Connor Dickie

Tech Generalist. ex @medialab, Co-Founder @Synbiota & Kameraflage. ex @ChooseMuse. ex Fellow @Mozilla & @Synberc. SXSW winner. Inventor Samsung Smart Pause

[Connor Dickie @c0nn0r]

In 2004 I designed my first wearable AI memory prosthesis: smart glasses that knew when someone was looking at you. They automatically paused your chats and blogged your interactions. Like most of today’s “AI memory” toys, privacy wasn’t even in the blueprint. ACM CHI demo 👇

[Connor Dickie @c0nn0r]

@danfaggella @samswoora No argument from me on this topic. That’s why I’m hoping for mitochondria 🤣

[Connor Dickie @c0nn0r]

@danfaggella @samswoora Symbiosis or merger imply we lose ourselves too. Symbiosis is abdication. Merger is amputation

[Connor Dickie @c0nn0r]

Every new medium starts by using an older one as its content. McLuhan noted: “The content of any medium is always another medium.” Film borrowed from theater; TV borrowed from film. Computers simulate them all. But LLMs go further: their current content is humanity itself. 🧵

[Connor Dickie @c0nn0r]

@danfaggella @samswoora I think our realistic best hope is that we become like mitochondria to the AGI

[Connor Dickie @c0nn0r]

@mkvenkit If you like that MIT project check out MiSTer FPGA. Open source FPGA effort with dozens and dozens of computer, console, and arcade platforms

[Connor Dickie @c0nn0r]

@Starship_20 Reporting for duty

[Connor Dickie @c0nn0r]

@Dirtride2005Men I made a similar DIY setup a long time ago. It was awesome!

[Connor Dickie @c0nn0r]

@yacineMTB Thx! Yes, this old version uses mediapipe. I have an in-progress prototype using my in-house ML calibration-free, lighting-independent eye-contact sensing model. Demoing this summer (I hope) 🤣

[Connor Dickie @c0nn0r]

@yacineMTB Get you a multi voice-agent setup like this:

[Connor Dickie @c0nn0r]

@tomasmas Are we hearing audio directly from the device in this demo, or is it transformed/reconstructed in any way?

[Connor Dickie @c0nn0r]

This method becomes truly magical when we use real eye-contact sensing rather than just head-pose. Currently working on a foundation model for eye-contact (not eye-tracking!)

[Connor Dickie @c0nn0r]

This is neat, but I'm itching to see a model that can do proper multi-interlocutor turn-taking like the demo I made last year (vid in replies). Since the TM demo also incorporates vision, should be doable, but seems people are not thinking about this rw problem

[Thinking Machines @thinkymachines]

People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way.

We share our approach, early results, and a quick look at our model in action.

https://t.co/AFJZ5kH7Ku

[Connor Dickie @c0nn0r]

Here's another example of the same system showing how it can solve the issue of mis-interpreting non-user (environmental) speech

[Connor Dickie @c0nn0r]

@thinkymachines Nice! Is it able to do multi-interlocutor turn-taking like this system from last year?

[Connor Dickie @c0nn0r]

@mov_axbx Happy to help you test if needed 🍻

[Connor Dickie @c0nn0r]

@kandowontu @RealSunlitSpace Love this! Could FX3 get us an anamorphic widescreen option?

[Connor Dickie @c0nn0r]

@emax @claudeai Lots of manual use or large autonomous runs?

[Connor Dickie @c0nn0r]

@safishamsii Already have plans to integrate with my local swarm test harness. I'm using Nvidia's free models rn 😅

[Connor Dickie @c0nn0r]

@safishamsii Graphify is even more useful after adding a feature that labels communities w name of origin node. Sidebar much better for navigation! 💪