Enough AI copilots, we need AI HUDs

Copyparty – Turn almost any device into a file server

Tao on “blue team” vs. “red team” LLMs

Sign in with Google in Chrome

I designed my own fast game streaming video codec – PyroWave

Wikimedia Foundation Challenges UK Online Safety Act Regulations

Different Clocks

Pony: An actor-model, capabilities-secure, high-performance programming language

We&#x27;re living in this weird asymmetry where companies use AI to talk to us, but we&#x27;re still manually dialing them. Companies everywhere are adopting AI voice agents lately. Big retail, family dentist clinics, local pharmacy. This year, I&#x27;ve been in a few calls where it&#x27;s super natural sounding AI, which has been pretty cool to experience. But then it got me thinking - why are we, the consumers, still the ones making calls if they&#x27;re using robots for theirs?So I built Piper: basically AI that makes phone calls for you. You tell it what you need (book appointment, check on an order, dispute some charge, whatever), and it handles the entire conversation while you do actual work. Right now it&#x27;s a web app, Chrome extension is pending approval but soon you&#x27;ll be able to click any phone number anywhere and just let Piper handle it.Technical stuff that was harder than expected:Latency - every millisecond counts in conversation, had to optimize around kv cache, got it down to ~1000ms to first word over PSTN for telephony, which feels pretty naturalKeeping the voice agents on track - built custom context engineering logic that constantly updates the agent&#x27;s situational awareness, so it knows when it&#x27;s been transferred, when it&#x27;s on hold, etcDone ~50 successful calls with early testers so far. Main failures are when they need complex verification or documents. Also had to take down our IVR navigation temporarily :&#x2F;, found some edge cases that were causing unnecessary transfers but working on fixing that.I really think we&#x27;re heading toward this world where AI talks to AI for most routine things, and phone calls might be the first real example of this happening at scale!you can check out the a voice demo on our website. <a href="https:&#x2F;&#x2F;pipervoice.com" rel="nofollow">https:&#x2F;&#x2F;pipervoice.com</a>

Show HN: Companies use AI to take your calls. I built AI to make them for you

Rotring 600 Ballpoint Pen

Nothing to watch – Experimental gallery visualizing 50k film posters

Age Verification Laws Send VPN Use Soaring–and Threaten the Open Internet

Tour de France confronts a new threat: Are cyclists using tiny motors?

My 2.5 year old laptop can write Space Invaders in JavaScript now (GLM-4.5 Air)

Truchet Tiles

After training calculator agent via RL, I really wanted to go bigger! So I built RL infrastructure for training long-horizon terminal&#x2F;coding agents that scales from 2x A100s to 32x H100s (~$1M worth of compute!) Without any training, my 32B agent hit #19 on Terminal-Bench leaderboard, beating Stanford&#x27;s Terminus-Qwen3-235B-A22! With training... well, too expensive, but I bet the results would be good!*What I did*:- Created a Claude Code-inspired agent (system msg + tools)- Built Docker-isolated GRPO training where each rollout gets its own container- Developed a multi-agent synthetic data pipeline to generate &amp; validate training data with Opus-4- Implemented a hybrid reward signal of unit test verifiers &amp; a behavioural LLM judge.*Key results*:- My untrained Qwen3-32B agent achieved 13.75% on Terminal-Bench (#19, beats Stanford&#x27;s Qwen3-235B MoE)- I tested training to work stably on 32x H100s distributed across 4 bare metal nodes- I created a mini-eval framework for LLM-judge performance. Sonnet-4 won.- ~£30-50k needed for full training run of 1000 epochs (I could only afford testing )*Technical details*:- The synthetic dataset ranges from easy to extremely hard tasks. An example hard task&#x27;s prompt:&quot;I found this mystery program at `&#x2F;app&#x2F;program` and I&#x27;m completely stumped. It&#x27;s a stripped binary, so I have no idea what it does or how to run it properly. The program seems to expect some specific input and then produces an output, but I can&#x27;t figure out what kind of input it needs. Could you help me figure out what this program requires?&quot;- Simple config presets allow training to run on multiple hardware setups with minimal effort.- GRPO used with 16 rollouts per task, up to 32k tokens per rollout.- Agent uses XML&#x2F;YAML format to structure tool calls*More details*:My Github repos open source it all (agent, data, code) and has way more technical details if you are interested!:- Terminal Agent RL repo- Multi-agent synthetic data pipeline repoI thought I would share this because I believe long-horizon RL is going to change everybody&#x27;s lives, and so I feel it is important (and super fun!) for us all to share knowledge around this area, and also have enjoy exploring what is possible.Thanks for reading!Dan(Built using rLLM RL framework which was brilliant to work with, and evaluated and inspired by the great Terminal Bench benchmark)

HN