Top all-time · #6160
The espadrine blog.
Personal blog · espadrine.github.io/blogprogrammingai
The espadrine blog. is a personal blog at espadrine.github.io about programming and ai. Since 2014, 13 different Hacker News readers have linked it 33 times, most often “Chinchilla’s Death”. The last post was in 2024, but the archive is still widely read.
Too few sampled comments (7) to say how often HN recommends it.?
Why HN links it
HN commenters link this blog's posts on LLM training scaling (the Chinchilla estimate) and on pseudo-random number generators. One commenter found the Chinchilla post enlightening.
Summarised by Claude from a sample of 7 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Most-linked posts?
- Chinchilla’s Death4 people · 7 mentions · first linked 2023-07-26
- Show HN: Shishua – Fast pseudo-random generator4 people · 4 mentions · first linked 2020-04-18
- Recomputing ML GPU Performance: AMD vs. Nvidia2 people · 2 mentions · first linked 2023-07-26
What Hacker News says?
I made a short sum-up of changes in this specification here: http://espadrine.github.io/New-In-A-Spec/es2017/
Comment linking /New-In-A-Spec/es2017/
on “ECMAScript 2017 Language Specification”Submitted Show HN: Shishua – Fast pseudo-random generator as a story · 93 points
HN discussion> Don't filesystem journals ensure that you can't get a garbled file from sudden shutdowns? For once, I have a blog post that goes in detail into why that is not true! (Or rather, not true unless you implement it in a very sophisticated manner rarely used in practice.) https://espadrine.github.io/blog/posts/file-system-object-st...
Comment linking /blog/posts/file-system-object-storage.html
on “SQLite as an Application File Format (2014)”Submitted Sometimes, rewriting in another language works as a story · 56 points
HN discussionI made a very succinct summary here: http://espadrine.github.io/New-In-A-Spec/es2015/ There is also this, slightly more in-depth: git.io/es6features
Comment linking /New-In-A-Spec/es2015/
on “ECMAScript 2015 Approved”Submitted EcmaScript Sixth Edition as a story · 40 points
HN discussionI concur; I wish for a delta to be made more explicit as a non-normative section. I added concise explanatory additions to the spec here: http://espadrine.github.io/New-In-A-Spec/es2016/ . Currently, the changes for ES2017 are the following: http://espadrine.github.io/New-In-A-Spec/es2017/ . File I/O means different things in different contexts. If implementing JS required a file system to be present, it would not b…
Comment linking /New-In-A-Spec/es2017/
on “ECMAScript 2016 Approved”Submitted A Primer on Randomness as a story · 14 points
HN discussion> I'd be very happy with an environment that provided dynamically typed languages the same analysis, checking, completion Thinks like tern.js[0] or my own Aulx[1] provide that with varying success. Visual Studio's JS autocompletion is also stellar. The techniques used could be applied (with tweaks) to most dynamic languages. [0]: http://ternjs.net/doc/demo.html [1]: http://espadrine.github.io/aulx/
Comment linking /aulx/
on “Show HN: Hydrogen Brings Light Table to Atom”Submitted Memorable Passwords as a story · 4 points
HN discussionIn case SpaceX is reading: I would love to read a postmortem. After having built Spash[0], I have much more appreciation for the difficulty of correctly computing spacial locations. [0]: https://espadrine.github.io/spash
Comment linking /spash
on “SpaceX Dragon Rendezvous and Docking Waved Off for Today”Submitted Show HN: Spash – Spacetime Locator as a story · 3 points
HN discussionSome comments, in random order: - your test codes are not reproducible: running twice generates different sets of numbers because they have unknown seeds. As a result, if you change (a) compilers or compiler switches, (b) operating system versions, (c) host processors or (d) architectures, the question arises: what is wrong? What is different? This is known as a regression test. - try shishua as another speedy PRNG.…
Comment linking /blog/posts/shishua-the-fastest-prng-in-the-world.html
on “Show HN: LoopMix128 – Fast C PRNG (.46ns), 2^128 Period, BigCrush/PractRand Pass”Submitted Wordle Perfect Play with Generic Methods as a story · 2 points
HN discussionJSON offers data structures that everybody understands, so having a configuration language that has exactly those is what I want in a configuration language. TOML isn't it (some JSON can't be put in TOML by design). YAML is definitely easier to edit than JSON, but just like XML its flaw is in its complexity (both in syntax and in specialized structures, such as references or types). That's why I made dotset ( http:/…
Comment linking /dotset/
on “Universal configuration language”Submitted Faster Wordle Estimators as a story · 2 points
HN discussion> Is it only theoretical, or is there any sample code? Ah, not exactly. I saw it a something that could become a draft standard. That said, the test vectors do come from a program that implements it by hand[0]. I assume we could write one that would implement it for the command line (say, for cURL), but the design is really meant to be implemented by browsers. > You could associate (some or all of) them with an acco…
Comment linking /blog/posts/memorable-passwords.html
on “SQRL: Secure Quick Reliable Login”Submitted Recomputing ML GPU Performance: AMD vs. Nvidia as a story · 2 points
HN discussionI love WebAuthn for 2FA, but I feel like it will struggle to go from 2FA system to primary login (as full replacement for passwords). Indeed, exporting key information between browsers or devices are out of scope, causing browser makers to recommend never making WebAuthn log-in the only log-in mechanism, and allow passwords as fallback[0] (not really a replacement for passwords then). Additionally, it is more comple…
Comment linking /blog/posts/webidentity.html
on “SQRL: Secure Quick Reliable Login”Submitted Schemes: Performance, Conformance, Openness as a story · 1 points
HN discussionIsn't the Chinchilla estimate considered to be wrong now? https://espadrine.github.io/blog/posts/chinchilla-s-death.ht...
Comment linking /blog/posts/chinchilla-s-death.html
on “The Mathematics of Training LLMs”Submitted Chinchilla’s Death as a story · 1 points
HN discussionWhere did you find 1.8T for GPT-4 Turbo? The Turbo model is the one with the 128K context size, and the Turbo models tend to have a much lower parameter count from what people can tell. Nobody outside of OpenAI even knows how many parameters regular GPT-4 has. 1.8T is one of several guesses I have seen people make, but the guesses vary significantly. Regardless, I’m not the one making that comparison. The person I r…
Comment linking /blog/posts/chinchilla-s-death.html
on “New models and developer products announced at OpenAI DevDay”Submitted Rant About Haskell as a story · 1 points
HN discussionYou’re just spouting theoretical numbers here, not actual real-world numbers. First, ChaCha20 can easily be around 1-3 cpb, which translates into around 4-12 c/u32 (source: AVX2 perf of Go ChaCha20 package: https://pkg.go.dev/github.com/aead/chacha20#section-readme ). So that’s already a lot better than your theoretical claim of 60 c/u32. Second, a multiply-add PRNG is going to have abysmally poor statistical proper…
Comment linking /blog/posts/shishua-the-fastest-prng-in-the-world.html
on “Need a PRNG? Use a CSPRNG”Submitted Chinchilla’s Death as a story · 1 points
HN discussionYup. I found this article quite enlightening: https://espadrine.github.io/blog/posts/chinchilla-s-death.ht...
Comment linking /blog/posts/chinchilla-s-death.html
on “RedPajama v2 Open Dataset with 30T Tokens for Training LLMs”Submitted A Base32 Checksum as a story · 1 points
HN discussionPossibly a lot. See: https://espadrine.github.io/blog/posts/chinchilla-s-death.ht...
Comment linking /blog/posts/chinchilla-s-death.html
on “TinyLlama project aims to pretrain a 1.1B Llama model on 3T tokens”That’s the one thing that I feel is a bit misleading in the article (to be fair, it was initially written years ago, and got rewritten a bit recently). FLOPS comparisons given in the wild are not always apple-to-apple (eg. not including Tensor cores for NVIDIA, but including V_DUAL_DOT2ACC_F32_F16 for AMD), while on the flip side, AMD’s WMMA should address the same goals as Tensor cores. I have an article on compari…
Comment linking /blog/posts/recomputing-gpu-performance.html
on “Which GPU(s) to Get for Deep Learning”