bariscan

@maxedapps Isnt claude subscription problem? How do you manage it?

Thanks to pi, I now expect all my software to be self-improvable.

@chenzeling4 If anyone needs container manager github.com/capsule-native…

Docker Desktop on Mac is a resource hog. Apple quietly built their own. container runs Linux containers as lightweight VMs on Apple silicon. Written in Swift, optimized for M-series chips. OCI-compatible. Pull from any registry, push to any registry. No Docker daemon. 47.5K stars. People are paying attention. ⭐ 47.5K #Docker #Apple github.com/apple/container Follow for daily dev finds 🔔

@rasbt Is it possible to add GLM as well?

Upon request, here's an updated version with Grok 4.5 and Meta's Muse Spark 1.1. Grok 4.5 seems to sit at the Pareto frontier. Good bang for the buck. (Also added harness info).

@MaximeRivest Forbidden gpt math feature

What if we could do math and programming on pen and paper? This kinda blows my mind! So many possibilities!

For agentic coding, it just not worth it for your time and money. People are mistakenly promote this by doing single file edits or tasks that requires low context length. Local inference starts to struggle when you use 128k context and beyond. It just requires enormous hw requirements which are fairly expensive. However, there are multiple potential models for daily conversion if you already have the hardware. If you are buying it just for this purpose unless it is research purposes or curiosity, it still doesn’t make sense because that tier is $20.

So @theo published a YouTube video about local inference and its hard limits (you often can *not* run the fancy open weight models locally) and I agree with every word, but then with two M5 Max 128GB systems and RDMA I'm at 50 token/sec with DeepSeek v4 Flash Q5 even without speculative decoding (that wins another 15/20% max, it's a very sparse model, single session, no magic tricks). And I'm starting to thing we are actually exactly at the turning point of all that.

@MaximeRivest Does it work on remarkable 2?

Still a WIP but if you know Linux, ssh, and AI that should get you going: github.com/MaximeRivest/R… I announce more broadly when I have confidence beginners could make it work.

@jarredsumner We are getting blog post before gta6

rust rewrite blog post Monday Bun v1.4 hopefully Tuesday (video still editing)

@Love2Code @kskrygan You are right for daily chit-chat sessions, good luck for agent work

@barscnasln @kskrygan You can run it on a MacBook Air 32GB

Unpopular opinion: While everyone is so hyped about Fable, GPT5.6 and other huge and expensive models, I think the real hero of the last few months is *Qwen 27b*. Our ML/AI engineering teams are have been very surprised how good/fast this open model is.

@Orwelian84 @kskrygan Ability of running model does not mean you can do agentic work. Searching 2-3 files easily take 64k context, so you actually need at least 48-64GB vram and also if you have less then 800gb memory bandwidth your eyes and screen will stare each other

@barscnasln @kskrygan it runs in 24gb of vram at reasonable quants -- in ML world thats tablestakes

@rough__sea It is good indexer, they prioritize to read this if the file exists.

AGENTS.md/CLAUDE.md is largely an anti-pattern

@Alennushkaa Jai

So what programming language do I need to learn to feel that life has meaning again?

Sadece embedded yazilimdan bahsetmiyorum ama gercekten hakkiyla embedded yapan insanlarin ustun zekali olmasi sart. Uzgunum ama dunyanin en zor meslegi gercekten.

@sflorimm They can fork ladybird.org

Anthropic should build a browser. not just another Chromium clone, a genuinely different one. webKit-based, fast, private with Claude baked in at the engine level. who else wants to see this happen?

@levelsio If you once connected it from web it does this, try to disconnect them or reconnect

Why Claude keeps telling me to connect MCP to Google Drive etc?

@jun_song You still need big model for huge context

As I said, fine-tuned small local LLMs for specific task works better than general frontier models. Learn post-training.

@ClaudeDevs @trq212 Nice will try

Claude Desktop is now available on Linux (Ubuntu and Debian) in beta. Alongside the browser and terminal, you now get a first-class desktop experience with Claude Code, Claude Cowork, and chat on all paid plans.

@jun_song Their API pricing is low

My god, who is still using Gemini? How is there still demand left?

@Keleesssss Tebrikler Alperen

Some good news, my Ph.D. is done after 5 fantastic years. So long, and thanks for all the fish!

@cursor_ai I thought it was sub plans get some percent off, wanna try new cursor tho

Introducing Cursor for iOS. Build from anywhere by launching always-on cloud agents. Or remotely control agents running on your computer from the app. Composer 2.5 is 75% off in the app now through July 5.