Dmitry Babokin

Looks like Anthropic just knocked down the last wall protecting my work-life balance 😱

@bcherny Any hints how to avoid slash commands duplication in the worktree? I guess CLAUDE.md is also duplicated in the context, right?

Introducing: built-in git worktree support for Claude Code Now, agents can run in parallel without interfering with one other. Each agent gets its own worktree and can work independently. The Claude Code Desktop app has had built-in support for worktrees for a while, and now we're bringing it to CLI too. Learn more about worktrees: git-scm.com/docs/git-workt…

@bcherny Ha! it worked on the second attent, but didn't work on the first try. Probably it had to create some folders and failed silently 😕

@bcherny Bug report: -w doesn't work (while --worktree works).

One not very hot take - The Claude C Compiler has the best internal architecture docs of any compiler I've ever seen. Far, far, better than any compiler I've ever written, lol :-)

The best products of the 21st century have yet to be shipped. With 21st, any idea can find its aesthetic and design, even if you aren’t a designer. Meet the first vibe crafting tool for everyone…. So every product can have soul.

@sama Codex is good, but please make sure that all the essential features are there, like "resume the session" in CLI. I'm also looking forward to subagents!

people seem to really like the new codex features!

Oxmiq Labs, founded by GPU legend Raja Koduri, emerges from stealth with a dream team and $100B+ in past impact. Rebooting the GPU stack from atoms to agents™ with breakthrough IP, hundreds of patents, and a bold vision for AI and graphics. Visit oxmiq.ai

🚀 New ISPC v1.26.0 Release! 🚀 🔹 Improved ARM support & performance 🔹 New "generic" targets simplify ISPC design & ease new targets support 🔹 Better code generation for x86 & ARM 🔹 Lots of stability fixes github.com/ispc/ispc/rele…

Recently we released ISPC v1.25, featuring exciting updates: improved code generation, faster compile times, enhanced templates support, and new standard library functions! Dive into the full details here: github.com/ispc/ispc/rele…

@iruletheworldmo 🍓

🍓in the comments chat. early access for all!!!!!!

@karpathy It looks like Telegram solved exactly this problem when introduced "star" likes in Telegram. One star is approximately $0.02. People can like with as many stars as they want.

I feel like a large amount of GDP is locked up because it is difficult for person A to very conveniently pay 5 cents to person B. Current high fixed costs per transaction force each of them to be of high enough amounts, which results in business models with purchase bundles, subscriptions, ad-based, etc., instead of simply pay-as-you-go. As an example, I'd like my computer to auto-pay 5 cents to the article/blog that I just read but I can't, and I think we're worse for it. In a capitalist system, transactions between entities are the gradient signal of the economy. Because our pipes don't support low magnitude terms in the sums, the gradients are not flowing properly through the system. I'm not familiar enough with payments to have an idea of specific solutions, but I expect we'd see a lot of positive 2nd / 3rd order effects if the gradients were allowed to flow properly, frictionlessly and with much higher resolution.

@ashvardanian Based on github search it’s not that popular :) Actually I haven’t found any legitimate calls of it, only definitions. Apparently people are not educated enough about the existence of this gem!

... the perennial programming quandary: "How do I take good data in string form and painlessly turn it into garbage?" Source: GNU C Library How have I survived without ever using this 😂 #SEC80" target="_blank" rel="nofollow noopener">ftp.gnu.org/old-gnu/Manual…

@ashvardanian I use free tier Oracle Cloud Arm instance with 4 cores and 24GB of RAM to run self-hosted Github runner. It's pretty fast, possibly it can handle your 10 projects.

Recipe to generate x86 and Arm binaries in GitHub CI 1. start Docker buildX 2. pass SSH credits into it 3. configure Dockerfile to install recent GCC, build, and 4. reverse SSH into parent runner & scp file into it ... burn all and switch from GCC to LLVM for cross-compilation

@ashvardanian And SierraForrest is Atom cores and it will be 288 cores. But it's not going to be and an edge chip - at least 288 version.

@ashvardanian Looks like you are mixing things up. D-line is for edge and the current offering for D is IceLake. Sapphire/Emerald is not available for edge, it's big servers only. So Granite vs IceLake comparison looks fair. Note Granite is Big Core product, it's not going to be 288 cores.

288-core CPUs on edge! Release planned for 2025 🎉 Hardware is amazing, but PR teams in semiconductor suck! Last week I saw a GPU vendor comparing its card to a 2015 CPU. And now Intel compares Granite Rapids to Ice Lake instead of Emerald/Sapphire 🤷‍♂️ anandtech.com/show/21276/int…

@science_dot And by saying that autovectorization problem is not "solved", I mean it doesn't work perfectly out of the box. To get optimal code you either need to put a bunch of pragmas, rewrite your code in specific way, or write explicitly vector code.

@science_dot Thanks for the slides. As I understand, there are pragmas/directives to guide the vectorizer. But it's not much different from modern Fortran/C/C++ autovectorizers - they have a bunch of pragmas to guide them or make their work easier.

Hey, CS twitter, I need your help with CS archeology. Some time ago my senior compiler colleague told me a story that when Cray released the first vector machine in 1975 (Cray-1), they promised to solve autovectorization problem in one year. Almost half a century has passed (1/2)

@JukkaSuomela @science_dot Yeah, hardware was quite different, but from software perspective the autovectorization problem is very similar to autovectorization problem on the modern hardware.

@science_dot @DmitryBabokin For me Cray-1 started to make sense if I imagine that "vector registers" are just a special cache that you control explicitly, and you've got magic instructions with which you do batch operations that stream data between the cache and (scalar, pipelined) execution units. :)