Wouldn’t that comparison be quite a bit closer vs the 17 Pro? I thought that was a decent camera upgrade year but I can’t keep track.

Generationally yes, the oppo is more recent, but (AFAIR, I'm speaking from memory) the 17 improved just on the selfie and accesory lenses, the main camera remained mostly untouched.

I would argue it was moving towards a more Apple-like distribution model that has hurt them so much.

The 5G standard is a bit of a mess. In the US we heavily rolled out 5G NSA, which means you still have a 4G network core with both 4G and 5G radios at the edge. You get a chunk of the performance benefits without having to replace the network core first.

This usually has similar (in practice) performance characteristics to LTE Advanced.

5G SA is an actual, end to end 5G network. The two biggest wins here for an end user are reduced latency (a big focus in 5G) and reduced power usage (the phone doesn’t need a simultaneous 4G connection).

Yeah. In india jio was SA from the start. Airtel began NSA then I think now they have mostly moved to SA.

Hah, I was aware of 5G vs 5G mm-wave, didn't realize there was another way in which most 5G was "kinda fake" too.

26ghz mm wave was a bungle as well. See https://news.ycombinator.com/item?id=49675724

Is the speech to text happening on the watch and then the transcription is getting summarized on the phone? If so, the argument may not be as clear cut IMO.

The speech to text is on the phone and the summary is on Apple's cloud.

Replying to you again because I would like to know what you think of the evidence presented above? Maybe download the data and try and run it yourself? What info do you need to see that the temperature is changing?

Free subscription + a media education module for their teachers. Without making them read it, they showed high usage of the free subscription.

Yeah, I got that from the abstract too. I just don't understand where the “making teenagers read newspapers” angle comes from.

There's a more subtle form of 'make' where you set conditions to induce a result. The fact is, giving anything away for free will increase its consumption.

Yes it will. But that doesn't mean that “making students read” doesn't arouse a very different impression. “Making students do homework.” does not imply “Making it free for them to do it.”.

Like just having a broom in the house makes my kids sweep the floors? There's no way to interpret the headline that makes any sense as truth.

good point about the title.

> USHCN

This was the network that had heat island issues because it was volunteer stations highly biased towards urban areas.

It was superseded by USCRN (which you linked to) in 2005. USCRN is 114 stations intentionally situated away away from development. USCRN shows MORE warming than global networks, around .3* C per decade, and it recorded 2024 as the warmest year on record.

The article you are indirectly citing (https://dailysceptic.org/2023/04/08/u-s-government-has-an-ac...) is easily disproven if you analyze the numbers: https://www.theclimatebrink.com/p/the-most-accurate-record-o...

What they all deliver is benchmaxxed models that are awful in actual use.

But it gives the lab good publicity, as seen in the article.

I'm giving out more unpopular opinions today for entertainment and thought :)

Every model is benchmaxxing today. Many of them are awful until tuned for any particular task.

Qwen is a really awesome model to use!

Everyone warns about folks jumping to personify AI but there is probably just as big of a risk of AI-ifying people.

Not necessarily a "risk".

Some people treat their agents better than they do their colleagues. Sometimes the colleagues have it coming, sometimes not. Sometimes it's one's own fault. Agentic interactions can help you orient yourself through what's what. Lots of posts about engineers getting their first taste of project mgmt through this recently, for example.

Or take customer service. Dehumanizing people is generally bad, but sometimes it's downright appropriate. You don't want to get bogged down by some rando throwing a tantrum, that's in no one's interest.

Two things:

- check your temp settings vs. Qwen's recommendations, they specify what it should be in the model card for thinking on, and that reduces some "over" think.

- the model appears to be intentionally designed to do a lot more test-time compute, if you anthropomorphize the tokens, it looks like overthinking and anxiety, but it is just spending compute to get to the end result, so it may not actually help it to prompt down the token spend (depending on the problem)