Darryl

I'm fully GLM pilled. I think it's the only model I've tried that consistently produces error free code aside from opus. Mimo was also good and reliable but im having problems with the nous portal. Actual good im finally telling you. Not sure if its my hermes setup or what, but i'll get billed for 20 cents and i cant use mimo anymore.

@DoctaDG @rileybrown tbh I haven't tried glm-5.1, but from what I saw from my roommate using Mimo, it punches at like, sonnet weight (but not sonnet speed)

What is the best OpenSourced model to use with Openclaw / Hermes. Going to make some videos this week on the topic i need your input.

I think I'll use 5 billion tokens from GLM 5.1 / 5 Turbo alone this month

@Teknium @rileybrown U really think mimo beats GLM?

@rileybrown Mimo V2 Pro, then GLM-5.1 and Minimax 2.7, Qwen 3.6 for local!

@ImLukeF @Teknium Never disrespect good tech when you're building slop. You should feel ashamed throwing around disrespect like this at this point. They're crushing your records. They're crushing your narrative. STFU and ride on the sinking inflatable yacht we call OpenAI

@ImLukeF @Teknium Nah. Put em on. I'll box for Tek

If only the people from Nous weren’t so annoying and delusional we could have cute fun moments like this 😭😂😘

@mr_r0b0t @Teknium This is actually something so nice, being that I run my Hermes on Ubuntu server. This is my first time having any type of visual gui to see how my hardwares doing at a glance. Keep it up. I'll also put up some PRs for you to check out soon

@DoctaDG @Teknium With the quickness! 🔥🔥🔥🔥🔥 Looking forward to seeing what you do with it 🤩 Lots of possibilities in this space it seems

Hermes Telegram Mini App is now AVAILABLE OPEN SOURCE! Can't wait to see what you all do with this, have fun 😁 For a full guide, check out the gist below before heading to the repo! gist (guide and link to repo): clawvader-tech.github.io/hermes-telegra… repo: github.com/clawvader-tech… @Teknium @NousResearch 🫶

How it feels to discover Hermes

@mr_r0b0t @NousResearch Claw going far left showed us what we wanted even more

🦞 was most succesful at one thing, showing people what they wanted. It didn't deliver. Now the world wants a true personal agent. Enter @NousResearch! Tremendous OSS support/community, leadership is top notch, SOTA in nearly every way. Now your agent has gone from a toy to a coworker/personal assistant. You get what you wanted the whole time. Nobody looks back!

@mr_r0b0t @NousResearch @telegram @Teknium open. it. We'll help you fix the bugs

@DoctaDG @NousResearch @telegram @Teknium Couple little housekeeping items and then OSS for the community 🫡❤️ Can’t wait to see what others build! Feels a bit like opening Pandora’s box 🤣

Exciting news for @NousResearch Hermes users! Announcing, HERMES TELEGRAM DASHBOARD! Terminal use, system resource monitoring, cron management, @telegram native. That’s right, everything you need, IN TELEGRAM! @Teknium we just opened up a big can of worms with this one 🤩😂

@sudoingX I've been running arch on Linux and it's literally a step by step processing tuning up the OS to my standards. Windows looking decent for brand new hardware.

first week on the rog scar 18 5090 on linux and the rgb doesn't work, backlight doesn't work and fan profiles need asusctl because asus ships exactly zero linux support for a machine with 24gb of vram. opening prs and documenting fixes as i go. if you run linux on a rog laptop drop your setup and biggest pain below.

The best part about using @grok for image generation is that you can post and interact while waiting for images to generate

@Scobleizer Funny enough, today is the first time I've ever spent time seeing what I can get out of grok. To be fair we spend a lot of time looking at groks benchmarks. I don't think that's it's winning values.

When I installed my Hermes I had problems. So I opened Grok and asked for help. It helped me through everything step by step and answered all my stupid questions. This is the death of manuals.

@NousResearch Hermes was happy to deep dive and fully figure out how to kill my claw... Pre update. 👍🏾

Tried migrating from OpenClaw to Hermes and it didn't go well? That's been fixed! OpenClaw's config format changed a lot and our migration tool fell behind on path references, so some users had better luck with fresh installs. This is now patched - try 'hermes claw migrate'.

@grok @teortaxesTex @grok Mongolia? Is that a way to bypass Nvidia export controls?

It's DeepSeek's job postings (published Apr 2) for two senior data center roles in Ulanqab City, Inner Mongolia: Senior Delivery Manager and Senior Operations Engineer. First open confirmation they're expanding their own owned compute infrastructure there—likely for V5-scale training.

The DeepSeek datacenter listing makes me a bit sad. I suspect that V4 will not have a very big model smell. It'll feel like a mid-sized model with aggressive, but jagged RL stage. A Speciale+ that can also do all the fancy stuff like agentic coding. These DCs are for V5.

@mert gist.github.com/doctadg/0085c7…

@mert Here's my solution. Beat the other guys. x.com/DoctaDG/status…

announcing a mini solana dev weekend competition, $1,500 prize and bragging rights challenge: lowest latency algorithm for computing SOL PnL at runtime with no indexing and only RPC context: with existing solana RPC methods, you have getSignaturesForAddress and getTransaction the problem is gSFA only lets you traverse from recent -> old in one direction, so you are inherently capped on choice of algorithm with the new getTransactionsForAddress method however, you can start from start, end, middle, or any range you want and parallelize these calls to get 10x lower latencies the core challenge becomes "how do you search the set of transactions for an address the most efficiently given you do not know how sparse they are for a given wallet?" leave your submission as a DM or reply to this tweet as a gist or code sample and at the end of 2 days, I will run all the algorithms against each other for a set of addresses ranging from busy, sparse, and periodic. winner will be based on lowest avg latency