New Proof Dramatically Compresses Space Needed for Computation
Scientific American · New Proof Dramatically Compresses Space Needed for Computation
Administered by:
#computation
0 posts0 participants0 posts today
Universal pre-training by iterated random computation
https://arxiv.org/abs/2506.20057
#HackerNews #Universal #pre-training #iterated #random #computation #AI #research #machine #learning
arXiv.orgUniversal pre-training by iterated random computationWe investigate the use of randomly generated data for the sake of pre-training a model. We justify this approach theoretically from the perspective of algorithmic complexity, building on recent research that shows that sequence models can be trained to approximate Solomonoff induction. We derive similar, but complementary theoretical results. We show empirically that synthetically generated data can be used to pre-train a model before the data is seen. We replicate earlier results that models trained this way show zero-shot in-context learning across a variety of datasets, and that this performance improves with scale. We extend earlier results to real-world data, and show that finetuning a model after pre-training offers faster convergence and better generalization.
Several of the LLMs have produced inaccuracies which have been uncritically communicated to our customers by CrowdStrikers who failed to exhibit due diligence. Those errors were caught by said customers, and they were embarrassing to us all.From @brianmerchant@mastodon.social 's latest newsletter: https://www.bloodinthemachine.com/p/how-ai-is-killing-jobs-in-the-tech
...
Now we have an engineer, if you can call him that, working on a project that will introduce more than 30k lines of AI generated code into our codebase, without a single unit test. It will be impossible to do a proper code review on this much code and it will become a maintenance nightmare and possibly a security hazard. I don't need to tell you how much management is cheering on that.
A ticking timebomb in the making. It's especially galling that CrowdStrike is doing this, given their epic fail just last year.
A while back I wrote in a post here:
under Taylorism the workers who actually do the work and know it best no longer have a say (opinion) in how that work gets done. Pseudo-scientific principles (scientific management, the astrology of MBAs) dictates all. Computers, from the very first, were intended and designed for this purpose.riffing on what a lousy person Charles Babbage was and the lousy anti-worker plans he had for the proto-computers he designed. Among other things generative AI is another manifestation of the MBA pseudoscience known as scientific management and exists in a long line of digital technologies stretching all the way back to Babbage's.
#AI #GenAI #GenerativeAI #AgenticAI #tech #dev #computation #ComputerScience #labor #organizing
Blood in the Machine · How AI is killing jobs in the tech industry
"it's interesting how many things in computation follow this pattern"
X (formerly Twitter)Aleph (@woke8yearold) on Xit's interesting how many things in computation follow this pattern
#OnThisDay Birth Anniversary of Alan Turing (1912) - highly influential in the development of theoretical computer science, providing a formalisation of the concepts of #Algorithm and #Computation.
Christoper Lotham Sholes received patent for Typewriter (1868).
Happy Birthday Zinedine Zidane (1972).
Today is UN Public Service Day, International #WidowsDay.
Replied in thread
I have a feeling that the study of systems composed of feedback loops (all of creation, I dare say!) will be of interest to you.
I understand it is called "cybernetics".
Please find attached an amusing 1949 Letter from the US Library of Congress to Norbert Weiner, asking him in which section his book "Cybernetics" is supposed to be.
Peano arithmetic is enough, because Peano arithmetic encodes computation
Mathematics Stack ExchangeCan PA prove "each Goodstein sequence can be proven in PA to reach zero"?This is one of a pair of questions trying to understand this comment on the xkcd forum contest My number is bigger than yours!. For a definition of Goodstein sequences, see this question.
Let $G(n)...
Today, I discovered that Postgres materialized views do _not_ inherit their parent table's indexes. This is why my spatial query is still running after 24hours.. 9000 × 50,000,000 = 450 billion geometry comparisons. Doh!
Continued thread
♬ 
“Astonishing #discovery by computer #scientist: how to squeeze space into time”
#KelseyHoustonEdwards PhD / #ChalkTalk / #computation / <https://youtube.com/watch?v=8JuWdXrCmWg>
AI: Cure All Diseases - Eric Schmidt on AI
"Rather than rely on clever computation to make the best of conventional camera hardware, some engineers are instead experimenting with optical components such as new apertures, and animal-inspired sensors, that together can gather high-quality visual data that requires less intensive processing."
www.nature.comHow the natural world is inspiring the robot eyes of the futureIntelligent machine vision could get a boost from unconventional hardware design.
My CPU is producing a lot of heat in recent days. Most of the time I'm doing #wakegp experiments to fine tune parameters. And I'm also compiling various versions of #Luanti to find out or fight with regressions and report them upstream.
I'm just glad I invested into buying this CPU plus a decent AIO. I can compile debug enabled Luanti from scratch in just about 5 minutes. Interestingly, compiling a debug enabled wakegp with incremental builds takes at least the same amount or even more.
wakegp is written in Rust, Luanti is written in C++. Luanti is a far larger codebase. But I guess the problem's that many of Luanti deps are shared libraries so I don't have to do the compiling or linking for them.
TOSID Update:
Added formal paper presentation
https://github.com/ha1tch/TOSID/blob/main/tosid-paper.md
As PDF:
https://github.com/ha1tch/TOSID/blob/main/tosid-paper.pdf
The TOSID (Taxonomic Ontological Semantic IDentification System) is a comprehensive taxonomic framework designed to uniquely identify and classify any conceivable entity across the universe. The system uses a structured alphanumeric code that embeds hierarchical information about an entity's fundamental nature, origin, scope, and specific identity.
Latest working (non-paper) version:
https://github.com/ha1tch/TOSID/blob/main/tosid-01.md
GitHubTOSID/tosid-paper.md at main · ha1tch/TOSIDContribute to ha1tch/TOSID development by creating an account on GitHub.
New Project!
The TOSID (Taxonomic Ontological Semantic IDentification System) is a comprehensive taxonomic framework designed to uniquely identify and classify any conceivable entity across the universe. The system uses a structured alphanumeric code that embeds hierarchical information about an entity's fundamental nature, origin, scope, and specific identity.
GitHubTOSID/tosid-01.md at main · ha1tch/TOSIDContribute to ha1tch/TOSID development by creating an account on GitHub.
X X^t can be faster
https://arxiv.org/abs/2505.09814
#HackerNews #X #X^t #can #be #faster #arxiv2505 #Xmatrix #computation #optimization #speedup
arXiv.org$XX^{t}$ Can Be FasterWe present a new algorithm RXTX that computes product of matrix by its transpose $XX^{t}$. RXTX uses $5\%$ less multiplications and additions than State-of-the-Art and achieves accelerations even for small sizes of matrix $X$. The algorithm was discovered by combining Machine Learning-based search methods with Combinatorial Optimization.
Why agency and cognition are fundamentally not computational
https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2024.1362658/full
FrontiersFrontiers | Naturalizing relevance realization: why agency and cognition are fundamentally not computationalThe way organismic agents come to know the world, and the way algorithms solve problems, are fundamentally different. The most sensible course of action for ...
Cardiac: A CARDboard Illustrative Aid to Computation [pdf]
In short, intuition, insight, and creativity are not computable.
—Aloisius H. Louie, More Than Life Itself
#computability #computation #abduction
—Aloisius H. Louie, More Than Life Itself
#computability #computation #abduction
…what if life itself is, even in principle, nonsimulable? There are, of course, many things that mechanization by rote does better than life, in terms of speed, repeatability, precision, and so forth. On the other hand, a living system…may be characterized by its ability to handle ambiguities and take chances, indeed, its ability to err. These are precisely the processes that cannot, by definition…, be modelled algorithmically.
—Aloisius H. Louie, More Than Life Itself: A Synthetic Continuation in Relational Biology
#life #computability #computation #algorithms
—Aloisius H. Louie, More Than Life Itself: A Synthetic Continuation in Relational Biology
#life #computability #computation #algorithms