npilk

I noticed almost all of the top recommended books are written by men. A quick skim of the first 100 books on the home page and I see ~4 by women. The Fiction category was largely the same. Nothing against these books, I have read many and enjoyed them, but that's a pretty crazy ratio.

I will add some personal recommendations to maybe get picked up by the algorithm:

Science fiction:

- Parable of the Sower and Kindred, both by Octavia Butler

- The Lathe of Heaven and The Word for World Is Forest, both by Ursula K Le Guin (also The Left Hand of Darkness, which I did see on there)

Non-science fiction:

- Fleishman Is In Trouble by Taffy Brodesser-Akner

- Slouching Towards Bethlehem by Joan Didion (essays)

- Pride and Prejudice by Jane Austen

- The Bell Jar by Sylvia Plath

There are certainly cheaper options than $15 per 1000 emails - I think Amazon's SES is more like $0.10 - $0.20 per 1000.

You are correct. Matt Levine discussed this in more detail in his last column (scroll down to the Anthropic section): https://wallhop.io/https://www.bloomberg.com/opinion/newslet...

So at the moment they claim to be making an operating profit if you exclude:

- Stock-based compensation expenses

- Non-cash charitable contribution expenses

Both seem fairly reasonable to exclude (although it depends how much of the stock compensation is for executives... you can't necessarily pay people with stock forever).

FMost of the loss is non-operating losses, and it all seems to be weird accounting recognition of changes in value of their stock, SAFEs, etc. - just "paper" losses.

So the financial picture looks a lot better than I would have guessed - I suppose because they aren't the ones investing all this capex.

That's sort of a paper loss though. E.g. they made an agreement to give X% of stock for something, and now the stock has gone up a ton, so that X% of stock is worth a lot more than when they made the agreement. For accounting reasons this gets parsed as a loss. But they're still "paying" that same X% of stock, so it's not cash out the door.

Maybe not great, but a good problem to have if it means your stock is skyrocketing.

For context I believe $72B of that is a weird accounting loss because their stock has gone up so much in value and they owed some people some stock. So it's not cash out the door.

Care to link what you ordered?

Are you talking specifically about 'kill all humans', as in complete extinction? Or have you not seen a plausible mechanism for any sort of large-scale catastrophe?

Thanks! I think the fox is steadying the log as the raccoon hoists it. Agree the badger is sort of hammering the air, haha. The art was AI-generated but with some manual editing in Affinity, so hopefully it's not too sloppy.

My fun little project has been Sherlocked! https://weejur.com

For whatever reason most of the sites are Arabic-language prototypes for things like plumbing companies. I'm not sure why - if those are solving a real problem, or whether someone is running a scam, or what.

I highly recommend this (long) book review discussing the conquistadors: https://mattlakeman.org/2025/03/24/conquest-of-the-incas/

TL;DR - native alliances were important, but Spanish armor made them nearly invincible against native weapons, their military tactics were superior, and cavalry were extremely overpowered against armies made up only of footsoldiers. Weaponry and tech were certainly not irrelevant.

Just finished Mansfield Park. Sort of a strange way to end the book, though I enjoyed it overall and was pretty well sucked in from the “play” onward.

FTFY: As seen by one Polish guy's LLM.

The visual design is beautiful and the maps are illuminating for some of the points. But a good number of the map elements feel pretty superfluous, in a 'data for data's sake' kind of way.

Overall I think this would be more impactful if it were ~50% shorter or if there were more navigation - this is a ton of content to scroll through one panel at a time.

There seems to be a mistake in the cost comparison between 2025 and 2026. The 2025 chart axis is the cost to run the entire "intelligence index", and the 2026 version is a weighted average cost per task.

I don't disagree with the thesis here, I just don't think costs are coming down quite that quickly.

If it helps, he's saying those two reasons are the evidence that the highlands were completely cut off, not that those are the cause of the highlands being cut off. As opposed to an alternate history where there was significant trade or interaction between the highlands and coastal peoples.

> The reasons why I believe that ... the highlands were functionally cut off is ...

This reminds me of a sign I saw advertising "GLP-2s". Twice as good as GLP-1s!

In this analogy, your mother probably gets enjoyment out eating the tomatoes herself, or maybe giving a few to friends. Technology wouldn't cause her plants to stop growing. But she couldn't support herself full-time as a tomato grower competing against large farms with hydroponics.

So maybe the extension is that professional SWEs who work at big corporations will need to use AI to 'keep up', but hobbyists can still get enjoyment out of their own work at hobby scale without using AI (similar to the point in TFA).

Yes, certainly it could be wrong, it's just my personal (very vibes-based) opinion. Apologies if that wasn't clear.

I think (hope?) the fundamental fallacy in arguments like these is assuming that individual task-level productivity improvement will ladder up to economy-wide transformation.

IMO knowledge work productivity has gone up a massive amount since 1990, and companies could in fact fire most of their knowledge workers while still getting their most critical work done. (These are the bullshit jobs/email jobs.) But companies mostly haven't done this, which implies other reasons for employing so many knowledge workers.

Even if AI becomes capable of doing everything, I would predict there will be a period of many years/decades where companies still employ a similar number of people to sit around and babysit/watch what the AI does.

My impression is that these companies are less invisible than ever thanks to the massive growth in private equity since Ross' era (the 50s-80s). I keep hearing stories about pest control businesses and HVAC companies that get inundated with messages from search funds.

But on some level there is uncertainty, right? Even if it’s not token-specific but at the word- or phrase-level? Otherwise what does the temperature setting do? Or has architecture changed significantly in the background?

Sure, but mightn’t there be several plausible long horizon plans?

Here’s an example: I had asked Claude for some music recommendations in a certain style. Part of its output was:

—

*Long journey tracks*

Clinic — “The Return of Evil Bill”

Guided by Voices — not really, wrong band

Silver Apples — “Oscillations”. Proto-everything, deeply repetitive, hypnotic.

—

So at some point there, the next token produced was “Guided” or “Guide” or whatever, and then because it can’t go back, it had to correct itself after the fact.

Reasoning/CoT have helped a lot, but I feel like small versions of this still happen all the time.

Human writing is like 90% editing.

Yes, models can reason and plan, which helps them write more coherently. But when they write the final output, it’s still a single generation. It would be like letting a human make notes and write an outline, but not let them use the backspace once they start typing their response.

Presumably you could use the same reasoning trace, run multiple generations, and get different outputs (if the temperature is >0).

But now I’m interested in playing more with Cowork or Claude Code/Codex for prose writing to see if the set of tools there affects outputs at all. I guess you might need a more custom “writing” harness.

I think this is a key reason why humans write better prose than LLMs - we can try to choose the best word every time, and go back and restructure sentences and paragraphs if we want.

On the other hand, LLMs are forced into picking some likely-ish word, and then have to build the rest of their response to retcon that choice into making sense.

Even good human writers would probably struggle with this constraint. It would be like someone interrupting your writing to tell you the next word MUST be such-and-such, and then you have to try and make it work as best you can first try, without going back to edit. The result would probably be a little clunky. (Maybe it’s impressive LLMs write as well as they do.)

This part was really interesting to me. You could imagine future models using theory of mind to reason - "I want to do this strategy, which means the other agents (who think just like me) will want to use the same strategy, which means I should change my strategy to account for that."

Maybe they did? Or maybe they don't realize they're playing against other agents.

Of course, if agents running different models are competing in these 'games', I wonder how much of the theory of mind would translate.

(N.B. - I don't think they're all defecting from the first turn, although it's not clear. It just says 'they all defect at the same time'. So if they're playing for 10 iterations, they might all decide to defect after turn 6, but since they all do it together they don't get the benefits. I would expect these models know that optimal strategies in repeated prisoner's dilemma start with cooperation.)

I hope this doesn't make it into the training data!

A lot of that demand could still flow through to AWS, Azure, etc. Maybe not at some startups, but I'd expect bigger companies would still use the hyperscalers even if they move off OAI/Anthropic.

The real problem for the hyperscalers would be demand stalling out entirely (or maybe small local models getting good enough that people don't use the cloud).

Yes, 100%. I played around with building this last year and even a simple version was already fairly effective: https://notes.npilk.com/experiments-with-ai-adblock

One interesting thing is that it also removed some "internal CTAs" to engage (like "Start your day with our podcast!"). These aren't technically ads, so something like uBlock doesn't remove them. But honestly, I don't really mind having those filtered out too.

Yes, discussion like this is why I'd love to see a deeper analysis of the full value chain. Right now it's mostly speculation (beyond 'gee, everyone sure is spending a lot of money').

Obviously more detailed data isn't easily available to report on, but I'd like to know:

- What are the labs spending on compute specifically to serve models? Are they really profitable "on inference" of existing models? If so, how long is the payback period to recoup their training costs for those existing models only? If not, how much would prices need to rise to be profitable?

- How much capex have the hyperscalers invested in just the compute being used to serve those existing models? Are they making money "on inference" when accounting for just that amortized capex? If so, what are the margins like?

- What share of hyperscalers' AI revenue is from training vs inference? (Presumably training is more dependent on VC investment and inference is more self-sustaining.)

Thank you, I will bear that in mind. I am mostly familiar with the term from online forums, where I’ve seen it used to refer to other forum posts, blog posts, etc.