mixedmath

Claude models seem to work far better in Anthropic harnesses. I don't know if this is because they are trained in that context or because they offer different models through external APIs.

Would someone be interested in explaining what in the world is going on in that article?

This is true in my case! I've had more people actually buy me a coffee after interacting with my site than to click on my Buy Me a Coffee button.

I find a link to "Buy me a Coffee" to not at all look like an ad. It's a funny-shaped "donate" button. And I much prefer it to actual ads.

I wasn't aware that there were any public, commonly-used voting machines that we could check.

I was very surprised to see that the article explicitly says it will not consider this answer!

Are there leaderboards that you follow or trust?

Also, do you have preferred OCR models in your experience? I've had some success with dots.OCR, but I'm only beginning to need to work with OCR.

How does this relate to the mean entropy of 1 noted in the OP article?

I wrote a similar app when mathbin was shutting down. It allows about 1500 characters of mathjax-displayed notes. [1]

[1]: https://davidlowryduda.com/mathshare/

I'm not very familiar with proper exception usage in C++. Would you mind expanding a bit on this comment and describing the misunderstanding?

That seemed to me like a radically high estimate of people who use ad blockers. But I see that the first page of results on MyFavoriteWebSearchEngine support that claim.

> This was a tough decision, having used Gmail since 2007/2008. However, I had to draw the line and stop giving Google my data for free. > > The problem with email is that everything is transmitted in plain text.

Interestingly, one of my biggest problems with Gmail is that they don't allow actual plaintext. I used to routinely collaborate with developers who were vision-impaired, and the official Gmail phone app wouldn't let me send them plaintext email. Instead, it was some sort of HTML thing. Unfortunately, we sometimes sent code snippets to each other over email, and though admittedly it looked more or less fine, Gmail changed the underlying representation enough that my collaborators' screen readers would mess up on the parsing.

This led to me leaving Gmail on my phone, which led ultimately to me leaving Gmail entirely.

Google is completely inadequate at mathematical search. But here is a concrete problem that no search seems to handle: given some complicated integral (say, some contour integral involving a K-Bessel function), find where it appears in the literature.

Most search will totally fail, because this is made of math symbols. Embedding-based search will give various related things involving, say, integrals and Bessel functions. But then I end up opening Gradshteyn and Ryzhik and trying to find where in this book the relevant terrible integrals appear.

This is a common experience for analytic number theorists. And it's a lousy experience.

From the title, I had thought that this would be a new tool for searching science, such as searching the arxiv. But this is actually a survey.

I quote the conclusion of the survey:

---

In conclusion, rapid advancements in artificial intelligence, particularly large language models like OpenAI-o1 and DeepSeek-R1, have demonstrated substantial potential in areas such as logical reasoning and experimental coding. These developments have sparked increasing interest in applying AI to scientific research. However, despite the growing potential of AI in this domain, there is a lack of comprehensive surveys that consolidate current knowledge, hindering further progress. This paper addresses this gap by providing a detailed survey and unified framework for AI4Research. Our contributions include a systematic taxonomy for classifying AI4Research tasks, identification of key research gaps and future directions, and a compilation of open-source resources to support the community. We believe this work will enhance our understanding of AI’s role in research and serve as a catalyst for future advancements in the field.

---

I jumped at this because I'm a mathematician who has been complaining about the lack of effective mathematical search for several years.

I've never seen this before. But yes, this does work. And that's very nice.

Thank you for the tip.

Do you use maxima? Would you say it's still worth using and learning?

This will not be taught in textbooks. Wildberger works in a math world where "infinity" or even "really really big numbers" don't exist. There is nothing mathematically wrong about that world, but others (like me) do not find this problem context to be a useful framework to work in.

I haven't read anything from the London Review of Books before. But this seems excellent. Is this indicative of their general quality?

I've been using it as my primary search engine for a couple of months. It's not great as a search engine. I find their locality of search to not be well-supported (e.g. the search "food near me" works good in google and not great in ecosia).

Ecosia doesn't emphasize recent events, news, or posts in search results as much as I'm used to --- but I haven't decided if this is good or bad.

It's not so bad that I've changed. But I do sometimes use a better search engine when I want better results.

I made a partial replacement that doesn't allow user-submitted content at https://davidlowryduda.com/static/MathShare/. It just stores the content in the URL, and is limited by URL size limits. In practice this means you have approximately one page of text.

I wrote about making this at https://davidlowryduda.com/mathshare/. I was trying out the LLM interaction that made it near the top of HN recently, and it worked very well.

I'm confronted with a similar problem frequently. I have a growing bash script and it's slowly growing in complexity. Once bash scripts become sufficiently long, I find editing them later to be very annoying.

So instead, at some point I change the language entirely and write a utility in python/lua/c/whatever other language I want.

As time goes on, my limit for "sufficient complexity" to justify leaving bash and using something like python has dropped radically. Now I follow the rule that as soon as I do something "nontrivial", it should be in a scripting language.

As a side-effect, my bash scripting skills are worse than they once were. And now the scope of what I consider "trivial" is shrinking!

About 5 years ago, StackOverflow messed up and declared that they were making all content submitted by users available under CC-BY-SA 4.0 [1]. The error here is that the users-content agreement was that all users' contributions are made available under CC-BY-SA 3.0 (and not anything about later). In the middle there were also some licensing problems concerning code vs noncode that were confusing.

I remember thinking that if any of the super answerers really wanted, they could have tried to sue for illegally making their answers available under a different license. But I thought that without any damages, this probably wasn't likely to succeed.

But now I wonder whether making all content available to AI scrapers and OpenAI in particular might be enough to actually base a case. As far as I can tell, StackOverflow continued being duplicitous with what license applies to what content for half of the year 2018 and the first few months of the year 2019. Their current licensing suggests CC-BY-SA 3.0 for things before May 5 2018, and CC-BY-SA 4.0 for things after. Sometime in early 2019 (if memory serves, it was after the meta post I link to), they made users login again and accept a new license agreement for relicensing content. But those middle months are murky.

I should emphasize that I know nothing.

[1]: https://meta.stackexchange.com/q/333089/205676

What does an /ideas page mean to you?

> The STEM movement intentionally excludes humanities and socials on purpose.

Could you say more?

I'm not sure if I would expect any bridge to survive being struck by an enormous container ship.

I would also be interested to know how they decide to pick up a site. I was very surprised to learn that a technical note posted only to my website was picked up somehow. (I am a mathematician and so there are other things on my site, but it’s some custom static site generator thing and I’m still astounded).

What does AI assisted reading mean to you?

This is an interesting problem. I love my library. And I use both physical and digital books.

On the one hand, it's probably true that more people can read a digital book. Ob the other hand, I have read fewer than 20 pages of a bunch of ebooks - enough to see if I might like them.

Do these count as checkouts? Am I a very expensive library user?

(And I also am happy to have several ebooks that I'll read sometime, casually renewing them until I get to them).

I don't follow the implication. I think you're suggesting that in France, they're not 100% beef. What are they?