why_only_15

Over and over again they talk about how Anthropic's margins are getting squeezed. They're so squeezed in fact that Anthropic has merely 85% gross margins! They're earning merely $6 in revenue for every $1 they spend!

Why do you think the Chinese labs figured this out before the western labs? No reason to believe that whatsoever.

yeah there's some KV cache offload

This is just wrong unless you're using some confusing definition. Notice that companies trying to do lots of inference aren't looking for CPUs, they're looking for GPUs/TPUs.

CPU DRAM can't really be used for inference efficiently -- inference mostly wants memory bandwidth, not memory capacity, and GPU DRAM has >10x more bandwidth. The fabs can switch between them but you can't switch after the fact.

They have various benchmarks, e.g. how much time it takes them to do wikipedia page -> page games. Jev seems to take the same or fewer hops but in ~10x less time and for ~10x less money.

It's totally reasonable to compare against LLMs doing chain of thought if it gets comparable performance.

To what degree can you say something like "if the kernel doesn't have metaprogramming [or some other set of features] it's fine". When were the last bugs with a reduced feature set?

The whole point is that these are security bugs that didn't necessarily bother people so far but could be exploited

I really don't like this. Don't post your AI-generated articles here. https://www.pangram.com/history/fce67fb4-4f6a-4eb4-a323-5048...

The correct comparison is not sonnet, but qwen3.5-27b on a cloud. Alibaba's pricing [0] is $0.20/m input $1.56/m output, so $0.72 for the prompt processing or $0.22 for the generation. Yours is still cheaper but the margins are less.

My guess is that this math gets less good with MoE (because you will be limited by VRAM, but clouds won't).

[0]: https://openrouter.ai/qwen/qwen3.5-27b

Why is the post AI-generated? If you're going to make something for us to read you pay us the courtesy of actually writing it first.

https://www.pangram.com/history/61fbc90e-180d-4d91-a85d-16ff...

The post itself is totally AI-generated. It has tons of tells, in addition to Pangram saying so.

https://www.pangram.com/history/61fbc90e-180d-4d91-a85d-16ff...

why do you think so? they provide some evidence of this in the article, but there have been several improvements in e.g. nanogpt-speedrun or openai parameter golf made by AIs

Does China subsidize housing construction? Is your claim something like "China encourages/forces consumers to save at high rates which lowers interest rates and makes there be more construction"?

what's the origin of your $20k/vuln estimate?

The formal Ottoman name was Kostantiniyye=Constantinople until the empire's fall in 1922. The official shift happened in 1930, with the Turkish Postal Services Law changing the name to Istanbul.

This is a quite old technique. The idea, as I understood it, was that lots of data at Google was stored in triplicate for reliability purposes. Instead of fetching one, you fetched all three and then took the one that arrived first. Then you sent UDP packets cancelling the other two. For something like search where you're issuing hundreds of requests that have to resolve in a few hundred milliseconds, this substantially cut down on tail latency.

I am in this world, but am not familiar with this specifically.

My guess is that they found a bug with their implementation of the model using the weights Google released. These bugs are often difficult to track down because the only indication is that the model is worse with your implementation than with someone else's.

The point is that if DoD's supply chain restriction does what Hegseth seems to want, all contractors involved with Anthropic would have to divest. That includes Amazon and Google, who are both DoD contractors who provide massive quantities of capital and compute to Anthropic. It's irrelevant that Anthropic provides Claude through Palantir.

This is not really possible. My guess is that the government is not willing to spend the necessary quantity of money to get e.g. Amazon or Google to divest of Anthropic and stop providing them computing resources.

The point of the supply chain risk designation was not just to have the DoD stop using Anthropic (they could have done that by just cancelling the contract). Their intended effect was to force every company that sells to the US government, no matter how indirectly, to not use Anthropic in any way, which would effectively destroy them because almost every company is in the supply chain (for example my company is https://calaveras.ai/ because we sell to AI companies who in turn sell to DoD).

if your bucket name is ever exposed and you later delete it, then this doesn't help you.

Why did you post an AI generated article (against HN rules)? https://www.pangram.com/history/0943da35-d51a-4d81-8207-fae7...

This seems like a really poorly thought out article. You should take more care on making sure your understanding is correct before publishing in the future.

Taking the Amazon example in Part 2:

For e-books (simpler), Amazon gets 30% for running the store, doing advertising, etc. and then authors get 70% [1].

For print books, I'm a little less clear but it appears Amazon buys the books for roughly 50% of list[2] which for Hachette in 2025 is $26.50 so Amazon pays $13.25 to the publisher and then Amazon retails the book for $14.84. So for $100 of books sold on Amazon, $89 goes to the publisher and $11 goes to Amazon. It appears that the cost to produce these books is maybe $2/book (though I'm very unsure on this, this is a guesstimate from public data) and then the rest flows back to authors, advances, etc.

Amazon.com (not AWS) has a 7% profit margin in North America (FY25), so of that $11 they get in revenue they get $0.77 in operating profit.

Ok and this also annoyed me: you say $1.7T/y is $10.5k/worker, which is accurate. but then you say for the average household it's $26k/y. This is not true. There are 134m households in the US [3] so it's $12.6k/y for the average household. Maybe you meant something else like the median household but it seems more likely you just said ~2.6 people/household and multiplied the number of people/household by cost/worker. This is obviously wrong and you should have caught errors like that earlier.

[1]: https://kdp.amazon.com/en_US/help/topic/G200644210 [2]: https://www.readersfirst.org/publisher-price-watch [3]: https://fred.stlouisfed.org/series/TTLHH

Very cool stuff! Thank you for making.

Pangram labs thinks this post is fully AI generated, for what it's worth. https://www.pangram.com/history/47460dd9-f9cc-45a8-81dd-dc59...

I think Cosmo's refutations were mostly not very useful and based on misunderstandings of what I was trying to say. This is fine and we discussed it prior to their article being published.

The point I was trying to make with "RL is only necessary once" is that you can embark on a single self-play loop getting better and better, and this will get you to something close to the frontier. Once you're at the frontier, the frontier doesn't move very much, so you have quite a while (decade?) where it's totally fine to distill from the RL games.

On correction histories -- imo I correctly described what they do. Cosmo was annoyed by the word "adapt" but what I described was the adaptation.

On SPSA -- you don't have a gradient! you don't do backprop! this is what i was trying to get at.

The chess people seemed to think my article was reasonably accurate. But I'm not really sure.