nyrikki

22 ms·

271.
by nyrikki 1y ago
Published, and if it holds...Ouch! > Our empirical results show that the perceived reasoning abilities of LLMs stem from the exemplar-query similarity and approximate retrieval rather than any inherent reasoning abilities, thereby leadin
272.
by nyrikki 1y ago
Those projects seem to devolve into a boil the ocean style projects and tend to be viewed as intractable and thus ignorable. In the days everything was http I use to set a proxy variable and have the proxy save all downloaded assets to comp
273.
by nyrikki 1y ago
Structure composition is a form of inheritance.
274.
by nyrikki 1y ago
While there are real use cases for the above, if you are looking at bloom filters make sure you check the above works for your need. First-order with least fixed point FO(LFP) == P As P=co-P but we think NP!=co-NP, bloom filters often have
275.
by nyrikki 1y ago
This paper works because it explicitly is a problem domain that was intentionally constrained to ensure safety in the Amateur high-power rocket hobby. Specifically with constraints and standards that were developed for teenagers of vario
276.
by nyrikki 1y ago
> What precisely is meant by simulating the laws of physics? IIRC That is in reference to the work around the Church–Turing–Deutsch principle which just classic physics. Basically Turing machines only have access to the computable real
277.
by nyrikki 2y ago
> But in the AI's defense, they have a point: You never specified if the towels can be dried simultaneously or not. Maybe you have to use a drying machine that can only do one at a time. This one seems to consistently work: This is
278.
by nyrikki 2y ago
The original paper about the bias variance tradeoff, that the double decent papers targeted, had some specific constraints. 1) data availability and computer limited training set sizes. 2) they could simulate infinite datasets. While challe
279.
by nyrikki 2y ago
You don't avoid it, you realize that there are fundamental limits and try to find ways to still get work done. While you can't have the territory road, plat, and topographical maps may be incomplete, but all have their uses.
280.
by nyrikki 2y ago
While I am to old to say what is taught today, this is exactly the map vs territory issue I mentioned in my other comment. It is all there in what you would have been taught, but hidden because we tend to avoid the hay in the haystack probl
281.
by nyrikki 2y ago
One of the huge sources of map/territory relation issues is that we use systems that let us avoid the problem discussed here Definitely worth spending time on.
282.
by nyrikki 2y ago
Avego wanted PLX switches for enterprise storage, not low margin PC/server sales. Same thing that Avego did with Broadcom, LSI, Brocade etc... during the 2010's, buy a market leader, dump the parts that they didn't want, leav
283.
by nyrikki 2y ago
While probably not useful for everyone, the best method for myself actually leverages that. I am using a modified form of TDD's red/green refactor, specifically with an LLM interface independent of my IDE. While I error on good co
284.
(How) Do reasoning models reason?(arxiv.org)
2 points by nyrikki 2y ago|0 comments
285.
by nyrikki 2y ago
A reminder, for those who find TAOCP useful, Knuth released "Volume 4 Fascicle 7, Constraint Satisfaction" last month. It connected a lots of dots for me, but like most of TAOCP may/ or may not be useful standalone depending
286.
by nyrikki 2y ago
The problem is that topics like descriptive complexity theory are typically reserved for graduate level courses, and what the author has access to is basically a survey. There are entire hierarchies that describe what is "NP-Harder&quo
287.
by nyrikki 2y ago
Note this paper that will be presented at ICSE in a couple of weeks too. https://arxiv.org/abs/2408.02487v3 The point being is that this is very much a very real and yet unsolved problem with LLMs right now.
288.
by nyrikki 2y ago
Funny, gemma will do this all day long. >>> Please write a hello world app in java ```java public class HelloWorld { public static void main(String[] args) { System.out.println("Hello, World!
289.
by nyrikki 2y ago
From a CS background, several people I know have raved about the following book[1], of which will be friendly and useful for future needs anyway in the field. The first part of the book is what appears to be a pretty good refresher path. IM
290.
by nyrikki 2y ago
You are assuming input-transformation based defenses in the image domain transfer to the music recognition domain, when we know they don't automatically even transfer to the speech recognition domain. But 'protection' of any
291.
by nyrikki 2y ago
To add to this. Both the Bayesian vs frequentist interpretations make understanding the problem challenging, as both are powerful interpretations to find the needle in the haystack, when the problem is finding the hay in the haystack. A bet
292.
by nyrikki 2y ago
The problem is that a common method is to seed the linkers symbol generation with a relative file name. This reduces entropy across binaries and may enable reliable detection of bas addresses or to differentiate gadgets in text regions. It
293.
by nyrikki 2y ago
My primary task isn't ML, and 64bit is needed for numerical stability. For the Titan V, the F64 was 1/2 of F32, it was the only and last consumer generation to have that. For Titan RTX and newer NVIDIA cards, the ratio from F32
294.
by nyrikki 2y ago
I have a machine with 3 108 TI's that I do a batch with, sending the question first to a LLM and then an LRM, returning to review the faster results and killing the job if they are acceptable. Ollama or just llama.cpp on podman makes
295.
by nyrikki 2y ago
That is pretty much a universal problem. If you look at the problems anyone's models has solved, they are all well represented in the corpus. Remember that AIME is intended for high schoolers with just pencils, erasers, rulers, and c
296.
by nyrikki 2y ago
Cloudflare doesn't officially support SSE, but if you send keepalives events every 15 or 20sec or so you can reliably use SSE for 40 min + in my experiance. No server traffic for 100+ sec officially results in a 524, so you could p
297.
by nyrikki 2y ago
The Performant claim there is counter to research I have heard. Plus as the PGO profile data is non-deterministic in most cases, even when compiled on the same hardware as the end machine "Committing profiles directly in the source
298.
by nyrikki 2y ago
This is one of the "costs" of reproducible builds, just like the requirement to use pre-configured seeds for pseudo random number generators etc. It does hit real projects and may be part of the reason that "99%" is call
299.
by nyrikki 2y ago
The typical solution is to have the interpreter call a wrapper program or better yet use RBOC etc.. to remove the need for suid in the first place. Sudo, ..., group membership, even using udev to emit devices with those permissions automati
300.
by nyrikki 2y ago
You don't, privilege escalation is trivial so it will ignore suid/sgid, as an example with a shell in the shebang $ cd /tmp $ ln /path/to/setuid_script -i $ PATH=. $ -i Now

More ›