New video out!! If you’ve been hearing “harness engineering”, this one is for you! And it’s not “just” a new term. "Harnesses" matter more than ever because agents got good enough to be both useful and dangerous. They now can do more than generating text, or token. Useful
@whats_ai
-
HF Storage as Local Filesystem for Agentic Workflows
By
–
Mounting HF storage as a local filesystem is genius for agentic workflows. No more downloading entire datasets just to process a few files. Will play around with this for sure!
-
Real Lab Robotics Demo Without Teleoperation Shows Practical Results
By
–
One take, no teleop, actual lab tasks. That's the kind of robotics demo I want to see. Most robot demos are carefully staged, this looks like it actually works in a real setting.
-
Q-Priming Improves Agent Reliability Through Clarification Requests
By
–
The Q-priming part is really interesting. Making models 5x more likely to ask for clarification instead of guessing wrong is exactly what we need for agentic workflows. Right now most agents just confidently charge ahead with bad assumptions.
-
Distilling Opus reasoning into 27B local model for production
By
–
Distilling Opus reasoning into a 27B model that runs locally is incredible. The fact that tool calling works too makes this actually useful for production, not just benchmarks. Super excited to test its limits 😀
-
Internal AI Cheatsheets for Claude Writing Tools
By
–
I just added all our internal cheatsheets at Towards AI in Markdown so Claude can also refer to them when building for us You (or your agents) can use them as well! We basically have 3 main markdown files we use: (1) our AI slop cheatsheet. Incredible for all writing
-
ColBERT retrieval model balances performance and computational efficiency
By
–
ColBERT seems to hit a nice sweet spot for retrieval. Easy to train and still gets great results, especially when you don't want to go full cross-encoder compute 🙂
-
AI Generated Applications Flooding Algorithms, Burying Quality Submissions
By
–
The worst now is all the AI generated applications too.. The algorithm just favours volume so much that real applications get buried. Interesting to see where this goes.
-
Capability Safety Tension in AI Model Interactions
By
–
The tension between capability and safety in computer use is real. Negotiating (with frustration) with the model on what it can and can't do is the new normal for power users.
-
Inference Scaling: Where Real AI Money Goes
By
–
Inference scaling is where the real money is going to be. Training happens once, inference happens millions of times. Makes sense that NVIDIA is pivoting hard.