At least from an interview with Amanda, key philosopher at Anthropic: https://www.youtube.com/watch?v=I9aGC6Ui3eE&t=1912s

why though? I doubt subscribers are moving the need for ARR, even for OpenAI

I do think this question comes up a lot-- I can understand why.

For some well-explained reasons, check out https://darioamodei.com/essay/the-adolescence-of-technology and search for "CCP".

As somebody from neither the US nor China, this argument would be much stronger if the US hadn't started acting like a rogue state - starting wars of aggression and messing up the world's energy supply, actively speeding up climate change, kidnapping leaders of sovereign nations, threatening its allies (!) with invasion, etc etc.

The CCP's not great either, sure. But the Americans don't really have a leg to stand on anymore.

To be fair the US has been doing this its entire history, it just stopped pretending. C.f. Hawaii, Guatemala, Cuba, Chile, "The Jakarta Method", etc. etc. etc.

As far as AI is concerned, it looks like you will need to pick 1?

Then I pick the country that hasn't been bombing people and starting wars nonstop over the past 40 years.

Hint: It's not the U.S.

So is your concern more about reducing the risk of an authoritarian China "winning" the AI race? And less about reducing the risk of your data being used against you personally?

To me, the risks of an individual helping China to continue to develop their AI by being a customer is pretty marginal compared with the personal risks of my data being used against me.

I see. Though if you follow the argument set forth by Dario, it seems you'll not only have your concern to worry about (i.e., personal risks of your data used against you), but many more as well on top of that.

A blog by Dario of all people. Totally not bias towards non-US models.

90%+ cache hit rate is common, and so you'll see on places like openrouter that Deepseek cache cost is indeed a magnitude cheaper than the rest.

My usage thus far from api.deepseek.com

  - input_cache_hit_tokens: 1,265,646,976 x 0.0000000028 = $3.5438115328
  - input_cache_miss_tokens: 18,208,088 x 0.00000014 = $2.54913232
  - output_tokens: 9,615,178 x 0.00000028 = $2.69224984
  - request_count: 10,837 (no price)

Total cost: $8.7851936928 (approximately $8.79)

Cache:

  - Hit: 1,265,646,976
  - Miss: 18,208,088
  - Total input tokens: 1,283,855,064

Hit rate: 98.582% (1,265,646,976 / 1,283,855,064)

So your cost would be $40.68 with another provider that has one less zero in the cache hit price.

Agreed on hoping books will be rewritten. It seems to be happening, but slowly. E.g., Harvard Machine Learning Systems: https://mlsysbook.ai/vol1/

Thanks, missed this and will try to remove

"The apps are open source, and Hickisch doesn't expect to make money from them anytime soon. "

I'd like to try out the settlers-style game, but can't find them? Nothing here, either: https://github.com/raffael-hickisch

The game is here: https://siedler-oesterreich.exe.xyz/

Thank you! Not sure why it's so hard to find (still don't see where you got this from), esp if it's using exe.dev's platform.

(exe.dev co-founder here) The platform is designed make it easy for you to share if you like, but we are not typically in the business of pointing directly at products users make. Maybe we should change that! It would be fun to have a "best of exe.dev" on our site.

Ask creators before featuring them, but medium and substack and YouTube are popular platforms because of their discovery mechanisms, not because hosting text or video is especially hard these days.

Sorry for this being hard to find. I've put the links to the live apps and repositories here https://kohlschwarz.at/

I wonder if .exe in the domain sets of all kind of overenthusiastic "safety" mechanisms in search engines, AV software, etc.

Interestingly, it's the .xyz on our VM domain that causes problems. It's apparently far more of a problematic gTLD than I realized. E.g. it appears Southwest Airlines blocks the entire thing, as do some enterprise networks are working with.

(We have a fix for this coming in a few weeks.)

I'm also curious, as this could pay for a trip out there, especially if buying for friends.

Anybody have experience with https://github.com/chopratejas/headroom? They seem to have similar goals in token reduction, but headroom appears to be broader in scope.

They provide benchmarks as well. I like headroom’s approch a lot better. More transparent.