own cost-benefit analysis, especially since the frontier labs are raising prices.

Marketing only takes you so far in creating noise.

Its weird seeing this focus on bench marks again - PC's did this for quite some time. But in the end it came down to - what does all this additional horsepower let you do? Oh create interesting apps, multi-tasking etc. Which was really the value-add.

[johnQdeveloper]

> You speak so authoritatively about quality and performance of these models, yet there are no quantitative metrics that correlate to real world outcomes that indicate that the quality and performance of these models is anything but subjective noise and classic benchmark nonsense.

I'm responsible for AI roll out at a small business and we've had data science go over these things internally in terms of what results we get for 12+ months now. Its just my experience that is roughly the results we've seen using Deepseek, etc. and comparing cost/results vs. Anthropic/ChatGPT.

> A company consumed half a billion dollars worth of tokens in a month and nobody noticed anything until the bill came due.

It was sourced from one anonymous source. Its highly unlikely to be true in my view, but hey, you do you.