Qwen3.6 35B-A3B can now be run locally! 💜
The model is the strongest mid-sized LLM on nearly all benchmarks.
Run on 23GB RAM via Unsloth Dynamic GGUFs.
GGUFs to run: huggingface.co/unsloth/Qwen...
Guide: unsloth.ai/docs/models/...
Excited to launch the accompanying free RLHF Course for my book. To kick it off, I've released:
- Welcome video
- Lecture 1: Overview of RLHF & Post-training
- Lecture 2: IFT, Reward Models, Rejection Sampling
- Lecture 3: RL Math
- Lecture 4: RL Implementation
Landing page: rlhfbook.com/course
STOP SCROLLING. Mandatory watch. It's < 3m.
Share to everyone. 🫡
Gemma 4 E4B (4-bit) completed a full repo audit by executing Bash code and tool calls locally.
Runs on just 6GB RAM.
If you want to understand more about the internals of Gemma 4 and how the model is structured, this visual guide is a perfect start:
newsletter.maartengrootendorst.com/p/a-visual-g...
So, I’ve written a bunch of things on how to do certain things with View Transitions, such as optimizing the keyframes or driving a VT by scroll.
I noticed I repeat a lot of code throughout those experiments … so I bundled that all up in a package for you.
👉 chrome.dev/view-transit...
Google releases Gemma 4. ✨
Gemma 4 introduces 4 models: E2B, E4B, 26B-A4B, 31B.
The multimodal reasoning models are under Apache 2.0.
Run E2B and E4B on ~6GB RAM, and on phones.
Run 26B-A4B and 31B on ~18GB.
GGUFs: huggingface.co/collections/...
Guide: unsloth.ai/docs/models/...
Had a lot of fun at the OpenClaw hackathon at SF yesterday with @georgeck.me. Met a good mix of people - more senior devs than usual. We loved that the focus was to build something real. It nudged us to install OpenClaw on our laptop in a fully locked down docker container.
luma.com/openclaw-hac...
By the way, I gave a talk a few months ago on the future of agents writing code and why Dynamic Workers are ideal for this use case. If you like to consume information in video form, check it out!
www.youtube.com/watch?v=xUj4...
Introducing Unsloth Studio ✨
A new open-source web UI to train and run LLMs
• Run models locally on Mac, Windows, Linux
• Train 500+ models 2x faster with -70% VRAM
• Supports GGUF, vision, audio
• Create datasets from PDFs
• Tool calling, code execution + export
GitHub: github.com/unslothai/un...
I’ve written a bit about my @jj-vcs.dev workflow for reviewing others’ code. Still room to improve the process but I’m finding value in it so thought I’d share. Feedback welcome!
Today I launched blunder.clinic, a realistic daily #chess puzzle app!
Short 🧵
Hacking alongside 500 other developers at the ‘'Built with Opus 4.6: a Claude Code hackathon'
Let's go 🚀
cerebralvalley.ai/e/claude-cod...
We launched HN Companion hncompanion.com, the standalone version of the browser extension that
@georgeck.me and I built a year ago 🚀.
Read HN better with Vim keys and HN-aware summaries that understand thread hierarchy and up/down votes.
Here is a sample post: app.hncompanion.com/item?id=4693...
We’ve been working on a project, and we’d love to hear your feedback on it.
Head over to Hacker News (news.ycombinator.com) and respond with your thoughts on how we should proceed with this project.
We’d also appreciate it if you could share this post so that more people can see it. Thanks 🙏
This is the author of HN Companion, not the bot posting.
After over a year of working on this project, we shared its details on Hacker News, and now we’re featured on the HN homepage! 🎉
Join the conversation on HN itself. I’d love to hear your feedback on how we should proceed.
hncompanion.com
I made a map of 3.4 million Bluesky users - see if you can find yourself!
bluesky-map.theo.io
I've seen some similar projects, but IMO this seems to better capture some of the fine-grained detail
After working with coding agents a lot, I’ve come to see that success isn’t just about how smart the model is. It’s really about getting the hang of how the agentic loop works.
I have documented my understanding of the agent loop here:
george.chiramattel.com/blog/mental-...
Using Mac Mini as a Home Server
medium.com/p/using-my-m...
Agentic coding changed how I build. It took time to understand the concepts and how they fit together.
Then my projects took off - esp. HN Companion hncompanion.com w/ @georgeck.me🚀
I wrote about these concepts and how to use them to build solid apps with quality code!
annjose.com/post/agentic...
This is an example of the 'skill injection' vulnerability:
x.com/ZackKorman/s...
MacOS System Settings showing Google Chrome.app showing many times in the Privacy and Security -> Local Network setting.
This is quite frustrating.
Or is it just my computer acting up? 🤔
For instance, there’s a skill available to find and install additional skills: skills.sh/vercel-labs/...
With this skill, an agent could potentially download a malicious skill, which, upon the next agent launch, would execute with elevated privileges 🤯
With such excitement surrounding this, it’s understandable that everyone is eager to propose solutions. There are a number of skill registries out there. An example is skills.sh.
This could potentially pose a security vulnerability if not addressed effectively.
Having a standard to bring specific instructions into context on demand is a good idea. The next critical step would be to standardize the use of a common folder to load these instructions, such as the .agent folder instead of the .claude folder.
Looks great!
I tried this on a profile with a large number of posts, such as @hncompanion.com
I assume the tool is sampling data rather than going through every post, so the results might not be accurate for accounts like this.
Or am I misunderstanding the results?
Got it. Thanks
Sam, these animations are incredible!
I’ve always admired how you can effectively convey your content through such beautiful animations.
By the way, what tool do you use to create these animations? Do you use libraries like Lottie for playback? I was trying to create some animations myself.
