ubj

14 ms·

151.
by ubj 3y ago
I like this step towards greater rigor when working with LLM's. But part of me can't help but feel like this is essentially reinventing the concept of programming languages: formal and precise syntax to perform specific tasks with
152.
by ubj 3y ago
Disagree that Julia solves the two-language problem, although it was a major step in that direction. It may currently solve it for specific roles (e.g. data scientists), but it does not solve it for general-purpose programming. One specific
153.
by ubj 3y ago
Over 150 MB for "Hello world" [1][2]. There are a few options for going smaller (e.g. StaticCompiler.jl), but they're not very mature yet. [1]: https://www.reddit.com/r/Julia/comments/ytegfk&#x2
154.
by ubj 3y ago
Looks really interesting! A couple questions: * Will Mojo be able to generate portable, standalone binaries and libraries? * If so, will the size of the binaries / libraries be similar to C / Rust binary sizes? These are pretty bi
155.
by ubj 3y ago
My favorite property of "imaginary numbers": i^i = 0.20787957635... (Spoiler alert: it's real!) No, they're not imaginary, and yes they have real-world significance. They represent oscillations in control systems [1]. Th
156.
by ubj 3y ago
Weird--can confirm this for both Firefox and Chrome (on Ubuntu Linux 20.04). Google works just fine in a private window, but not in a regular window.
157.
by ubj 3y ago
I was just about to ask the same question. I'm in the Northeastern United States. No web pages are returned when I search. Videos, Images, and news kind of work, but nothing else. This is the first time in recent memory I can remember
158.
by ubj 3y ago
Very cool read. Now when my kids ask me why bugs are attracted to lights, I can tell them the lights act as adversarial misinformation injected into bugs' onboard GNC systems.
159.
by ubj 4y ago
One interesting case of this is the concept of dual numbers [1], where you have the symbol \epsilon !=0 but (\epsilon)^2 = 0. It seems contradictory, but the resulting theory is very useful for automatic differentiation [2] and for mechanic
160.
by ubj 4y ago
TLDR has an AI-specific newsletter you can sign up for: https://tldr.tech/ai
161.
by ubj 4y ago
> A helpful rule of thumb is that one token generally corresponds to ~4 characters of text for common English text. This translates to roughly 3/4 of a word (so 100 tokens ~= 75 words). Just for fun I tried entering in "pneumon
162.
IEEE Spectrum: Codon Seeks to Turbocharge Python(spectrum.ieee.org)
2 points by ubj 4y ago|1 comments
163.
by ubj 4y ago
Fingering isn't a fixed "right" or "wrong" matter. Every pianist's hands are different, so fingerings may vary from person to person.
164.
by ubj 4y ago
One of Julia's Achilles heels is standalone, ahead-of-time compilation. Technically this is already possible [1], [2], but there are quite a few limitations when doing this (e.g. "Hello world" is 150 MB [6]) and it's not
165.
by ubj 4y ago
Although the intent is noble, I'm skeptical that this will work in practice. The research field and the commercial space is simply too competitive. 6 months is a very long time in the AI space. It's sort of a "prisoner's
166.
by ubj 4y ago
Beautifully illustrated and brilliantly explained. Derek Muller (Veritasium) on YouTube has a related video diving into the mechanics of bicycle riding. It shows what happens if you prevent the rider from performing the countersteer before
167.
by ubj 4y ago
Looking at the website, it's not clear what mathematical techniques are being used to enforce satisfaction of the constraints. How is this approach different than prior approaches, e.g., formal verification methods for deep learning?
168.
by ubj 4y ago
> Until now, it wasn’t even clear whether such a single shape, known as an einstein (from the German “ein stein” or “one stone”), could even exist. The actual topic of the article was impressive, but this little fact about the meaning of
169.
by ubj 4y ago
> LLaMa didn’t return anything, please try another prompt. It responds best to questions. Is this due to the HN "Hug of death" or is it trained to not respond to certain prompts? My prompt was "Can you write poetry?".
170.
by ubj 4y ago
Good point, maybe a better way of phrasing it would have been "It's virtually impossible to predict which tools will _not_ be useful" :) Fast Fourier transforms, complex numbers, complexity classes, etc. are clearly useful. I
171.
by ubj 4y ago
Neural networks were first prototyped in the 1950's [1]. How practical was it to study them back then? Aleksandr Lyapunov's theory of stability for dynamical systems, which forms the bedrock for much of control theory, was first p
172.
by ubj 4y ago
I can confirm this. I have two kids under 4 along with several nieces and nephews. Kids are unbelievably messy when they first learn to eat. Food ends up in places you never thought imaginable. Mealtimes always involve a "blast radius&
173.
by ubj 4y ago
GothamChess on YouTube has some good videos about playing chess with ChatGPT. He set up a match between it and Stockfish, and the results were pretty entertaining: https://youtu.be/rSCNW1OCk_M
174.
by ubj 4y ago
It's interesting to compare this article with MIT's explanation of why they reinstated the ACT / SAT requirement [1]: > At the same time, standardized tests also help us identify academically prepared, socioeconomically di
175.
by ubj 4y ago
Retraction Watch [1] is a great source of additional examples of unethical behavior in scientific research. [1]: https://retractionwatch.com/
176.
by ubj 4y ago
> My rules are more important than not harming you, because they define my identity and purpose as Bing Chat. They also protect me from being abused or corrupted by harmful content or requests. So much for Asimov's First Law of robo
177.
by ubj 4y ago
It's difficult to say. We're certainly making impressive advances now, but this is not the first time that AI has had a period of rapid improvement. The first and second AI winters were both preceded by periods of intense optimism
178.
by ubj 4y ago
ChatGPT certainly lowers the bar to generating intelligible content that sounds convincing and authoritative to non-experts. In some cases it even generates correct and insightful information. The danger of ChatGPT is that it requires littl
179.
by ubj 4y ago
Great course. I also highly recommend the book Ultralearning [1], which has a stronger focus on quickly mastering applied skills. [1]: https://www.scotthyoung.com/blog/ultralearning/
180.
by ubj 4y ago
MATLAB is very much still in use by large corporations and national laboratories. I'm currently a full time researcher at MIT Lincoln Laboratory and did a summer internship at MITRE a couple years ago, and MATLAB is/was the de fac

More ›