maeil · HackerTrans

anno scorsodiscuss

Sonnet 3.7 non-reasoning is better on its own. In fact even Sonnet 3.5-v2 is, and that was released 6 months ago. Now to be fair, they're close enough that there will be usecases - especially non-coding - where 4.1 beats it consistently. Also, 4.1 is quite a lot cheaper and faster. Still, OpenAI is clearly behind.

maeil

anno scorsodiscuss

It isn't.

maeil

anno scorsodiscuss

Fringe but incredibly similar to Typescript. I don't think there exist any other language that is not a direct dialect of Javascript that's as similar. I didn't find there to be any language barrier, especially now with LLMs who are perfect for answering one-off questions about the few diffefences. The tooling is better as well, which means even less of a barrier.

maeil

2 anni fadiscuss

Not sure if the irony is intended here. The entire point is that the Chinese people aren't a monolith, hence CCP != The Chinese people.

This will also hold for whether they believe us - in that too, Chinese people won't be a monolith. Plenty of those who aren't the biggest fans of the CCP will, as they understand where we're coming from better than anyone.

maeil

2 anni fadiscuss

Surely on HN of all places we're aware that the CCP for decades now has been as communist as the Democratic People's Republic of Korea has been democratic?

You're looking for "authoritarian" or "dictatorial".

maeil

2 anni fadiscuss

> requires a multi-faceted approach

Proof enough that this has been trained directly on GPT input/output pairs.

maeil

2 anni fadiscuss

"If you're China" clearly refers to the government/party, assuming otherwise isn't good faith.

maeil

2 anni fadiscuss

What I'm getting is that Google did push too hard, they gave Bard/early Gemini demos when the product was still miles from ready and compared awfully to GPT, and they're still suffering from it.

maeil

2 anni fadiscuss

The frontend of the B2C product, on desktop, sure. I was talking about the LLM itself, i.e. API-based usage. And there's a good number of tasks it's better at than any other model. Same goes for GPT and Claude FWIW, each have tasks they beat the competition on.

Even on the consumer side though, the overwhelming majority of usage will be through mobile apps, and I find the Gemini app significantly better than e.g. the Claude app - one of their two main competitors.

maeil

2 anni fadiscuss

> One thing I don't really like about Satya Nadella by the way, his AI strategy is a mess. They're slamming stuff on the market way too quickly (see the recall fiasco)

Google did an even worse job in that regard, Gemini's reputation is still greatly tainted in the tech world despite now being competitive with the other frontier models.

maeil

2 anni fadiscuss

Any recommended resources to learn about this to those starting out?

maeil

2 anni fadiscuss

FWIW here in Korea the 1940 statistic still holds, yet I don't think youth mental health is any better.

maeil

3 anni fadiscuss

Not sure how this is a compromise by any means as it still means abiding by some arbiter to decide who gets to be a country and who doesn't. Outsourcing the arbitration to the ISO doesn't make much of a difference.

comments