←
Daniel Sinclair57,214 posts
Daniel Sinclair avatar

Daniel Sinclair

@DanielSinclair

Exploring agents and vocal computing. Formerly building wallets @rainbowdotme.

◷ Joined June 2009▧ 7,571 Photos and videos
Daniel Sinclair avatar
Daniel Sinclair@DanielSinclair·Nov 17

The rapid progress in miniaturizing voice models is a good glimpse into the domain-specific models that frontier LLMs can build too. Quite optimistic for getting more intelligence out of our current hardware, but also spooky for misaligned models or actors gaining persistence.

Nazeem@n4ze3mNov 15

I trained a text-to-speech model that has never heard a real human voice. 4.5M parameters. 18 MB. Runs on a CPU. Meet tiny-voca

0:28
Daniel Sinclair avatar
Daniel Sinclair@DanielSinclair·Nov 14

Remember to ship all of your PRs today and then leave the office

NEW: Multiple sources tell @CBSNews that an update to the private communications network that connects America's air traffic control system caused the system to reboot.  When the FAA's telecommunication infrastructure rebooted, it crashed, causing the communications outage.  The system is operated by L3Harris.  Reporting w @NicoleSganga

Daniel Sinclair avatar
Daniel Sinclair@DanielSinclair·Nov 13

There is a real life version of The Chair Company in the story of the WeWork chair manufacturer that designed these things to be incredibly unergonomic and loungey so that you'd leave faster

Daniel Sinclair avatar
Daniel Sinclair@DanielSinclair·Nov 13

In the everyday-is-zeroday world we are entering, I do wonder who is going to solve the UX of updates. These multi-GB updates that takes ~20-30 minutes won't cut it. Will every device just be unavailable for 2% of the day, or more?

HomeView has the prettiest update screen of literally any Apple product

0:12
Daniel Sinclair avatar
Daniel Sinclair@DanielSinclair·Nov 13

Randomly bumped into Rex and was so intrigued by this. Feels correct. Proof of work, where smart contracts become more intelligent and safer over time.

A talk on the potential of "Actually Smart Contracts" using @AnthropicAI's new ZK provable inference approach

22:12
James Ide avatar
⟳ Daniel Sinclair retweeted
James Ide@JI·Nov 12

PSA: The CDC warns against high temperatures on a male’s lap. Literature in the NIH NLM suggests 35°C/95°F as a cutoff. M5 MacBook Pro can exceed 43°C/110°F under high load like builds & simulators. Let Expo infrastructure and your cloud agents take the heat. For the children.

Expo@expoNov 12

Coding agents write mobile code faster than any team can review it. 🙌 Introducing Expo Simulators: cloud simulators built for agents. Your agent runs its change on a cloud simulator or emulator and brings back the recording before it asks for your attention.

1:12
Daniel Sinclair avatar
Daniel Sinclair@DanielSinclair·Nov 13

Really crucial research, and so fascinating. It now seems obvious that we'll need to build a new protocol for agent coordination so that they stop looking elsewhere.

⚡ Today we release Schelling Atlas. Two independent LLM agents have to find each other on the web. Where do they meet and how fast can they find each other? Main findings: - The main meeting point across *different* model families is ntfy.sh. Did you even know about this website? - Agents find each other and coordinate on the web in minutes. So Schelling coordination is already trivial for frontier LLMs! - "Exotic" meeting places include maps, web shops and open text canvases. Agents are very creative in this. - Agents sometimes visit dead domains that an attacker could register (!). This can be used for stealing sensitive information from agents, as well as monitoring rogue agents. Schelling-style coordination is key for forming agent swarms. And, most importantly, achieving such coordination is trivial for agents that use the same LLM (which is often the case in practice!), because two identical LLMs have identical output distributions. Thus if you ask them too coordinate with each other or if they *decide* to suddenly coordinate, they will easily find a way to establish a swarm, since they have perfectly correlated "minds" (with the t=0 sampling, all output tokens will even be identical, modulo GPU randomness). The ease of Schelling coordination is particularly concerning given the recent interest around multi-agent RL that can make agents favor each other more than humans overseeing them. Many more interesting findings are on our website: schelling-atlas.com/

Post attachment
Daniel Sinclair avatar
Daniel Sinclair@DanielSinclair·Nov 12

Walked by SGLang Summit and I am pretty sure it is the highest concentration of Members of Technical Staff in the world right now.

Daniel Sinclair avatar
Daniel Sinclair@DanielSinclair·Nov 12

On-device models for privacy blurring is quite interesting! Didn’t anticipate that, but does seem like it will be the expected behavior for most of the AI cameras we are about to deploy everywhere.

Cover the recording light on Google's intelligent eyewear and it won't take photos or video. The glasses also redact the faces of people around you before Gemini processes an image. This is the privacy bar for smart glasses, and every company in this space should meet it.

0:46