camdenreslink

What is considered a big event? Some papers published or letters sent between academics have invented entire new categories of mathematics that didn’t exist. Is it is big as calculus or Euclid’s elements, or Godel’s incompleteness theorem? Or Hilbert’s program of formalism?

I’m skeptical!

In my area many of those use cases are handled by Facebook pages already. For the general public this might still seem too technical.

I tried creating a similar concept at https://www.codembark.com but felt it would be too hard to monetize with the headwinds of AI. I ended up just getting rid of all of the paywalls and just offering it for free.

It definitely fell off before ChatGPT by at least a few years.

We know inference costs scale with usage. Unless they found a way to make it much more efficient (which is possible). I think it's a fair presumption.

Presumably the expenses would also be higher in 2026 as well, because inference costs scale with usage and I think training their new large models would be in there as well.

Some of these numbers are crazy. The scale of them is hard to comprehend.

- $42 billion net loss in 2025

- $518 billion in infrastructure obligations coming up (EDIT)

- 1/4 of revenue coming from 2 customers

Why was the title of the HN post changed from the actual title of the blog post? I thought it was clear in its meaning.

If you only fund “commercially viable” mathematics, then you are going to miss out on the majority of the most commercially important mathematics.

At least historically, we haven’t really been able to tell what will be transformative in terms of real world applications ahead of time.

You used a calculator every time you needed to multiply or divide simple numbers while reducing matrices to echelon form in introductory linear algebra? Or when doing simple arithmetic necessary to find integrals and derivatives in calculus? I bet you in fact did have those tables memorized, even if it wasn't a conscious effort.

At much much lower rates than accidental overdose rates assuredly.

There is more to development than domain knowledge. There are decades of wisdom about development best practices on how to keep a large codebase maintainable, performant, testable, etc. that LLMs frequently don't follow. And now humans aren't practicing those skills either or passing down lessons learned to the next generation of programmers.

Maybe it will be no big deal, and nobody will read code anymore, but it is understandable why somebody might be concerned about it.

Not possible with jobs with on-call responsibilities (most senior software engineering jobs in my experience).

You could do this for code reviews. Have Claude include one subtle bug into each PR. If the reviewer does not spot it, then the PR is not mergable.

I don't think the details of the company's benefits plan is considered your private information. But I am concerned about errors in this parsing leading to incorrect results (and that compounding everywhere throughout society when this sort of workflow becomes ubiquitous).

One lesson that could be applicable, is that your passion doesn't necessarily have to be your "job". You can perform music on the weekends, or be an active participant in your local indie dev gamer community while having a totally boring corporate job. I've seen a lot of people burned by trying to make their passion their actual career.

I think your ability to market will matter the most (which is also pretty much the same as ever).

I think their business model was sustainable, but depended on SEO. AI made it so people would either use the LLM generated tailwind syntax, or the "AI Overview" at the top of the Google search results and never land on the official tailwind documentation site. Less visits on the official docs meant less people would see the paid offerings, which means less people would buy.

It's possible they could have explored other marketing channels (youtube?), but it seems like the death of SEO for this sort of content lead to the main issues.

Probably “Alcoholics Anonymous”

You can imagine with more operations being available to be done more cheaply and quickly the LLM doesn't need to "one shot" a solution. It could try many solutions, test them, throw some away, wiggle some of the parameters like a genetic algorithm, see how that changes the result, and converge on an optimal solution (based on whatever the cost function is). Basically producing a good result could become like an optimization problem. That would be way too expensive and slow right now.

I think the scrapers have gotten pretty good and can imitate real traffic. And it isn’t just scraping for model training. A large amount would be LLM tool calls searching the web (and HN) for info about some prompt.

There is caching to consider. Switching models throws away the cached tokens.

This is basically what my life is like (going on 9 years now). I don't think it has necessarily led to burnout for me. Keep in mind that time spent with your children isn't all work. There is some fun and leisure in there as well (although it is a lot of work too).

When you work at an enterprise that depends on Github, frequent outages are actually a pretty big deal and begin to impact productivity in a real way.

There are nasty things we can do in software that can indeed make debugging difficult (for instance in the “make everything reactive” craze, some random subscriber could trigger from some piece of state changing that isn’t obvious at all and fully asynchronous and in a totally different part of the code).

LLM output is not like IKEA furniture at all. IKEA furniture is mass produced, identical, furniture that can be combined to accomplish some lifestyle goal.

LLM code is different every time. Sometimes it’s better than the finest custom carpentry. Frequently it is much worse than IKEA quality (in the metaphor).

Some people (not Anthropic necessarily) are claiming 5x or 10x productivity improvement. So it is relevant, as it is being somewhat oversold by some.

I find that LLMs also generate a lot of false positives, or extremely minor issues that don't warrant a fix (that are always overstated by the LLM as very important!). Signal to noise is still not great and requires somebody to wade through and pick out the actual good findings.

I have had to force quit Spotify a couple times recently because it was using many GB of memory to the point where it was unresponsive.

Agreed, considering "taste" to only mean UX or aesthetics is taking things too literally in my opinion.