theahura

Next time I'll hedge more

And here's what people don't know: its not poetry, it's passion -- and that makes the difference.

A freedom sucking job (lots of travel to places I don't want to be) or 90h weeks is a no for me.

Actually I think you agree with the central premise of the article, even if the examples didn't resonate

Interestingly, two words that aren't mentioned even once in this entire article: "co-founder" and "hire"

I actually debated including this!

But I ended up not doing so, because your co-founder and the people you hire will end up being more aligned to you (your values, judgements, whatever) than even the market you end up in, especially early on. And if they don't, that's an even bigger problem. Founder breakups are the #1 cause of startup death.

Also, agreed, all of this is out the window once you _already_ have the ability to hire a bunch of other people, but that rarely happens right out the gate, which is where most of my interest lies

straw man examples

not straw men, all real but unnamed examples (also not really what a straw man is but whatever)

Perhaps you could submit something you're curious about, instead?

Hi, author here. I write about what I'm curious about. I don't think it's promotion (both in spirit or letter) to share what I am curious about. I hate just linking to <openai blog / anthropic blog / karpathy twitter / tech crunch / engadget> without any commentary, since that has very low value-above-replacement.

I'm curious why you went into my posting history -- do you do that for every post?

[EDIT] to be clear this isnt written by AI, despite a few people seemingly tripped up by it. Check pangram if you doubt

Let's spend $100 billion + on Iran, but also cancel student funding because don't you know the budget is too high?

Do you really want to make the case that the Federal government should fund these?

No, but I don't think means testing is a better alternative. You and I both know that this is a wedge to punish perceived political enemies, as much as you know that the total number of people who graduate with a degree in "somatic body work" is such a small fraction of people as to be negligible

Nationally, about 10,600 to 10,700 students graduate annually with a degree or certificate in the broader category of Somatic Bodywork & Therapeutic Services. The vast majority (over 10,500) specialize in Massage Therapy, while purely focused Somatic Bodywork programs see much smaller cohorts, averaging roughly 40 to 50 graduates per year across all US institutions. (From Gemini)

This is kinda like saying we should get rid of encryption because a few people may use it for bad things, or that we should require identification before anyone can get on the Internet because some people may be hurt by it

Education isn't a binary, so I'm not sure why we should arbitrarily draw the line at highschool as we become richer as a nation

Yes we should also do this

There is no evidence that students were actually doing this in any significant numbers.

premature optimization is the root of all evil. Seems like we shouldve actually shown that kids would do that before putting it into law

whats the goal of government education subsidies

a more educated populace is a public and civic good on its own terms. Public funding for education is maybe partially for economic returns, but is mostly because education is a necessary part of a functioning democracy and a necessary part of living a good fulfilling life

imo this is pants on head backwards. The whole problem with the current university system is that it has become exclusively a credentialing system that everyone uses to justify higher salaries. We’ve completely left the education part of it by the wayside…except for the liberal arts majors who are actually there just to learn! This rule is just encoding the existing tulip mania into federal law directly, by making it clear that the ONLY reason one goes to school is for future $$$

always a fun day when someone rediscovers these. Lucky 10000

in case folks are interested, i wrote up a ~layman's review of each paper over the course of two years a while back. Several of those reviews ended up doing reasonably well on hn. Full analysis of the ~23 docs that were papers and not massive books

https://12gramsofcarbon.com/p/ilyas-30-papers-to-carmack-tab...

(note that I am the author!)

Like, if you have complaints about the thing, perhaps you should address them to your supplier directly. None of your readers can help, and nobody's magic folk solution to your problem is better than yours.

I think you may just misunderstand the point of having / writing a personal blog. I write because it's fun! Whether the reader gets any value out of reading it is almost entirely beside the point.

(Also several comments here directly post a fix to the problem stated in the blog post, so readers can and do often help)

With the million-context-window models we never hit compaction, observed over hundreds of sessions. What are you doing that has you hitting compaction regularly?

interesting take. I think I disagree, but I like this take a lot and I had to think about it.

First, I think that models still need a context layer. One way to think about 'context' is as a form of compression. You provide the model context because it makes it easier for the model to figure out what to do. Even in a world with infinite model capacity and infinite model context, this is still useful because it allows the model to avoid rederiving everything from first principles every time. As long as models perform better using fewer tokens and as long as we care about token spend, context is a useful (necessary?) shortcut.

Once you bite that you need some form of context layer, the question is which. Here I do agree that it is better to work with what the models will find familiar (markdown files colocated with code, for eg). But this speaks to over-engineered solutions not understanding their main user (the agent) more than it does the need or lack there of.

starting a company that benefits from tokenmaxxing, mostly

Hi, author here. I'm probably somewhat financially benefiting from tokenmaxxing. I also just believe compounding correctness is right, based on my own experience using the tools (which is why I have structured my life to try and financially benefit from tokenmaxxing)

(its in the article, which predicts that there will be another round of tokenmaxxing with different underlying incentives)

that is a better title! Added it as a subheader

Dan Luu at MS: https://x.com/danluu/status/1487228574608211969?lang=en

Independent of everything else, very interesting to see how polarized the comments are here

I do abuse this title format, guilty as charged

“I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,” Commerce Secretary Howard Lutnick wrote to Anthropic’s chief compute officer Tom Brown Friday

why is the commerce secretary making this decision

usgov picking winners and losers in AI --> usgov picking winners and losers in every industry (speedrun, any %)

not to be dismissive, but isn't this the same as the ralph loop thing everyone was talking about 6 months ago?

No, we use subscriptions. We wrap the underlying Claude Code with ACP. See: https://github.com/tilework-tech/nori-cli

Not all the things that are good will rise to the top, but most of the things that rise to the top will be good. We've gotten pretty good at ranking systems as a species at this point, I'd say

most flash games were horrible too! You had to go through a load of crap to find games like boxhead, motherload, or bloons. I'm a big believer in volume here. You don't have to be an amazing programmer to be an amazing game designer, but before, the former was a prerequisite for even getting started. The beauty of AI tools applied to games is that you can just focus on the latter. Over time the gems will rise to the top