HNHacker News
TopNewBestAskShowJobs

dawnofdusk

353 karma · joined September 8, 2022

submissionscomments
dawnofdusk··on Why English doesn't use accents
>Similarly, Chinese and Korean names are usually written in the order they are pronounced, while Japanese names are reversed.

As a Chinese speaker this is maddeningly confusing when reading Western media. It's also a fairly new trend, I want to say a decade ago Chinese and Korean names were also read in Western order.

dawnofdusk··on Why English doesn't use accents
In French you can (and most people do) drop diacritics from all uppercase forms.
dawnofdusk··on Why English doesn't use accents
>but the spelling rules differ depending on whether the word was imported directly from Latin, or came in via Norman French

In fact it can be even more complicated because in English the words can come from Norman dialects and "typical" French simultaneously. For example, warden and guardian come from the same word in Old French, the former is closer to how Normans pronounce it and the latter is closer to its modern French pronunciation.

dawnofdusk··on Stalking the Statistically Improbable Restaurant with Data
Now that Google Maps has the AI summary for restaurants, I wonder if this can be queried in the API? It would probably have the keywords for Xinjiang food or whatever the OP wants to analyze. Checking two Central Asian restaurants I've been to in my area, one is tagged "Restaurant" and one is tagged "Pan-Asian Restaurant", so not very illuminating. However, in their AI summaries both have keyword "Central Asian" and one even says more specifically "Kyrgyz"
dawnofdusk··on Hilbert's sixth problem: derivation of fluid equations via Boltzmann's theory
The short answers:

1. It answers how macroscopic equations of e.g., fluid dynamics are compatible with Newton's law, when they single out an arrow of time while Newton's laws do not.

2. It was solved in the 1800s if you made an unjustified technical assumption called molecular chaos (https://en.wikipedia.org/wiki/Molecular_chaos). This work is about whether you can rigorously prove that molecular chaos actually does happen.

3. There are no applications outside of potentially other pure math research. For a physics/engineering perspective the whole theory was fine by assuming molecular chaos.

dawnofdusk··on The new skill in AI is not prompting, it's context engineering
>This really does sound like Computer Science since it's very beginnings.

Except in actual computer science you can prove that your strategies, discovered by trial and error, are actually good. Even though Dijkstra invented his eponymous algorithm by writing on a napkin, it's phrased in the language of mathematics and one can analyze quantitatively its effectiveness and trade-offs, and one can prove if it's optimal (as was done recently).

dawnofdusk··on Tools I love: mise(-en-place)
Yeah it's hard to get excited about these sorts of tools when Nix exists.

I think for development purposes it's a smart idea to use the language specific conventions/ecosystem.

dawnofdusk··on I deleted my second brain
This comment you wrote sounds a lot like an LLM, FYI. Not saying you are
dawnofdusk··on Reinforcement learning, explained with a minimum of math and jargon
Haha that's crazy I'm so used to reading RL papers that when the blog linked to a textbook about RL I just filled in Sutton & Barto without clicking on the link or thinking any further about the matter.

I think the other criticism I have is that the historical importance of RLHF to ChatGPT is sort of sidelined, and the author at the beginning pinpoints something like the rise of agents as the beginning of the influence of RL in language modelling. In fact, the first LLM that attained widespread success was ChatGPT, and the secret sauce was RLHF... no need to start the story so late in 2023-2024.

dawnofdusk··on Copilot Chat in VS Code is now open source
I mean you're right it's just a front end. And front ends can be open sourced? Obviously this has some public value: other people don't have to build a frontend starting from zero.

I don't think it's well-aimed criticism to say that the LLM design/training material itself should have been made open source. Pretty much no one in the open source community would have the computational resources to actually do anything with this...

dawnofdusk··on A Lisp adventure on the calm waters of the dead C (2021)
The author of this blog is clearly eloquent and, as per their interspersed quotations of David Hume and others, it is refreshing to see someone so well-read in the software/tech blogosphere.

I love Lisp. The last few paragraphs are a pretty good description. It's nice to have a very flexible set of tools, instead of being forced to conform to object-oriented design or whatever paradigm. IMO the only legitimate reason in sticking steadfast to a design paradigm is for performance reasons, but of course this can only really justify array programming/imperative programming. But at the point where you want some flexible abstractions, it's nice to have the power to do introspection, delayed evaluation, and so on. Disclaimer: my background is physics/math, so function abstractions are much more intuitive to me than objects, or whatever other structures are taught to CS students.

dawnofdusk··on Gemini CLI
Yes they do. Scroll to bottom of Github readme

>This project leverages the Gemini APIs to provide AI capabilities. For details on the terms of service governing the Gemini API, please refer to the terms for the access mechanism you are using:

Click Gemini API, scroll

>When you use Unpaid Services, including, for example, Google AI Studio and the unpaid quota on Gemini API, Google uses the content you submit to the Services and any generated responses to provide, improve, and develop Google products and services and machine learning technologies, including Google's enterprise features, products, and services, consistent with our Privacy Policy.

>To help with quality and improve our products, human reviewers may read, annotate, and process your API input and output. Google takes steps to protect your privacy as part of this process. This includes disconnecting this data from your Google Account, API key, and Cloud project before reviewers see or annotate it. Do not submit sensitive, confidential, or personal information to the Unpaid Services.

dawnofdusk··on Gemini CLI
https://en.wikipedia.org/wiki/First-mover_advantage
dawnofdusk··on Gemini CLI
I feel like you get more mileage out of prompt engineering and being specific... not sure if "fix all the bugs" is an effective real-world use case.
dawnofdusk··on Is mathematics mostly chaos or mostly order?
In the diagonalization you don't need to assume the existence of any real numbers. Just on the left hand side you write down, formally, any sort of numbers that have decimal expansions that may be infinite. Rational numbers have infinite decimal expansions too, it's just that they will eventually repeat, but at this stage it's not necessary to think about what the properties of these infinite decimal expansions actually mean. Then the diagonalization argument shows that this set of numbers with infinite decimal expansions are uncountable and also contain the rationals. This still doesn't define the real numbers yet: to do so one needs to think about the Euclidean metric on the rationals and how to complete it.
dawnofdusk··on Emergence of Diffusion Models from Associative Memory
The paper maps two energy-based methods to each other. I'm not sure what the innovation is.
dawnofdusk··on Chemical knowledge and reasoning of large language models vs. chemist expertise
Fast-moving field? This is a chemistry paper not an ML paper. ML people have their conferences which are on much abridged timeframes.
dawnofdusk··on Geometry from Quantum Temporal Correlations
Yes but Euclidean 3-space means you need to test 1) that it's 3-space, by counting dimensions and 2) that it's Euclidean, which is a question about local curvature
dawnofdusk··on Is gravity just entropy rising? Long-shot idea gets another look
This is just a question about the origins of inhomogeneity in the universe. The prevailing theory is cosmic inflation, I believe: in the early universe a quantum field existed in a high entropy state and then the rapid expansion of space magnified small spatial inhomogeneities in the field into large-scale structures. What we see as "low entropy" structures like stars are actually just high entropy, uniform structures at a higher scale but viewed from up close so that we can see finer-scale structure.
dawnofdusk··on Is gravity just entropy rising? Long-shot idea gets another look
Deriving existing theories of gravity is an important test of the theory, it's not a problem at all. It's only a problem if you can only do this with more free parameters than the existing theory and/or the generalized theory doesn't make any independent predictions. Seems like in the article the former may be true but not the latter.
dawnofdusk··on Geometry from Quantum Temporal Correlations
Not in this field directly, but first of all, talking about the "geometry of space" is more than just saying there are three spatial dimensions: geometry involves the local curvature of the object. Historically the Pauli matrices are discovered by assuming certain symmetries of spacetime. This paper shows the other direction also makes sense: if we assume certain structure on quantum observables, measurable only by temporal measurements and independent of the content of the quantum state (i.e., a measurement of any system will do), we can get the spatial symmetries we want.

I suppose the ideal outcome is that there is some sort of exotic algebra of observables which is well motivated somehow by purely quantum considerations and by serendipity induces all the usual spacetime symmetries + extra stuff we didn't know about. This paper itself is cute but not sure if it's very impactful, I would defer to domain experts.

dawnofdusk··on Why We're Moving on from Nix
If you like Lisp there is GUIX which is the Nix packaging idea but using a language that is slightly less trash. The ecosystem is much worse so it would not be great for a desktop end user AFAIK.
dawnofdusk··on Gaussian Integration Is Cool
They are the same for physicists who analytically continue everything. Steepest descent is technically for integrals in the complex plane, and Laplace's method is for integrating over the real line.
dawnofdusk··on Typing 118 WPM broke my brain in the right ways
Interesting! My personal experience is that using a vertical mouse very quickly cured wrist and forearm pain that I had, and that the reason is because the "handshake" position is the most neutral hand/forearm posture from an anatomical perspective. But everyone is different, of course
dawnofdusk··on Typing 118 WPM broke my brain in the right ways
Staying at home row is not really about ergonomics. The best thing you can do (without breaking the bank, i.e., no bespoke split keyboard setups) for your hand ergonomics at the computer is to get a vertical mouse and tilt your keyboard downwards and away from you (the opposite of what most keyboard feet do, which is tilt your keyboard up and towards you). Both of these keep your wrist neutral/in slight flexion, whereas the typical setup keep your wrists in constant extension which contributes to carpal tunnel.
dawnofdusk··on Typing 118 WPM broke my brain in the right ways
>I can't think of any benefit

If you want to minimize the time it takes to travel to a particular key, it makes sense to keep your fingers at the "center of mass" of the keyboard... which is the home row.

dawnofdusk··on Typing 118 WPM broke my brain in the right ways
Typing fast is an underrated skill for developers. A lot of the value added by various intelligent tab completion and LLMs is easily replicated by typing variable and function names at 100+ wpm.

Not a fan, however, of the desire to disparage the "right" way of doing things that's done in this blog. I type the "proper" way, home row and all, and can reach 150+wpm with high accuracy. So, based on this evidence, you're wrong and would be better off having learned the proper technique...

is an argument which I could make, and the evidence would be true, but it doesn't sound extremely convincing, does it?

dawnofdusk··on Applications of Classical Physics
Typically non-relativistic and non-quantum is called "Newtonian". Classical is just for anything which is not quantized, and so far no one knows how to quantize general relativity.
dawnofdusk··on Applications of Classical Physics
You learned about stress tensors and PDEs in high school?
dawnofdusk··on Applications of Classical Physics
Not really dumbed down, just that it prioritizes quantum physics instead of classical. One can debate whether this is a good set of priorities but it's flippant to say a curriculum focused on quantum mechanics is dumber than one focused on fluids and elasticity/continuum mechanics.
← PreviousPage 2 of 5Next →