
r3dlust
109 posts


@ZixuanLi_ does this regard stability at higher context sizes (ie over 100k tokens)? i’ve found that the GLM models all seemed to follow that same weird pattern of destabilization at around 100k tokens or more, tool calling accuracy dropped massively and the model spewed garbage tokens
English

@gokcan_akalin @zRdianjiao what do you mean? what could the service provider have to do with model stability at higher context (unless they were quantizing the model or compressing the context hyper-aggressively)?
English

@r3dlvst @zRdianjiao Dont exceed 100k context length. Otherwise you will immediately get gibberish results. I think this is not related to the model itself but the service provider.
English

You can try GLM-5-Turbo and GLM-5V-Turbo.
Maybe they are not quite Claude Opus 4.6 level yet, but they are absolutely strong contenders — and definitely worth a shot.
Boris Cherny@bcherny
Starting tomorrow at 12pm PT, Claude subscriptions will no longer cover usage on third-party tools like OpenClaw. You can still use these tools with your Claude login via extra usage bundles (now available at a discount), or with a Claude API key.
English

Claude literally just dropped the missing piece to becoming OpenClaw.
You can now send messages to your Claude Code through Telegram or Discord.
Thariq@trq212
We just released Claude Code channels, which allows you to control your Claude Code session through select MCPs, starting with Telegram and Discord. Use this to message Claude Code directly from your phone.
English

@dhh @FireworksAI_HQ @opencode a fast TTFT feels amazing, i’ve found myself using GLM-5 a lot lately (i got their 1y plan so might as well use it) and it has a TERRIBLE TTFT, feels like im waiting forever just to fix a single line of code.
English

Kimi K2.5 continues to be my daily driver for all the basic stuff where I don't need PhD-level intelligence. I just need it done quickly. Running it at 200 tps through @FireworksAI_HQ within @opencode is just such a delight.
English

oh wow… did not expect that from cloudflare but… i guess things are changing now, huh?
Cloudflare Developers@CloudflareDev
Introducing the new /crawl endpoint - one API call and an entire site crawled. No scripts. No browser management. Just the content in HTML, Markdown, or JSON.
English

When NixOS users get to share their dot files or configuration.nix
x.com/i/status/20021…
RAW EGG NATIONALIST@Babygravy9
Can we put this guy in a SpaceX rocket and send him on the first manned mission to the sun?
English

@HSVSphere real, they have like all the fucking CI in the world and all the proper things to check for breaking builds but they just go “you know what, fuck it, i don’t care if this breaks a whole bunch of user’s systems, lets merge this now!”, while actual important PRs get ignored
English

@zeddotdev @HSVSphere remote ssh workspace access is broken on machines that have nushell as their login shell due to an improper usage of the uname command, would you guys be willing to fix that or add a way to choose what shell to run commands in for the internal ssh access client?
English

@HSVSphere why do you use the single same SSH key for literally everything? was reading through and this kept popping up and bothering me quite a bit. like even for git remotes, you use the SAME ssh key as always
English

@r3dlvst github.com/RGBCube/ncc, I should really document it better tbh
English

@jtregunna @linuxopsys how would different architectures or hardware-specific firmware work?
English

@linuxopsys I download my entire operating system on whatever computer I netboot from every day.
Laptop or desktop, it's the same system.
English

@HSVSphere will study, still new-ish to nix, mostly frustrated with my experience with the existing LSP servers which i can’t say has made the experience of coding nix systems rather bad, but this is in all honesty a skill issue
English

@HSVSphere sure, except when i'm super nested and have to do super::super::super::super::utils just to get a teeny utils function
English















