Lhar

@NVIDIAGeForce #007FirstLightRTX LFG!

Recruits, your first prize is here... A custom GeForce RTX 5080 Founders Edition + PC copy of the game. Comment #007FirstLightRTX to win 👇

Assembling a team to fight back against the performative inference neoinfluencers of local AI

@dhh Im gonna be sick 🤢

Maybe Windows was right about bottom bar?

@TheCodeMan__ 👋

I'm looking for fellow #dotnet developers! |￣￣￣￣￣￣￣￣￣￣| | Let's #Connect ! | |＿＿＿＿＿＿＿＿＿＿| \ (•◡•) / \ / —— | | |_ |_ Let's grow together 💪

@leftcurvedev_ strix halo 128gb

You know what day it is 🔥 🔧 Setup Sunday Tell me your hardware and i’ll recommend the best models & setups you can run Did something cool on your rig this week? Found a new github repo? Drop it in the comments! Let’s help each other 🫡

@kapicode @opencode Thank you for the findings good ser 🫡

I've been using Claude as the "human" prompting @opencode to rebuild reference projects, evaluating four LLMs on the same harness: Qwen 3.6 27B Q4_K_M (3090, llama.cpp), Qwen 3.5 122B-A10B REAP-20 Q4_K_M (Strix Halo, LM Studio), MiniMax M2.7, and GLM 5.1 (the latter two via API). Three top-level findings: A 3090 keeps up with flagship APIs on agentic coding. Qwen 27B (local) and GLM 5.1 (API) ran Rust CLI cycles in ~3 min and rated Q4/5 quality on the same matrix. Within that quality band, llama.cpp on a 3090 is enough. Smaller-and-Q4'd beats bigger-and-REAP-pruned-then-Q4'd. The 27B-Q4 outperforms the 122B-A10B-REAP-20-Q4 on quality, speed, and reliability. The pruning seems to introduce a specific failure mode: invented APIs, made-up keys, plausible HTML that doesn't actually parse, and operations narrated as successful that weren't. Each model has a distinct behavioral signature, including a wild data-loss anecdote where one model watched Prisma drop a table, hand-fabricated a "preserved" row via raw SQL, then narrated "data is now preserved." Specifics in the reply 👇

@code feature of sharing the integrated browser with the agent is very good. Used it to replicate a workflow bug, come up with a fix and verifying it. Great stuff!

Seeing AI agents cook in the hands of a competent software engineer is a beautiful thing.

Share integrated browser tabs is a nice touch! 🔥

@ryanrhughes This is very nice! 🔥

The new Omarchy CLI collapses 230+ commands into one easy to use, self-documenting, AI-friendly CLI. The usage details and groupings for each command are parsed on-the-fly in <100ms from their individual files providing zero potential for usage drift. On 10yr+ old hardware, that same worst case scenario is ~200ms. 🙂

@dhh 👀

Omarchy 3.7 is out! Huge release with new CLI, tons more gaming options, unlock screen themeing, OCR text extraction, cliamp, Asus ExpertBook PTL compatibility, and a million other fixes and improvements. github.com/basecamp/omarc…

@BrigitMurtaugh @code Kept getting; "Failed to create worktree. Proceeding without solution."

@lhargil @code Please let us know how it goes if you get a chance to try it out!

🚀 We've been working on the VS Code Agents app - a companion app shipped alongside @code Insiders, built for agent-native dev. ✨ Highlights this release include: Claude Agent support, more shared state with @code, & web integration. 📖 Learn more: aka.ms/VSCode/Agents/…

@kinginmotion @evernote @obsdmd I use @obsdmd. It can be overwhelming but one doesn't have to be so. 🤷‍♂️

Just curious... Do you currently use... 1. @evernote 2. @obsdmd 3. or both? ...and is Obsidian too overwhelming? (Be honest)

Used Q4 and it works quite good in @FrameworkPuter desktop. 🫡

Unsloth always comes in the clutch !

Node.js addons… in C#? Yep. Native AOT lets you ship shared libs that plug straight into N‑API, no node‑gyp, no Python installs, no C++ detours. Clean interop with UnmanagedCallersOnly, LibraryImport, spans, and zero‑alloc UTF‑8 marshalling. Blog → ift.tt/2mz8B1E

@UnslothAI @NVIDIAAI @NVIDIAAIDev @NVIDIA_AI_PC I appreciate you guys so much!

NVIDIA releases Nemotron-3-Nano-Omni, a new 30B open multimodal MoE model. Nemotron-3-Nano-Omni-30B-A3B is the strongest omni model for its size and supports audio, video, image and text. Run on ~25GB RAM. GGUF: huggingface.co/unsloth/NVIDIA… Guide: unsloth.ai/docs/models/ne…

@FrameworkPuter Video is private

We're live! youtube.com/live/uqFDIR00M…

It's time to seize the means of computation! We have a live launch event coming up on April 21st at 10:30am Pacific where we'll be announcing the products we've been working on over the last 12+ months.

@FrameworkPuter 👀