Peter Bakkum
@pbbakkum
API Capabilities @openai, for fun: https://butterfi.sh — Previously internal ledger @stripe, platform @quizlet
A small project I'm publishing for Neovim people: 🐠 butterfish.nvim This is a Neovim plugin to use language models for very fluent code editing. It veers away from the chatbox or codebase Q/A approaches and instead focuses on specific local edits, for example, select and…
🎙️ Hopping on a voice AI panel with @pbbakkum at 10:15AM pacific! We're doing a fireside chat on "Building Real-Time, Reliable Voice AI: From Simulation to Production" - basically all the messy stuff that happens when you try to make voice AI actually work in the real world.…
I really enjoyed this discussion and love what @adhsu and Speak are doing with language learning. We’re approaching fully automated on-demand language learning over voice.
🆕 Personalized AI Language Education, with @adhsu, CTO of @Speak! youtube.com/watch?v=tIVKgz… For the first time, Andrew tells the story about building the next generation of AI Language Learning, starting pre-Transformers, getting advice from @karpathy, and why they went bet hard…
Realtime API
Releasing Shoggoth Mini!🐙 Soft tentacle robot meets GPT-4o & RL. I built it to explore the boundaries of weird: expressiveness, aliveness, and AI embodiment. Blogpost: matthieulc.com/posts/shoggoth…
Running python in prod is a sinful moral failure and punishment is an eternity debugging asyncio
It's amazing how bad and annoying auth is on the modern internet and we accept it. You are you, and yet you log in like 10 times a day. The GOOD solution is to log into 1password, sometimes you're texted a number. It's fundamentally silly, internet TSA shoe removal.
Voice AI hardware + WebRTC + Pipecat Here's a @pipecat_ai SmallWebRTCTransport client for the ESP32-S3 family of embedded devices. The SmallWebRTCTransport is a serverless WebRTC connection designed for voice AI. Link to code and resources below ... @aconchillo wrote the ESP32…
Combining various compositing effects in this one, I like the result. More to come.
Our latest speech-to-speech model is faster, more accurate, and excels at function calling. Watch @promptshant and @bfioca build a realtime voice agent that can search the web and hand off tasks to reasoning models with full context.