petters

Me neither knocks on wood

Perhaps the render should use the real sizes and positions in meters for all three objects, at a time given by the user

Exactly! I can not believe this is in the top spot.

You likely just got lucky.

I have lost commands in the Bash history as well and I don't know why. Fortunately, I store all commands executed (+ time, return value etc.) in an sqlite database so I run a program from time to time to import that into the Bash history. The Bash history is useful for Ctrl+R

Only 1000x overhead would make some image classification tasks go from 1ms to 1s. That’s viable for some applications!

These are also very cool: https://antfu.me/posts/ai-qrcode

Diffusion image generation constrained to also be a valid QR code

Why does figure 1 use the same color and style for two different graphs?

> Since connection poolers reuse connections between clients, the connection state of one client “leaks” into the connection state of another.

Wow this is very bad. This actually happens in typical Postgres setups?

5B revenue is not the same as 5B in the bank collecting interest.

> a hand tuned input loop written in C that takes <<1ms

Yes, I would certainly expect much less than 1ms. Perhaps 1µs should be the goal?

> If we give an LLM a prompt that reads “The following is a conversation between Julius Caesar and Genghis Khan,” it will generate a coherent dialogue between the two historical figures. But no matter how detailed the responses are, no matter how vividly they recount their respective historical accomplishments, we would never conclude that the LLM has conjured up digital re-creations of Julius Caesar and Genghis Khan, nor would we suggest that the historical figures are conscious

They might be in principle. It could be that the best way to generate a plausible dialogue is to bring up re-creations of the characters and have them act it out. LLMs definitely have been demonstrated to have world models in some cases. That helps generating text.

It also means you'll have to onboard a lot of new employees all the time. That sounds exhausting.

Wikipedia is amazing. The Swedish articles are even longer.

Karpathy replied to IS with ”agree” at the time

Many dismiss Dawkins here but Ilya Sutskever wrote in 2022: “it may be that today's large neural networks are slightly conscious.”

We have a very good idea of all math behind chemistry. But the equations are very difficult to solve.

He is much better at building hardware than he is writing software.

That's a good idea and it exists: https://www.johndcook.com/blog/2026/04/18/qlora/

It seems quite wastful to have two zeros when you only have 4 bits it total

You could add a feature where it will compute the global optimum of any function of a small number of variables. Branch and bound with interval arithmetic works well for a small number of variables.

Disjoint unions of intervals seems like a nice thing to have

Yes, that blog post could have been much shorter….

They have found a large number in OpenSSl

Dungeon Crawler Carl is not science fiction. At least I would not recommend it to someone looking for science fiction with “interesting ideas.” It’s a comedy about an RPG with magic.

But if that’s what you’re looking for, it’s pretty good

> the GPU is limited by the Thunderbolt port

Not everything is limited by the transfer speed to/from the GPU. LLM inference, for example.

> On a 34" ultrawide monitor, it was too easy to put YouTube running on the left side, and whatever else on the right.

Yes, if you were doing that, almost any change to your environment that stops that will be good. I don't think you'd have to give up your monitor.

That challenge was pretty stupid. I could read the question and I’m not even a native speaker. We can of course easily come up with much better challenges

As a TA, I've seen adults try to pass initial college calculus many times (and failing - you were allowed to try several times) with enormous effort. It's not a small multiplier

And this was still people selected from the small subset of the population choosing an engineering major. Human are much, much more different than you seem to think

I think that bias is not due to the proportion of books and more due to how they are fine-tuned after the pretraining.

We identify the real number 2 with the rational number 2 with the integer 2 with the natural number 2. It does not seem so strange to also identify the complex number 2 with those.

Going from 220 to 450 ms would be a disaster in my project. It has many thousands of files. Recompilation of almost everything happens from time to time.

If those made-up numbers were true, they would be very significant and an argument in favor of keeping the code in C