Two moves today that rhyme: Anthropic funds Canadian researchers with no strings attached, and Prime Intellect ships a modular RL stack so anyone can train agents on 6 GPUs.
The pattern? The frontier is being deliberately opened up, not locked down.
Meanwhile, a paper drops on AI models that self-improve while "sleeping." We're watching the full stack get rebuilt, from training infra to memory, in real time.
|
|
In Partnership with
|
·
|
|
|
|
|
|
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
|
|
Top Repo
|
|
|
|
PrismML open-sources a 27B model that fits on a phone at just 3.9 GB
|
|
3,264 Likes
|
|
Bonsai 27B just dropped, and it does something that wasn't possible before: runs a full 27-billion-parameter AI model on your phone. Here's the size problem it solved. A normal 27B model needs 54GB of memory. Even a compressed version sits around 18GB. Way too big for any phone. PrismML squeezed it down by reducing how precisely each number inside the model is stored, trading tiny amounts of accuracy for massive size savings. - Ternary variant: 5.9GB -- laptop-ready, keeps 95% of full performance
- 1-bit variant: 3.9GB -- phone-ready, keeps 90% of full performance
- Multimodal: reads images and text
- Handles coding, reasoning, tool use, and multi-step agent workflows
Why does size matter so much? Because phones only give apps a slice of their memory. At 3.9GB, this model actually fits with room to spare. You can now run private, offline AI agents locally. No cloud. No API costs per step. Your files never leave your device. It's fully open-source under Apache 2.0, available on Hugging Face now.
|
|
|
|
Presented by Agent Field AI
|
|
An open source just ranked #2 on Code-Review-Bench
|
|
It beats CodeRabbit, Copilot, and Devin on Martian's Code-Review-Bench - and it's fully open source. PR-AF runs on any open model, and the harness is why it wins: it plans a review strategy per PR, spawns parallel reviewer agents, verifies every finding against your source, and drops anything it can't prove. No hallucinated comments, ~10x cheaper per review than closed tools. Self-hosted, one drop-in GitHub Action.
|
|
|
|
Top News
|
|
|
|
Anthropic commits $10M CAD to fund AI research across Canada
|
|
2,185 Likes
|
|
Anthropic just dropped $10 million CAD in Claude credits across eight Canadian research institutions. No strings attached. Anthropic confirmed it won't control research directions or findings. So who gets the money? Eight institutions, each receiving $1 million in Claude API credits: - Amii, Mila, and Vector Institute (Canada's three national AI hubs) for safety and reinforcement learning research
- CHEO and CAMH for children's health and mental health AI, including predictive treatment models
- Université Laval to study how AI handles Quebec French and Indigenous languages
- University of Saskatchewan for agriculture, public health, and quantum computing
- University of Toronto to distribute credits through a peer-reviewed process
There's a startup angle too. Amii, Mila, and Vector are joining the Anthropic for Startups program, meaning hundreds of affiliated startups each get at least $5,000 USD in API credits. If you're building something connected to these institutes, that's free Claude access worth applying for.
|
|
|
|
Presented by Airbyte
|
|
Make Your Own AI Agents
|
|
Build production-grade agents today: - Connect the Airbyte SDK, plug into your existing stack & ship agents with full programmatic control
- Works with Claude, ChatGPT, Cursor or any AI assistant
- Ship agents in minutes, not days
|
|
|
|
Top Repo
|
|
|
|
Developer builds AR glasses interface to control and visualize robot sensor data in 3D
|
|
2,036 Likes
|
|
What if you could see a robot's eyes through your own? That's exactly what this open-source project pulls off. It connects Snap Spectacles AR glasses to a Unitree Go2 robot dog, letting you visualize and control it in real space. Here's the core idea: the robot's LiDAR sensor (a laser that maps the surrounding environment in 3D, like sonar but with light) streams live point clouds directly into your AR view. You see what the robot sees, overlaid on the real world in front of you. What this opens up: - Send navigation goals by tapping a floor point in AR, no joystick needed
- Watch the robot's planned path appear as a live overlay in your glasses
- Emergency stop the robot with a single gesture
- Works without a physical robot via replay mode for offline dev
The repo is MIT-licensed, built in Python and TypeScript, and runs on a regular MacBook. Clone it, run ./setup.sh, and you're in.
|
|
|
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 300,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.
Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
|
|
|
Looking to promote your company, product, service, or event to 300,000+ AI developers?
|
|
|
|
|