cbg0

Back in the day people used to upload new movies on public torrent sites that turned out to be something completely different in order to troll people. This is probably the same.

If not, this is one hell of a way to save their IPO.

Or it could be that the benchmaxxed models are getting pushed down by the introduction of unsaturated benchmarks. Muse Wonder also went down 52 -> 43 after the update.

It does seem like a very uncompetitive approach.

Never thought I'd meet someone who's eyes can't see "more than X Hz anyways".

If they had a model on par with Opus 5.5 there would be no need to undercut it by that much, that's probably the logic.

The "portal" layout style was very popular on Web 1.0, reminds me of PHP-Nuke.

What are you on about with niches? Most people can't afford a brand new EV car and have nowhere to charge it.

You haven't really answered my question. Germany shut down its last nuclear power plant in 2023, this guy you linked got appointed in December 2024.

You want to destroy the automotive industry in Europe so we can increase dependance on Chinese EVs?

"simply ban combustion engines and gas boilers" - there's nothing simple about this

How is the EU blocking nuclear power? Also the tariffs are on Chinese EVs, not those built in the EU.

How would you have avoided that "whole situation"?

> Agents doesn't cause new problems, they reveal existing weaknesses.

Agents absolutely can and will cause (new) problems if they are unsupervised even with guard rails in place.

Bold of you to assume they have many defenses left.

Claims to be on par with GLM 5.3 in DeepSWE (from https://thenextweb.com/news/mistral-releases-large-4-a-1-tri...)

I use AI on a daily basis and I still found all of your takes in this thread to be silly hyperbole.

You'll be surprised (or not) to find out a lot of people aren't using anything. Claude gets asked to build it, then some other clown uses their copy of Claude to do the PR review and then it gets merged.

> regardless of what you think about coding is done, agents and your harness is the best tool.

I guess you're all in on that Anthropic IPO with the "coding is solved" approach.

There are AI players already inside Follower Dungeons. (not LLM-powered though)

France isn't "broke", its debt-to-GDP ratio is actually lower than the US right now and it's not in a situation where it can't pay its loans.

There is currently no real trade war EU-China; Chinese EVs are streaming into the EU while the US is blocking them completely for one example.

While Germany and more specifically Volkswagen is having some issues, EU industrial production actually had growth last year.

The $100 billion aid packages to Ukraine are also a bit overblown based on public figures https://www.kielinstitut.de/media/news/ukraine-support-track....

I feel like your comment pulls on some real threads, then blows things out of proportion.

> The US Government is not a major funder of frontier labs (who are almost exclusively VC and private money funded)...

There isn't a lot of VC funding available in Europe compared to the US, so without government intervention catching up will be very tricky.

Assuming you are right, the "build-up" of more physical datacenters isn't really the difficult part, training frontier models is. The open-weight Chinese models are several months behind frontier models at this stage and as long as you're not interested in burning cash to stay in the lead you can just wait it out. Many signs (upcoming ridiculous IPO figures and benchmarks) point to the current LLM paradigm reaching a plateau.

I think that currently the EU is focusing on strengthening itself from a military perspective due to US policy shifts, so the EU just doesn't have the funds to divert to AI without impacting day to day life of people, thus prioritizing safety and perhaps energy production, as electricity is considerably more expensive here than in the US or China. If you ask the people who vote, they'll put safety, grocery and gas prices way above EU's LLM capabilities.

You've posted a link that doesn't support your statement.

Here's the front page from 12 years ago, not exactly riveting debates in top hats.

https://web.archive.org/web/20140929012305/http://www.reddit...

https://web.archive.org/web/20140929073818/http://www.reddit...

https://web.archive.org/web/20140929232807/http://www.reddit...

This has always been Reddit, aside from the niche subreddits.

> This is YET AGAIN another Sonnet model that is just a FAR worse version of Opus at every part of the cost AND speed curve.

It's been out for an hour and you've already concluded this?

You don't have to run Claude/Codex in auto-approve mode, you can manually approve its interactions with your machine without having to copy your code back and forth between the website and your local files.

Trying to appeal to the LLM is pointless, everyone would just claim that they're working on their own code.

I've found that if you're seeing refusals and you have a more permissive model available, you can use that to find the issues and pass them back to Claude for review and implementation.

This is some main character energy right here.