usgroup

17 ms·

181.
by usgroup 3y ago
I've spent a lot of time in academia up to PhD level. Personally I'd suggest you start by doing the Maths A-Levels. There is lots of solid material / books, there a lots of tutors if you need them, and you have exams to prove
182.
LLMs and recursive reduction in output diversity [video](youtube.com)
2 points by usgroup 3y ago|0 comments
183.
by usgroup 3y ago
Google/Bing/Yahoo/Yandex/etc search and Meta/Google/AMZ/Linkedin engagement algos seem like fairly clear front runners for world changing ML projects, but very many others come to mind too.
184.
1BRC in Awk, Haskell and Fortran(github.com)
3 points by usgroup 3y ago|0 comments
185.
by usgroup 3y ago
I think the skill-level required to start grinding boosted trees is relatively low compared to NNs. There's lots you can do without special hardware. Its more democratic. It trains very quickly compared to NNs. It works for big and sma
186.
by usgroup 3y ago
You have to learn to maths and to code, that takes about a decade. It requires a lot of output from you. There’s no way around that. Even if it was all redundant due to AI , it is essential mental scaffolding for higher thought unreachable
187.
by usgroup 3y ago
I'm not sure that is true. I think inference speed is often the bottleneck for the use cases stated, as is the need for frequent re-training. As a result algorithms like catboost are very popular in those domains. I think catboost was
188.
by usgroup 3y ago
I'm not sure this is surprising. Say you were to glue together 10 datasets with the same 10 explanatory features and 1 response feature, but distributed very differently to each other. This would be no problem for tree based model beca
189.
by usgroup 3y ago
I think its more intuitive for statistical applications where Python is grossly under-represented. This includes things like the design and analysis of experiments but also lots of domain specific statistics and algorithms such as in bioinf
190.
Some Data Science with Haskell(github.com)
5 points by usgroup 3y ago|0 comments
191.
by usgroup 3y ago
This is well put. Coincidentally in the example the results are the same , but they need not be. given repeated experiments with the same intentions one may expect different distributions. However, one could just move the argument up a leve
192.
by usgroup 3y ago
Love it: p(I saw E) and p(I didn’t really see E). Just move the argument one level down: “I saw E is false” and it turns out so is “E is false” . So then? Add “E was false was false”? Turtles all the way down. At some point something has to
193.
by usgroup 3y ago
Yeah I’d agree at some depth. We don’t talk enough about integers, rationals and real numbers and what they imply for our “normative rationality” or “epistemological commitment”. But aside from the integers, everything else is totally suspi
194.
by usgroup 3y ago
Well no because it’s talking about either a fixed sample size or stopping when a % total is reached. Neither imply a favourable p-value necessarily. I think the author means to say that it’s two methods incidentally equivalent in the data t
195.
by usgroup 3y ago
Yeah generally Jaynes book is very nice and easy to read for this sort of material.
196.
by usgroup 3y ago
So you know when you believe something and then you update your belief because you get some evidence? Yeah, and then you stack some beliefs on top of that. And then you discover the evidence wasn’t actually true. Remind me again what the no
197.
by usgroup 3y ago
I learned Haskell recently, and I thought ChatGPT was a great support for that effort. It helped me with definitions, standard recipes and disambiguation. What it wasn't very good at was coding: it couldn't solve anything I did no
198.
by usgroup 3y ago
I'd -- as usual -- suggest Prolog. Its elegant, easier to understand and more mature. It comes with "batteries included" if you want constraint solving over finite domains. Besides that MiniZinc is a phenomenal interface to a
199.
by usgroup 3y ago
GLMs are a non linear transformation on an output followed by linear modelling. They are referred to as “linear models” but you might as well then also consider NNs as linear models, or any model at all which ends with addition as the final
200.
by usgroup 3y ago
I'm not sure this argument is in any way specific to LLMs, and the space for their application is still enormous. Search results, ad targeting, recommendation systems, anomaly detection, content flagging, and so on, are all systems usi
201.
by usgroup 3y ago
Some of the comments reminded me of LeCun's claim regarding the error distribution of an LLM output conditional on content length. Namely, if "e" is the probability of an error, the probability of a sequence of length "n
202.
by usgroup 3y ago
I expect this is particularly borne out when there is no obviously good move, but many options. The grandmaster will "see" implications without calculating and then confirm them. I think most strong chess players can "evaluat
203.
by usgroup 3y ago
I don't think you've understood what I meant by "flattening". I wanted to imply something like an embedding of search paths into a high dimensional space within which a point in the space represents a whole path, and the
204.
by usgroup 3y ago
Grandmasters do search! They think many moves ahead for most moves, and obviously StockFish does search -- a lot of search, much more than a grandmaster. I feel that the sort of structures we implicitly operate over during search can be use
205.
by usgroup 3y ago
"A non-existent amount of mathematics" -- which the joke could have been interpreted as implying -- is not zero mathematics: that would be an existent amount. Another place these semantics arise is in talk about probability and po
206.
by usgroup 3y ago
Pedantry on my part, but would you say that a non-existent unicorn weigh some amount "None"? That'd be strange to me, and "no amount" would describe such a situation.
207.
by usgroup 3y ago
Ah good, because the sort of rubbish I suspect OP is referring to also usually contains "no amount" of mathematics too :)
208.
by usgroup 3y ago
This rings true in my experience too. It isn't even artisanal code lacking an end user, its garbage code that barely works which no-one wants. Luckily I've spent most of my time in the start-up space where things working is a more
209.
by usgroup 3y ago
I get this now but certainly didn't 20 years ago, and I suspect it will more easily register with functional/logical programming language users. I've heard it referred to as "wholemeal" programming. Take the first s
210.
What is good about Haskell?(doisinkidney.com)
5 points by usgroup 3y ago|0 comments

More ›