> Humans Who Are Not Concentrating Are Not General Intelligences^

But yeah, you should still treat them with humanity. (Related: you should treat LLMs well, not because they're human but because you are^^)

^ https://srconstantin.github.io/2019/02/25/humans-who-are-not...

^^ hmm couldn't find this tweet but didn't look too hard

A few caveats: you need to know the key to the function (used when generating the text) and you need to know the bias it would introduce

The point is that you do not need to know the full prefix, just a modest sample set of contiguous tokens

Also when people ask questions like this it usually means they're asking questions about a point in time, eg given today's numbers

But valuation should account for trajectory (where will they be in five years?)

In the hypothesized bull case for ai, they benefit dramatically from the secular tailwinds (ie unrelated to their own business strategy) of AI

So one way to consider this is it's a hedge by stripe in case AI becomes as big as some people think it is. And 7b to get in on the ai wave is a good deal in that case. And in all the other cases, it goes to zero and stripe is fine

Note, there are many ways to represent words visually on computers that look identical

You can carefully select which pseudorandom number generator (prng) you use to be able to id text of a certain length. I expect there is some performance characteristic you have to manage since you're doing this on every inference, but once you do that it doesn't change the output in any meaningful way (the prng is still a statistically valid prng, it just happens to let you check if the output used that prng)

The point is that you can do this simply by swapping to a different RNG, which isn't noticeable to the end user, and while it changes the output, it's not any different from how using a different seed or being lumped in a different batch will change the output.

^ excerpt:

> So then to watermark, instead of selecting the next token randomly, the idea will be to select it pseudorandomly, using a cryptographic pseudorandom function, whose key is known only to OpenAI. That won’t make any detectable difference to the end user, assuming the end user can’t distinguish the pseudorandom numbers from truly random ones. But now you can choose a pseudorandom function that secretly biases a certain score—a sum over a certain function g evaluated at each n-gram (sequence of n consecutive tokens), for some small n—which score you can also compute if you know the key for this pseudorandom function.

Do you think this is an impossible task and we shouldn't try to solve it? Or do you think it's doable and that some ai detectors might be better than others?

Many users are not smart enough to realize that the transcription step is where the ai (watermarks) were necessarily injected.

When you have a lot of free users the business demands that you serve them with the best cheap model you can build

And time spent building that may provide dividends (eg OpenAI has very good RL and reasoning) but it might take resources away from the larger model training

(I have no inside knowledge, so please consider this to all be speculation)

https://x.com/polynoamial/status/2083478171975082334

As a complete guess, it seems like they tested hundreds to thousands of problems with a relatively low per-problem budget

--

The linked tweet from Noam Brown at OpenAI reads:

> And yes we did try other major problems without success. Sadly no Millennium Prize problems (yet).

> But also, we didn’t spend a lot on each problem. It’s possible to push test-time compute much further.

I discovered the reset, but it took a minute and I expect given the way I read being able to explicitly walk the high watermark back a bit would be nice (maybe that's doable if you select text? I didn't try!)

The format is nice, I love standard ebooks

I opened Moby Dick but didn't see how to listen to it on my phone, not sure what the status of the "listen" effort is.

I listened to some books from standard ebooks using eleven labs, but I'm really excited about a better implementation

First, this is about not just a job or career, but someone's identity. And we must have compassion that they are losing something that is fundamental to who they are. To them, it doesn't just feel like they're losing it, it is being taken by these companies that have so often acted in ways we despise

And that is a tragedy! And there are so many tragedies like this that will happen regularly as the technology advances

But my second reaction is one of this great shared experience. I studied math. Many of my friends are mathematicians. And it is now possible for anybody to access mathematical insights or think deeply about strange conjectures and theorems that were previously incomprehensible to anyone who hadn't at least studied mathematics in college

I don't know if 3b1b's Grant Sanderson considers himself a mathematician or science communicator, but I think of him as both. And while I expect he is someone with the mental capacity to find and prove new things in the world, I am grateful for the time he spends instead understanding things at a fundamental level and explaining and celebrating those concepts with his audience.

Not every mathematician can be or wants to be _that_ kind of mathematician. But for the moment, it is enough for me that higher mathematics is more accessible than ever.

And with some trepidation, I predict that the "traditional" job of a mathematician will change (of course it will). And it will change in big obvious ways and also subtle little ones. How will it change, though?

Importantly, math is actually going to be one of the fields where the fundamental things we thought we knew are shaken, because in the next five years we are going to start to see connections between things that were previously considered completely separate.

And so one way the job will change is that anyone who discusses math regularly will need to learn new things.

To me, this is exciting. It's almost like finding a bunch of new dinosaur fossils that fundamentally reshape our understanding. There's going to be a lot of work to do!

Could they do it? Sure, but to what end? It would make more people hate them and feel even more "take our interesting work." Pitching it as a useful tool just makes more sense on all levels

As an aside, there's this idea in math that when you create a new field you shouldn't solve all the easy problems - you need to entice other people to learn about the field!

I expect there's some element of that here. It's much better for OpenAI and Anthropic if their users are the ones discovering and writing up the results of the AI solving hard math. Look at the high school and college age students who have become ai power users and potentially learned how to use git to contribute ai generated solutions

(Related: I believe Terry has also gotten all of the subscriptions gifted to him)

https://xenaproject.wordpress.com/2026/07/20/human-mathemati...

Agreed on linear vs n log n, ofc, and streaming considerations might also be relevant

> Consider the following real problem, one of the steps in scikit-learn’s gradient histogram boosting algorithm:

> You have a large array of floating point numbers.

> You want to assign them to the integer range 0-254, spread out evenly.

Naively I would consider sorting the initial array and then using something like `batched` from itertools to chunk them into the 255 buckets - binary search will give you a bunch of random accesses, and sorting can be cache-oblivious (eg efficient for arbitrary data sizes)

But I'm somewhat concerned I don't fully understand the underlying problem being solved with this step, so I might be misunderstanding the intended result