Tensor-Slayer

Exploring Direct Tensor Manipulation in Language Models: A Case Study in Binary-Level Model Enhancement: areu01or00.github.io/Tensor-Slayer.…

> You use hermes through Discord/Telegram > Data has grown , multiple GB, 100s of skills, 100s of plug-ins, 100s of commands, 100s of crons, 100s of tasks > You don’t have visual layer in your hand > Hermes dashboard is perfect for this > Created a visual layer where In v 0.1.1 - Huge update coming to make sure that install is super easy. Also new features you can try rn as part of mini update : 1. Resize text size 2. Fully functional hermes TUI right in your mobile

My workday: 1. Finish the work within 2 hours, already exhausted 2. 16 hours - Read papers, x, open source projects, x , read random 90s article on programming philosophy, x, read paper, keep up with new coding agents, x, hermes agent tasks are sprinkled throughout the day, x.

When do you reach for other models instead of Codex? What can we do better? Hit me with all of your frustrations. dms open. If you can give me detail (e.g. specifics/transcipts) - it'll help a lot in finding out exactly what we need to do to improve the next model

@wondering_camel LMK if any issues

@TensorSlay omg, i love you

Codex App Linux - Open Source Port Updates : 1. Got mobile paring working 2. UI stability 3. Voice input 4. New plugin/hooks rewrite Work-in-progress : - Further UI stability and keep providing future feature support from upstream codex app untill OpenAI releases official one. Source and one click installer : github.com/areu01or00/Cod…

Paper of the day! huggingface.co/papers/2605.13…

@mike64_t I often find myself asking 5.4xhigh / 5.5 to unlearn the behaviours it picks up during ICL for instance one off document creation turns into a chore and it doesn’t let go.

Codex is widely known to drop coding strategy philosophy bars, but if you dump too much of your thought process into your prompt in cot-like fashion, it starts speaking your language in an uncanny way. It’s smelling your naming process and overall word usage patterns in a way that’s non trivially accurate. It’s like it ICLed your CoT style and therefore starts to think more like you. Alien.

is this true

ported the /goal command from codex to a standalone mcp and slash command for arbitrary agents and harness (for now support for claude code and opencode) github.com/secemp9/goal

Hello Everyone, Early contributor of hermes agent here. Check out my new repo : Hermes-Agent Android Client Let me know if you would like to see improvements in specific areas or new features or basically anything related to repo or hermes-agent itself. Explore code, install Apk and try it out now.

@Teknium @NousResearch cc @tom_doerr

cc @Teknium @NousResearch

Open-sourced Hermes Agent Mobile Client for Android. It’s an Android native client for a running your existing Hermes Agents on your PC/VPS. Playstore and app store review pending but you can try the app on android now.

Thank you. It’s just one command to get it going but that one command becomes difficult to find for a non technical person so thought of adding “for agents” section, everyone these days anyway do that. Good practice to also check the source code and make sure everything is okay before install. This one is aimed at choosing path of least friction.

@TensorSlay Lol very modern, bro even have a Instructions for agents in the github

🚨Typical RL algorithms and on-policy distillation methods are blind samplers: they use privileged info to score rollouts, but not to *find* them. We ask: can we use privileged info to *actively sample* the rollouts RL wishes it can stumble upon with compute? ⤵️ Pedagogical RL

May I present, Elwood, a claude sdk replacement without subprocess/claude -p usage through AST instrumentalization of existing claude code installation (blog, github and npm link in replies) 1/4)