You do have to worry about scale with lambda though. It’s not super hard to hit a rate limit where warm up times start being a real problem.

I can tell it’s getting bad because I went to close some Safari tabs and got confused because the tab next to this was [a dupe of] https://news.ycombinator.com/item?id=49330597.

You've a tab from 9 days ago open? Maybe you should be closing your tabs more often :p

I have tabs from like 6 months ago. Modern browsers have gotten pretty good at hibernating unused tabs and restore when needed. And this website is extremely light, it gets restored in milliseconds.

I hording my tab in case I need them later (I don't, 99% of the case)

but I always have sense of fear that I might need it someday

Every 3-6 months I purge all my tabs, but always dump all URLs to an html file in case I will need them in the future. You never know!

I do that too but instead of html, I store them on discord and obsidian text file (double back up)

but I still keep it on my browser, maybe someday I would like to fetch LLM and describe my interest over time

That is just called your browser history?

The browser history contains a lot of junk and I don't want that saved forever. Curated tab history is a bit different. It's stupid for sure, but also a bit comforting to keep the most important articles and such. For anything I really care about, there's always ArchiveBox.

Bookmark?

Yes but why

It is certainly the case that isinstance(True, int) returns True, even today.

I got a bug for not remembering it, a couple of years ago: https://jpscaletti.com/p/8/true-false-one-and-zero

Hmm. Having read your post, surely the bug is having a function where

    set(foo, false)

removes foo entirely. What if you want foo to have the value False? Even besides the unintended behaviour where 0 is coerced to a boolean value, this function seems poorly designed.

I don't remember the specifics, it might have been a simplification for the example

Now imagine a model shipped in silicon that gives 5,000 tokens per second at Opus 4.8 quality

Others have addressed the validity of your comment, but I’d like to discuss another aspect. Even if it is filter, I think most people would not want to filter. Apple makes it very easy to use private relay and many people that buy Apple products do use private relay. Even if you know, it’s a private relay email address, surely you want a user who buys expensive technology products to use your website in almost all circumstances.

Suspiciously ignored is power consumption. That’s the biggest and most useful metric often ignored because it’s less flashy than speed.

I'm also interested in the security features of this new CPU. Where doesn't it sit compared to Apple and ARM (Qualcomm, Titan)?

Modern processor without memory tagging for example is a little..... dated, given how much the new AIs accelerated the vulnerability discovery in all the apps and system components.

I have heard it said that security is the new power when it comes to microarchitecture. I feel like that’s something security enjoyers tell themselves to be happy.

I don't understand your comment - did you want to say that hardware security is overrated or that security and microarchitecture don't mix?

Or maybe that security doesn't have a practical application?

Or maybe just a joke, that most of the world does not care so much about security, unlike security minded people would like to?

Good question.

I’m sure Apple’s chips could do better in single core if they were willing to push the power envelope way higher too.

Multi-core is easy to beat, just add more cores until you win. There were already chips that beat the M series just by having way more threads/cores.

Apple’s devices generally have no or very underpowered thermal solutions; but it also partly regulates battery life.

(Seriously, they rely on air cooling for Neo and MBA — these chips hit 105deg easily and throttle, intentionally)

Adding a thermal pad to a Neo or MacBook Air gives significant boosts peak and sustained performance; whether in benchmarks or real world video renders; but it also reduces your battery life by a little, as there’s less thermal throttling forcing wattage down.

And no, the battery is not connected to the chassis (heatsink), it doesn’t have increased temps.

And arguably the most important when it comes to mobile and small form factor usecases.

Apple doesn't optimize for raw performance but rather perf/watt. So comparing only performance is not really useful.

IMO bot that important practically - SiC batteries, active cooling, bypass charging etc - if one really cares about sustained performance (gaming etc), PRC has phones segments big on mitigating thermal throttling. Exclude benchmark drama, golden samples, retail units still vastly outperform passively cooled phones with "better" silicon on long sessions. Otherwise there aren't many usage case where power consumption matters between that can't be trivially made up by slightly better or larger battery and/or passive cooling.

At some point the whims of the CEOs don’t win. The shareholders demand more growth.

Demand from whom, if not the CEO?

The board, presumably

Of course the demand will be directed at the CEO and they will have to provide results or be replaced.

Yes. Let’s say I saw a person kill someone. That’s a fact, but my perception of them would be low; that is an opinion.

I think it’s really swell how the last 10 years have made it so the basic difference between fact and opinion is not totally obvious.

We're talking stated "risk" to the public.

Using your example --- as far as "risk" to the public is concerned, which would be more pertinent info --- the fact that this person is a murderer or your low opinion?

Still, that characterization falls under negligent or reckless depending on if the person knew or should have known the actual danger. It is different than intent.

I think we’re gonna see this criticism become less than less valid over the coming decades, as 1989 becomes more and more “history” and less “present situation”