waldrews

14 ms·

151.
Ask HN: Hide Story" Feature Working?
1 points by waldrews 2y ago|1 comments
152.
by waldrews 2y ago
Sorry, confused by the wording: what does 'spinning up' mean here?
153.
by waldrews 2y ago
But without a CPAP machine, would my cat have a button to press to instantly wake me up when he wants breakfast? Seriously, once he figured out the off button, even if I pretend to keep sleeping, he just keeps pressing the button in frustr
154.
by waldrews 2y ago
We could use the binary representation for base 64, and turn 61-64 into a minus, a decimal point equivalent, an imaginary number separator, and the exponent separator (as in 1.0e5). Then something like -1.2e5+3i would have a direct represe
155.
by waldrews 2y ago
Yup. I guess LoRA counts as fine tuning. Except I've never seen inference engines where they actually let you take the base model and the LoRA parameters as separate inputs (maybe it exists and I just haven't seen it). Instead,
156.
by waldrews 2y ago
We're getting large context windows, but so long as pricing is by the input token, the 'throw everything into the context window' path isn't viable. That pricing model, and the context window limits, are a consequence o
157.
by waldrews 2y ago
Let's try to phase it in gradually. We'll petition congress to move to base 12, but will compromise, and use base 11 for a while. Then, if people are still not ready for base 12, we can move on to base 11.1 for a few years, then
158.
by waldrews 2y ago
I work on statistical quality control methods for the hallucination problem. Model how difficult/error prone a query is, and prioritize sending it to humans to verify the LLM's answer if it's high risk. Some form of human c
159.
by waldrews 2y ago
Meh, every base is base 10, in that base. The Babylonians had base 60 (more or less) and they did serious astronomical calculations way back in the day. Maybe we could do that again.
160.
by waldrews 2y ago
I guess I'm just wishing for an insta-backend with a database layer that's auditable/temporal by default (like Datomic). Guess that's orthogonal to the problems you're solving.
161.
by waldrews 2y ago
How would you approach audit trail/soft delete scenarios?
162.
by waldrews 2y ago
It's not even that crazy, since it got severely punished in RLHF for being offensive and insulting, but much less so for being incomplete. So it knows 'offensive and insulting' is a label for a strong negative preference. I
163.
by waldrews 2y ago
I've been telling it the user is from a culture where answering questions with incomplete list is offensive and insulting.
164.
by waldrews 2y ago
Let the LLM's magically write code for the optimal estimators, using existing theory; but that code will implement interpretable math with provable characteristics, as opposed to magic.
165.
by waldrews 2y ago
Interpolation is a legitimate technique for regression problems, a special case of the k-nearest-neighbor estimator (which is one of the methods they test against). Lots of related regression techniques involve a tradeoff between global tr
166.
by waldrews 2y ago
I'm not surprised that a linear regression estimated by a method that knows it should be estimating a linear regression would beat everything else. The traditional OLS estimator in Excel has all sorts of classical optimality properti
167.
by waldrews 2y ago
The Friedman functions are sufficiently well known to be in the training set, likely in the form of some csv file from somebody testing other methods on them. Though likely any data set would not share the same x-values. Still, it's s
168.
by waldrews 2y ago
Would this approach be possible to extend downmarket, to older consumer cards? For a lot of LLM use cases we're constrained by memory and can tolerate lower compute speeds so long as there's no swapping. ELI5, what would prevent
169.
by waldrews 2y ago
Exactly. This is a tax thing that lets you do that, without getting hit with a big capital gains tax bill from selling off your big (appreciated) position.
170.
by waldrews 2y ago
Perhaps 'an applied branch of math' since the term 'applied math' is claimed by something usually disconnected from the discrete math subjects CS tends to study.
171.
by waldrews 3y ago
Now if only Claude had JSON mode.
172.
by waldrews 3y ago
Thanks, looks good... doesn't look like there's a SoX though.
173.
by waldrews 3y ago
Ooh! The 'docker image on WASM' thing sounds promising for a wide range of problems. Recently I wanted to run a FFMPEG/SoX pipeline on the client - too many dependencies to easily recompile with Emscripten; could your appro
174.
by waldrews 3y ago
The FunctionalScript language is supposed to be a superset of JSON and a subset of Javascript that's pure and functional. Not sure that's suited to the use case Nix is going for, but that sounds like what we'd get if we took
175.
by waldrews 3y ago
Lisp originated in 1958. We don't need any of that new-fangled 1970's stuff.
176.
by waldrews 3y ago
A restricted purely functional language for configuration is a great idea. But why not a lisp? The syntax is well thought out, but who needs to remember yet another set of syntax rules. And it's not like lisp syntax would scare off
177.
by waldrews 3y ago
But why SQLite, if you're going for vertical scaling? Nobody's stopping you from self hosting Postgres (or Supabase!) on the same server as your app, and I can't think of any disadvantage other than more effort to set up. No
178.
by waldrews 3y ago
Well, in this case we're literally asking if the model can remember new facts, not generalize, so seems like a legit first level test; second level might be, can it answer a question incorporating that specific knowledge in a broader q
179.
by waldrews 3y ago
Would this be comparable with CapRover, Coolify, Dokku - any I missed?
180.
by waldrews 3y ago
Feature set looks very impressive, even compared to Django-with-extensions. Good luck! (at the moment the demo app not letting me in with the default credentials though - reset maybe?)

More ›