Monk Zero

The only way we human are able to communicate and understand each other, is that across space and time, we are one and all Inspired by <The Egg> by Andy Weir galactanet.com/oneoff/theegg_…

@ludwigABAP @Yuchenj_UW Yep, actually I think somehow this become of better signal to know candidate has good fundamentals. Public github projects and profiles means a lot less now. In my experience working in most top tech companies , this remains one of top indicators regardless of what people say

@Yuchenj_UW x.com/ludwigabap/sta…

I’m so glad AI killed LeetCode interviews. For 10 years, tech companies made every engineer grind the same puzzles and prove they could invert a binary tree from memory. Today, the dumbest AI model can walk in and one-shot the entire interview. Thank you, AI.

@mycharmspace 啊才知道你也用twitter😂. 祝贺🎉! Search is indeed my most used grok feature ❤️

Today is my last day at xAI. I joined xAI a year ago and had the pleasure of leading the search and factuality post-training team. Over time, we developed so many recipe and engineering co-optimizations, making Grok the best AI for search and real-time agent. I am also particularly proud of working with a small group of talented people delivering the recent iterations of the instant mode of Grok - the one I personally liked and used the most. My thanks to all the friends and teammates for their support and help over the past year. They are among the brightest minds I’ve met in my career. I am sure the team will continue the mission to make better Grok and understand the universe.

@antirez This is the way 🫡. Next is bidirectional multi-stream.

Now DS4 implements the OpenAI Responses API and attempts to match the IDs in order to continue from the live KV cache without doing the efforts required in the chat completion API code path.

@badlogicgames @ShopifyDevs x.com/nocommas/statu…

looked into /goal in both cofex and claude codr and all i'm seeing are inferior versions of autoresearch. what am i missing? available from the @ShopifyDevs folks as a pi extension. shopify.engineering/autoresearch

@shafu0x Member of Technical Support

forward deployed engineer just means the guy is not fucking autistic

@jonasgeiping Got a feeling some of Thinky and OpenAI realtime api already does something like this. Great work, this direction feels right

Finally, we find that models with many internal streams allow us to more easily monitor their thinking, for example concerning evaluation awareness. With many parallel internal streams, it would be my hope that the model continues to subvocalize concerns in side-streams, even if the main CoT/thinking stream is occupied with solving a particular task.

We’re training models wrong and it’s due to chatGPT. Even the modern coding agents used daily still use message-based exchanges: They send messages to users, to themselves (CoT) and to tools, and receive messages in turn. This bottlenecks even very intelligent agents to a single stream. The models cannot read while writing, cannot act while thinking and cannot think while processing information. In our new paper, see below, we discuss LLMs with parallel streams. We show that multi-stream LLMs can … 🔵Be created by instruction-tuning for the stream format 🔵Simplify user and tool use UX removing many pain points with agents and chat models (such as having to interrupt the model to get a word in) 🔵Multi-Stream LLMs are fast, they can predict+read tokens in all streams in parallel in each forward pass, improving latency 🔵 LLMs with multiple streams have an easier time encoding a separation of concerns, improving security 🔵 LLMs with many internal streams provide a legible form of parallel/cont. reasoning. Even if the main CoT stream is accidentally pressured or too focused on a particular task to voice concerns, other internal streams can subvocalize concerns that would otherwise not be verbalized. Does this sound related to a recent thinky post :) - Yes, but I don’t feel so bad about being outshipped with such a cool report on their side by 23 hours. I’ll link a 2nd thread below with a more direct comparison. I actually think both are complementary in interesting ways.

@iridescence_dev @ST_Automation @ptr_to_joel @joshmo_dev Yes and we already did. And it is beautiful github.com/AntigmaLabs/an…

It's worth it, even if the network is the ultimate bottleneck. Users should have high quality software. Taking 0.5 seconds to load and 270-300MB at startup for a TUI is completely unacceptable when their competition, Codex CLI (Rust), can do it in a fraction of the time and half the RAM. Software Engineers have a duty to demand higher quality software that people can actually love using and stop making excuses.

holy wow they merged it

@heyandras Doing gods work

We made a fake repo with fake bounties, and the bots are applying fake PRs, so we know who is fake, and we can ban them from the Coolify repo. IQ over 1000

@badlogicgames @MaksShamihulau Haha I held the same strong opinion against GoLang; the only language I dislike with some passion

@MaksShamihulau but then i'd have to use rust, and i dislike rust a ton.

uhm i sort of disagree :p

@badlogicgames @hjanuschka Lua was always the most lovely one. But right now bash is the sweet spot.

@hjanuschka i don't want to push lua onto anyone.

@mitsuhiko Yep. Rust based agent should focus on resource footprint and reliability; there is room for both.

Pi wouldn’t make any sense in rust or go. Extensibility is key to it. That leaves ruby, python, js, php for the most part unless you want to ship an interpreter. None of those languages have any benefit over node.

@satory_ua @ThePrimeagen It has got too good recently since gpt-5.2. Miss the time when it is super easy to spot filter slop rust

@ThePrimeagen > Rust > Looks inside > Arc::new(Some struct::new(param.clone())) everywhere > CLAUDE.md in the root of the repository

Current meta

@SIGKITTEN @thdxr interesting how different team has very different priorities; building sdk was actually where we started and everything grows from how to best interact with LLM; never occurred to me to use any 3rd party client sdk library

@thdxr finally big enough to remove ai-sdk!

we're working on a library to abstract over all the llm providers there's very few teams that have dealt with the quirks between providers at the scale we have it's written in effect but will also have a vanilla api progress is in the opencode repo under packages/llm

@badlogicgames @rebelcrayon “Inline this” is probably my most typed words recently. I feed the entire deep module lecture into both and still see this shit from time to time

calling it slopex from now on so it can join its sibling slopus.

@tenderizzation him and Linus really built the pillars of our current digital world

wow

@thdxr there is alway an ancient Chinese proverb for these kind of things: 守正出奇

i pay attention to: 99%: using our product in dumb/simple ways and will never change behavior 0.01%: aliens who are showing us the distant future ignore the reminder: "pro" users who think they invented some clever workflow every week but get less done than the 99% group

@orzxh97 @badlogicgames honestly the first one reads like AI reply lol

@badlogicgames Oh I’m not saying to you. I’m agreeing you. Sorry that wasn’t clear.

but it's cool that frontier models are now basically regressing. maybe all this madness will come to an end soon.

@antirez Makes sense thanks!

@NoCommas Unfortunately this si what DeepSeek official API returns, I don't know why they are masked / place-holder-ed. So the test checks that the continuation matches without being able to check the logits values. But it's enough to spot issues given that we test with long contexts.

Welcome to DS4, a specialized inference engine for DeepSeek v4 Flash. github.com/antirez/ds4 This project would have been impossible without the existence of llama.cpp and GGML and the work of @ggerganov and all the other contributors. Thanks!

@eshear 🫡 you built something outlived your own fame

It’s an honor just to be nominated