thesz

8 ms·

121.
by thesz 7mo ago
Erlang has "die and be restarted" philosophy towards process failures, so these "bugs that happen to erlang systems in prod" may not be fixed at all, if they are rare enough.
122.
by thesz 7mo ago
> until all potential sources of error are close to being eliminated This is what PSP/TSP did - one has to (continually) review its' own work to identify most frequent sources of (user facing) defects. > if you
123.
by thesz 7mo ago
https://hackage.haskell.org/package/spectacle "Spectacle is an embedded domain-specific language that provides a family of type-level combinators for authoring specifications of program behavior along with a model
124.
by thesz 7mo ago
Time to completion (in appendix A9) should be treated as log-normally distributed, or by some other one-sided distribution because one cannot complete the task faster than 0 seconds. This transformation will rule out confidence ranges with
125.
by thesz 7mo ago
> We’re heading for a world of terrible code that can only be maintained by extremely good coding agents and are pretty much impossible for a human to really understand. I once figured out the algorithm of the program written in
126.
by thesz 7mo ago
Thanks. My memory failed me, I've implemented MIPS 18 years ago.
127.
by thesz 7mo ago
HashCash.
128.
by thesz 8mo ago
MIPS code is not much larger. There are two interesting differences of ISA between MIPS and RISC-V: that MIPS does not have branch on condition, only on zero/non-zero and that MIPS has 16 bit immediates with appropriate sign extension
129.
by thesz 8mo ago
An old paper on the expressiveness of the programming languages [1] had to add an implicit binary operator into whitespace to make Haskell not be ten times more expressive than most imperative languages. [1] https://www.researchg
130.
by thesz 8mo ago
> how to discover unsolved problems There are plenty of them and LLMs will not help you much there because unsolved problems are, by definition, out-of-distribution samples. Neural networks are interpolators, not extrapolators.
131.
by thesz 8mo ago
This looks very much like whirlpool. LLM researcher makes LLMs researching LLMs. The quote from old post from Karpathy [1] look very appropriate here [1] https://karpathy.github.io/2015/05/21/rnn-effectiveness
132.
by thesz 8mo ago
I speak Russian and some English, but the question was about universal quantification: author declares that LLMs generate code of better quality than "any codes" he seen in his career.
133.
by thesz 8mo ago
The question was about universal quantification, not grammar error. As if author of the comment had not seen any code that is better or of equal quality of code generated by LLMs.
134.
by thesz 8mo ago
I guess you can guide it to write in any style. But what set me off is an universal qualifier: there was no code seen by you that is of equal quality or better that what LLMs generate.
135.
by thesz 8mo ago
> LLM code is higher quality than any codes I have seen in my 20 years in F500. "Any codes"?
136.
by thesz 8mo ago
Exercise rises CO2 levels in blood and there are specific exercises to increase CO2 tolerance. Also, extra ventilation during very long exercises (hours) lowers CO2 blood level. As the recovery from aerobic and resistance exercises also inc
137.
by thesz 8mo ago
> So NNs too really (damn autocorrect, it's rely, rely) on human insight and structuring upon which they can then build over. If so, one can repeat it with other basic models as well. It is possible to train deep decision tr
138.
by thesz 8mo ago
> Ironically, NN can detect such latent features Like they detect magazine logo on images of horses?
139.
by thesz 8mo ago
I read hundredths (100ths) of lines of code for one line of code I write. My last PR of three lines of code moved into conditional tasked me to read about 8000 lines of code to understand and justify the reason to do exactly that.
140.
by thesz 8mo ago
> but I think it's unfair to say that LLMs _cannot_ recognize or generate CFGs. They recognize and/or generate finite (<800 chars) grammars in that paper. Usually, sizes of files on a typical Unix workstation follow
141.
by thesz 8mo ago
And this Figure 12 is not about Dyck/balanced-brackets grammar. This figure is about something not properly described in the paper.
142.
by thesz 8mo ago
> it actually does seem to be 100% For all Dyck grammar sequences, infinitely many of them? ;)
143.
by thesz 8mo ago
Most of the software engineers out there do the support, augmenting source code behemoths the least possible way to achieve desired outcome. I believe that more than 90% of software development was support roles as early as 2K or so. Not th
144.
by thesz 8mo ago
Dyck grammar (balanced brackets) are not an a^nb^n, there are several kinds of brackets. I cannot find probability of success in paper you linked. Is it 100%? I believe it is less than 100%, because LLMs are intrinsically probabilistic mach
145.
by thesz 8mo ago
Transformers are able to recognize balanced brackets grammar at 97% success rate: https://openreview.net/pdf?id=kaILSVAspn This is 3% or infinitely far away from the perfect tech. The perfect tech is the stack.
146.
by thesz 8mo ago
If I need to change someone's code, I also read it. several times.
147.
by thesz 8mo ago
> I do read the code, but reviewing code is very different from producing it, and surely teaches you less. If you don’t believe this, I doubt you work in software. I work in software and for single line I write I read hundredths
148.
by thesz 8mo ago
> 39% adoption in two years (internet took 5, PCs took 12). Adjust for connectivity and see whether it is different (from pure hype) this time.
149.
by thesz 8mo ago
An old one: https://news.ycombinator.com/item?id=33755016 Stylometry can match not only people, but ethnic groups. No LLM required.
150.
by thesz 8mo ago
You said AI: https://github.com/stassa/louise Louise (Patsantzis & Muggleton 2021) is a machine learning system that learns Prolog programs. Louise is a Meta-Interpretive Learning (MIL) system. MIL (Muggleto

More ›