striking

Yeah. It's "engineering".

Wikipedia's introductory paragraph to the subject includes the following sentence:

> Engineering involves balancing competing demands such as safety, performance, aesthetics, cost, laws and regulations, and time, while operating within fundamental limits such as the laws of physics.

(from https://en.wikipedia.org/wiki/Engineering)

I realize the word is overloaded now and that people are confusing writing programs and other cerebral pursuits with engineering, but that doesn't change its original definition.

I think the fine article says it best when it says

> After doing a bit of research, we decided that the most common real world use cases of zero-knowledge proofs (age verification, crypto, etc) aren’t particularly interesting to us. We enjoyed the graphs and theory of computation and networked computing though. We hope you had fun playing around with interactive proofs too.

The inclusion of this statement makes your comments look rather bizarre and strangely combative by comparison. Maybe we can just enjoy a bit of math together?

Is the table not a per-metric comparison, though? And if it is a result of finetuning, isn't that by definition not something inherent to how LLMs are?

As to the latter bit, it just seemed a weird framing to me.

If the model were just repeating the distribution baked into the training data, could you help me understand why the R^2 values in Table 3 are mostly below 0.1?

Could you explain what you meant when you wrote "the women who published this"?

I do think code review is worthwhile, not sure how you read that from my comment.

A cap on PR size isn't inherently going to make an LLM do a good job of segmenting PRs. It requires careful prompting or manual action, the kind of effort typically exerted by people who already cared enough not to hit such a cap. You may as well just ditch the cap, to save yourself from having to reject a series of PRs rather than just the one.

Yeah, that's a fun way to get massive stacks of PRs that are individually incomprehensible.

Fine. But walking is also just kind of nice and unobtrusive for most people. I'd understand if you were replying to someone telling you to do strength training, but I'd argue walking is just pleasant.

I don't see him as blaming us regular people. Just liars (which he takes pains to enumerate thoroughly). Of course most of us are just living our lives. But sometimes even us normal folk have an opportunity, when we bump into one of these liars or have a conversation about one, and we can at least remind folks this is still a problem.

The main thrust of the article is that the attention to the uncomfortable issue of climate change has been replaced with noise about everything else and comfortable lies about things being fine, and that the drop in attention correlates strongly with climate change worsening as such making the case that even just bringing more attention back to the issue could help prevent it.

Exposure Notifications is no longer available in Android as of like 2023. https://developers.google.com/android/exposure-notifications

If that worries you, you'll be aghast at the fact that Ed's actually been on Bloomberg TV a couple times now.

If it weren't for all the supermassive cars rolling around, purchased because they're ostensibly safer in a crash for the person driving them (while markedly decreasing the safety of the other driver) then maybe I could agree with you. But with so many people intentionally electing to value their individual safety at cost of the safety of others, I actually think that it makes for a great reason for them to not think about road safety! Who cares if the road is unsafe if I'm driving a tank?

He was lucky to be managing any money at all, the West Coast smiling upon him did not translate into East Coast excitement:

> When this star of San Francisco arrived in New York during his fund-raising tour around last summer, however, he received a relatively cool reception, according to three people from whom he tried to raise money, who declined to be identified talking about a private fund. They said they viewed him as a lightweight and a one-hit wonder. The asset management colossus Blackstone, the world’s largest investor in hedge funds, passed on investing, according to two of those people.

> One wealthy New York investor who did take the meeting welcomed Mr. Aschenbrenner into his downtown office, and then gave him a grilling, according to the investor, who declined to be named publicly because he had agreed to keep the contents of the fund’s pitch private.

> What was Mr. Aschenbrenner’s plan if the A.I. revolution didn’t pan out quite as hoped? The hedge-fund founder had no detailed response, the investor recalled. Mr. Aschenbrenner simply truly believed it would all work out.

(as quoted in https://www.bloomberg.com/opinion/newsletters/2026-08-03/hed..., originally from https://www.nytimes.com/2026/07/31/business/situational-awar...)

Asking politely is not how we got a 40-hour work week or workers' comp or most other labor standards we take for granted, just historically speaking. I think a positive outcome is very likely, and I think it will be a lot more work than loudly yelling, but I don't think anything will happen if we try to build everything up from first principles instead of taking a moment to consult history as many are wont to do in this AI era.

I think this too is a kind of denial. As in, while some are in denial about the usefulness of AI, others are in denial about whose living standards are actually going to be uplifted.

And it's sad, really, because I think these two groups would make a great pairing if they could stop arguing against one another for a moment. They'll both be impacted about as much and probably have the same ultimate goals (to lead dignified lives).

But it seems these days everyone is more interested in Kayfabe and feeling like they're in the right than working together, so maybe I should just keep quiet rather than attract the ire of both groups...

95% of farms are "family owned". Large scale family owned farms make up about 4% of all farms but 48% of agricultural output. https://www.nass.usda.gov/Publications/Highlights/2025/Censu...

There's a tendency to want to be nice to farmers who own "family owned farms" because it evokes imagery steeped in quaint classical Americana. It is mostly from misinformation that this impression arises. Don't feel bad for a family owned farm that isn't going well. They get special bankruptcy rules, they can sell off their very valuable land as a backstop, and most of all they are probably not as small and quaint and cute as you think they are.

From a look at https://en.wikipedia.org/wiki/BBC_Micro#Specifications I think the 6502 versions of the beeb didn't have banked RAM so to keep it loadable from tape the limits might be as stated.

But with substantial additional effort, maybe some banked ROMs could be added..?

Second chart (it's to the right of the first chart and clicking it to scroll is easier than otherwise)

He does have some choice words for you if you're Russian, though:

> As to sending me a revert patch — please use whatever mush you call brains. I'm Finnish. Did you think I'd be ‘supporting’ Russian aggression?

(from https://therecord.media/linus-torvalds-russian-linux-kernel-...)

I think this shows that there is a blurry line between moral and practical judgment, and that the type of judgment might not even matter in the way you think. Sure, this episode may have been started by OFAC sanctioning various Russian companies, but Linus is happy to invoke more personal moral terms; and ultimately whether it's practical or moral doesn't matter, a bunch of Russian folks found themselves without a commit bit seemingly at random and the same could happen to anyone regardless of a particular platform's or company's judgment.

The fixed consumption plans are offering several times their worth compared to API pricing with completely free cache reads: https://she-llac.com/claude-limits

I look at that and think that they must be losing money hand over fist on something like this, not that this shows what their margins are like. If their margins are like this then I don't see why they'd be raising money and shuffling it around in circles.

> If it's really that easy to get better coding performance, why haven't the chinese labs replicated it?

Nobody said it would be easy! I just think it's possible, and that presumably they will get around to doing it at some point.

I'm arguing we can't trust retail prices because the marginal pricing isn't meaningfully connected to it anyway.

But if we have to look at what we think margins might look like, DeepSeek continues to host v4 Flash at the existing price despite competitors beating it in price (https://openrouter.ai/deepseek/deepseek-v4-flash), so there's at least one example of a Chinese lab charging a predetermined price despite competition. And no one but Moonshot is hosting Kimi K3 yet (https://openrouter.ai/moonshotai/kimi-k3). Perhaps there's room in the market for those who release their models to make margin on them.

And I believe my Composer example speaks for itself. The open models are behind but there's tangible proof they can be tuned for pareto frontier efficiency. See "Cost per Task" at https://artificialanalysis.ai/agents/coding-agents.

Sure, let's have a look...

> I highly doubt that Chinese models are cheaper to serve on a marginal cost basis, they just seem cheaper because Anthropic and OpenAI are so supply constrained that they are charging far more than they would if there were sufficient supply to meet the demand for intelligence. [emphasis mine]

I guess I'm missing the part of this article where they bring hard numbers in to back up the argument here. What work was attempted? https://cursor.com/evals shows the previous generation of open models (Kimi K2.7) trading blows with the others, cost effectively. Composer 2.5 is itself a fine-tune of K2.7, and it's apparently quite token efficient, so why would it be impossible for a Chinese lab to achieve something similar? GLM 5.2 Max is also ranked above the lower end OpenAI models and is not far off in price.

It's weird to have this entire discussion about tokenomics without mention of the circular financing and debt raised by labs in the West, which can then essentially give away their capacity to end users. OpenAI giving away quota resets to subscribers like candy on Halloween while their compute partner Oracle's bonds is reevaluated to be one grade above junk? How?

I don't think you can make an argument about the future one way or another by arguing using the listed prices. The math is not internally consistent enough for it.

I've played through the whole of story mode with no more than a few rare world streaming issues and that was a couple years ago. You could give it another shot, I recommend binding the gyro to one of the back grip buttons for precise aim.

Better yet, get the original release of GTA: San Andreas. It holds up to several replays in a way that GTA V does not imo.

> Movi M10 uppper brace and bolt holder

They're cat ears. You can just admit it. We won't judge.

It just is what it is. Sometimes you want to write the obvious query without the DB getting in your way, and other times you want to know as soon as possible that you're doing something that won't scale under exponential load. At this point in my career I prefer the latter, but the former will always have a special place in my heart.

> Maybe one day I’ll learn to read a query plan.

With SQLite's `.expert` mode you can delay that day a little longer: https://www.sqlite.org/cli.html#index_recommendations_sqlite...

  sqlite> CREATE TABLE x1(a, b, c);                  -- Create table in database 
  sqlite> .expert
  sqlite> SELECT * FROM x1 WHERE a=? AND b>?;        -- Analyze this SELECT 
  CREATE INDEX x1_idx_000123a7 ON x1(a, b);

  0|0|0|SEARCH TABLE x1 USING INDEX x1_idx_000123a7 (a=? AND b>?)

  sqlite> CREATE INDEX x1ab ON x1(a, b);             -- Create the recommended index 
  sqlite> .expert
  sqlite> SELECT * FROM x1 WHERE a=? AND b>?;        -- Re-analyze the same SELECT 
  (no new indexes)

  0|0|0|SEARCH TABLE x1 USING INDEX x1ab (a=? AND b>?)

Also wrt

> My approach so far has been to just do these cleanup operations in small batches so that I don’t need to do database queries that take more than 5 seconds to run. This whole experience has given me more of an appreciation for why someone might want to use a “real” database like Postgres which can have more than one writer at the same time though.

The advice for those " “real” " databases is generally to also do cleanup operations in small batches, they just tend to make it less obvious you're doing something unperformant in the smaller case. You're more right than you thought!

I believe it's 5 minutes on API pricing by default, though you can turn off caching or force it to 1 hour. Subscriptions are special, seemingly because Anthropic doesn't want to expose casual users to all of the tokenomics.

Anthropic's cache expires after 1 hour when using subscription endpoints, and for those cached tokens cache reads are free. It's generous (compared to API pricing) but it's not 100% free.