daxfohl

47 ms·

631.
by daxfohl 2y ago
Same thing in sql. Non null guarantees are great until you do a left join.
632.
by daxfohl 2y ago
Does TAO help? I imagine it's a lot easier to embed all this additional logic and metametametadata in a single service than distribute.
633.
by daxfohl 2y ago
"fuck the wallfacer project"
634.
by daxfohl 2y ago
So _would_ it possibly be useful for business apps? Or is that still a long way off?
635.
by daxfohl 2y ago
I remember reading (~10 years ago) that F* was created as part of a "Project Everest" a long time back, with the goal of creating a provable TLS implementation. I never saw that anything came out of that though. If it's that
636.
by daxfohl 2y ago
Eh, probably not. It's not like web components are a big threat to react. Nor that react is any kind of a profit driver anyway.
637.
by daxfohl 2y ago
Embrace, extend, extinguish?
638.
by daxfohl 3y ago
Wouldn't this be "transformer models" rather than LLMs?
639.
by daxfohl 3y ago
The same should be said for engineering design docs tbh
640.
by daxfohl 3y ago
Though note that was a question of possible redundancy, not possible inconsistency. Still, the outcome is probably roughly the same. We'd replace the inconsistent axiom with something similar and 99.99% of existing practical math would
641.
by daxfohl 3y ago
Seems like you could interview half the candidates and get a lot of time back.
642.
by daxfohl 3y ago
It's astounding to me that space is so empty that in the billions of light years from here to the edge of the universe, there's orders of magnitude less total interference than what you get from a small cloud. It's also astou
643.
by daxfohl 3y ago
Probably. So given I haven't bothered to look up this feature yet, and I've established for myself I'd rather have this than a dumbphone, I don't think it makes real world sense to buy a dumbphone, even though it sounds
644.
by daxfohl 3y ago
I mean, it could just be built into android, or maybe even an app with enough permissions. But agreed, even at that minimal level of friction, I'd probably turn it off most of the time, and deal with it via old fashioned self control.
645.
by daxfohl 3y ago
Yeah, or charge every night, or look for when you're in a hurry to get out the door, etc. Or buy, for that matter. Probably what I'd rather have is a phone that has a low power low function mode, that I can configure what apps are
646.
by daxfohl 3y ago
Or a good way to teleport out of local minima while training. Create a few clones and take the one with the steepest gradients.
647.
by daxfohl 3y ago
You could use this and try to tease out something similar to https://news.ycombinator.com/item?id=39487124 , but for NNs instead of images. Maybe it's possible to have this NN diffusion model explain the pieces of the N
648.
by daxfohl 3y ago
I wonder if you could do this on the NN-generating NN from https://news.ycombinator.com/item?id=39458363 . Then you could have it start to create a vocabulary and an understanding around what all these parameters and weights
649.
by daxfohl 3y ago
I remember a wittgenstein thing asking to think of a tree or something, and then point to the place in your head where that thought exists. It's kind of like that.
650.
by daxfohl 3y ago
Seems like it could be useful for resizing the networks, no? Start with ChatGPT 4 then release an open version of it with much fewer parameters. Or maybe some metaparameter that mucks with the sizes during training produces better results.
651.
by daxfohl 3y ago
To be fair, there was a paper a week ago showing how GPT-generated responses were easily detectable due to their "averageness" across so many dimensions. Maybe they ran ChatGPT through a GAN and this is what came out.
652.
by daxfohl 3y ago
Haha, that's exactly what I'm trying to avoid. We're looking at adopting openpolicyagent. Which frankly, a modern JSON-oriented datalog may be ideal. But this whole time I keep thinking, this would be so much easier in Haskel
653.
by daxfohl 3y ago
Looks like it has become abandonware, or at least the component they open sourced. Oh well, such is life in the Haskell ecosystem. https://github.com/facebook/Haxl
654.
by daxfohl 3y ago
I still wonder why Haskell sans IO isn't a thing. Seems like it would be a really great way to define/implement rules for a rules engine that supported it. I guess you'd still have to worry about OOM or CPU thrashing, but in
655.
by daxfohl 3y ago
Not really. Sure it adds some overhead but it's well optimized and not overly significant, especially for incremental collections. It can even pay for itself by defragging your memory, making subsequent allocations cheaper. The big pro
656.
by daxfohl 3y ago
Fwiw I'm going back to my original hypothesis and saying there's no limit at infinity, but there is a class of questions that are in none of the levels, essentially meaning you can't prove anything about their solvability. It
657.
by daxfohl 3y ago
> One annoying thing about both the Collatz and Goldbach conjectures is that if they are unprovable, it's also impossible to prove they're unprovable (unless math is inconsistent). This has been proved! So is there a countable
658.
by daxfohl 3y ago
Quality aside, there's a reason producers pay millions for A-list stars instead of any of the millions of really good aspiring actors in LA that they could hire for pennies. People will pay to see the new Matt Damon flick but wouldn&#x
659.
by daxfohl 3y ago
We won't and the world will go into a massive depression, destroying the market for AI produced garbage and staving off global warming for a few extra years in the process. So even better than UBI.
660.
by daxfohl 3y ago
The third batch, the one with the cat, the guy in bed has body parts all over, his face deforms, and the blanket is partially alive.

More ›