thesz

11 ms·

181.
by thesz 9mo ago
Catastrophic forgetting is overfitting.
182.
by thesz 9mo ago
As far as I remember, article stated that he found same problematic behavior for many prompts, issued by him and his colleagues. The "stupid prompt" in article is for demonstration purposes.
183.
by thesz 9mo ago
>>> It is a very contrived under-specified prompt. No True Prompt can be such contrived and underspecified. The article about degradation is a case study (single prompt), weakest of the studies in hierarchy of knowledge. Ca
184.
by thesz 9mo ago
> I think the missing bit here is that this only works for cases where there's a really large test set (the html spec, the linux kernel). I'm not convinced that the models would be able to maintain coherence without this,
185.
by thesz 9mo ago
There is tinycc, that makes it three compilers. There is a C compiler implemented in Rust from scratch: https://github.com/PhilippRados/wrecc/commits/master/?after=... (the very beginning of commit histo
186.
by thesz 9mo ago
> Yeah, this is a bullshit article. There is no such degradation, and it’s absurd to say so on the basis of a single problem which the author describes as technically impossible. It is a very contrived under-specified prompt. I s
187.
by thesz 9mo ago
> No we don't and yeah we would expect them to only report positive results (this is both marketing and investigation). This is matter of methodology. If they train models on that task or somewhat score/select models on
188.
by thesz 9mo ago
> It's still really, really impressive though. Do we know how many attempts were done to create such compiler before during previous tests? Would Anthropic report on the failed attempt? Can this "really, really impressi
189.
by thesz 9mo ago
> As tweaks are made... > ...how you layout different files and functions in memory have different trade offs and are hard to know up front without doing actual real world testing. These are definitely not an algorithmic opt
190.
by thesz 9mo ago
> the huge gains in coding performance in the past year have come from RL, not from new sources of training data. This one was on HN recently: https://spectrum.ieee.org/ai-coding-degrades Author attributes past y
191.
by thesz 9mo ago
We are talking about compiler here and "performance" referred above is the performance of generated code. When you are optimizing a program, you have a specific part of code to improve. The part can be found with profiler. When yo
192.
by thesz 9mo ago
> This test sorta definitely proves that AI is legit. This is an "in distribution" test. There are a lot of C compilers out there, including ones with git history, implemented from scratch. "In distribution" t
193.
by thesz 9mo ago
SQLite was a library for TCL, implementing "sqlite" command, and from TCL one can use it in a variety of interesting ways, approaching and even surpassing "stored procedures" concept. For example, it is easy to create SQ
194.
by thesz 9mo ago
This is a story of how one became better golf player by increasing his strength: https://www.youtube.com/watch?v=sr2pgBTRpK4 One can enhance cognitive functions by strength training: https://pmc.ncbi.nlm.nih.gov&
195.
by thesz 9mo ago
> There is research that shows that banning calculators impedes the learning of maths. Please share what you know. My search found a heap of opinions and just one study where use of calculators made children less able to calculat
196.
by thesz 9mo ago
You can model systems in your mind using past experience with different systems, reliably and repetetively.
197.
by thesz 9mo ago
https://www.youtube.com/watch?v=8kUQWuK1L4w APL inventor says that he was developing not a programming language, but notation to express as much problems as one can. He found that expressing more and more problems with the
198.
by thesz 9mo ago
> When you know half a dozen programming languages or web frameworks, the next one takes you a couple hours to get comfortable with. Learn yourself relational algebra. It invariantly will lead you to optimization problems and the
199.
by thesz 9mo ago
VLDL, a precursor for LDL, is produced in liver. Both are more or less the same chemically, but differ in the amount of fat carried. LDL is VLDL but somewhat processed by body, HDL is a VLDL (LDL) completely processed by body. Bile is used
200.
by thesz 9mo ago
But why?
201.
by thesz 9mo ago
I was on the other side, I thought types are indispensable. And I still do. My own experience suggest that if you need to develop heavily multithreaded application, you should use Haskell and you need some MVars if you are working alone and
202.
by thesz 9mo ago
If you do a single set of half of exercises you need to train each day of the week, rotating these halves, you get 3 and a half sets of each exercise per week. Training volume of Bulgarian Method is not much bigger than that of regular trai
203.
by thesz 9mo ago
Bulgarian Method does not have rest days: https://www.mashelite.com/the-bulgarian-method-is-worth-a-lo... Alexander Zass (Iron Samson) also trained each day: https://en.wikipedia.org/wiki/Alexander_Zass
204.
by thesz 9mo ago
> AI assisted development isn't all or nothing. > We as a group and as individuals need to figure out the right blend of AI and human. This is what makes current LLM debate very much like the strong typing debate about
205.
by thesz 9mo ago
Weightlifting and weight training was invented long before forklifts. Even levers were not properly understood back then. My favorite historic example of typical modern hypertrophy-specific training is the training of Milo of Croton [1]. By
206.
by thesz 9mo ago
This one was on HN recently: https://spectrum.ieee.org/ai-coding-degrades The quality of code generated with LLM decreases over time. Let us talk again after, say, five years.
207.
by thesz 9mo ago
> There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. This is exactly my experience with using Haskell after having c
208.
by thesz 9mo ago
> That's when you start *learning* the techniques for using these tools for maximum effect and where to apply them. https://paulgraham.com/popular.html?viewfullsite=1 "They're perfectly justified:
209.
by thesz 9mo ago
The paper about type classes [1] includes exceptions as an example. [1] https://web.cecs.pdx.edu/~mpj/pubs/springschool95.pdf This [2] paper generalizes exceptions, allowing them to be first class citizens of a la
210.
by thesz 9mo ago
> The `Either a b` type in Haskell is equivalent to the `Result<T, Error>` type in other languages. The only difference is in the naming: This is false and misleading. The Result <E, T> usually (C#, at the very least,

More ›