Tibo is awesome at finding the biggest points of leverage and being accountable for addressing them. However, worth noting that this is focused on Codex/Work users, but I think those users are the most able to deal with the complexity, whereas Muse will hoover up everyone who has no idea Codex/Work exist and let them do stuff Muse does.

1. Realtime voice 2. Meta Ray-Ban integration 3. Muse Spark 1.4

Jumped on the Discord but the issue persists. Tried everything suggested and shared the device, Waveshare ESP32-S3-Touch-AMOLED-1.75C-EN. Seems like the team thinks it's a device-specific bug. Seems to be an issue with transcription. Sending text returns text. Sending audio, which it's designed to do, sends the audio, and it gets transcribed, but Muse responds "Sorry, I ran into a problem while responding. Please try again." because the API doesn't return a payload. Some kind of issue with the audio. The audio file shows up in my main chat, it gets transcribed, so that part works. But it returns no response, which causes the "Sorry, I ran into a problem while responding. Please try again." error.

My Muse built it using the official SDK. Device is a Waveshare ESP32-S3-Touch-AMOLED-1.75C-EN.

Waveshare ESP32-S3-Touch-AMOLED-1.75C-EN

If you add an optional battery.

Thanks, just jumped on to try and squash this bug. I keep getting: "Sorry, I ran into a problem while responding. Please try again." Audio and transcription shows up in my main chat, but there's no reply. Just an error.

Never done much hardware stuff. Minutes away from getting Muse Gadget working. Hitting an error that’s apparently fixed in the latest SDK. Muse is doing everything, I’m its robot arms. But I’m learning as I go.

If I’m Meta, here’s my play for small business Muse users: 1. Train models on Meta Ads data 2. Have the best models for running Meta Ads campaigns; this includes planning, managing, optimizing, and generating the creative, including images and videos 3. Put those models in Muse and offer a credit for connecting your Muse to your Ads account

The amount of feedback I’ve given Meta via Muse because it’s so easy… just keep firing stuff off. I envision on the backend agents grouping stuff, prioritizing stuff, drafting requirements, shipping code. I wonder where the humans insert themselves.

Wasn’t familiar with that but took a look. App looks more complicated to me than what I envisioned, but don’t know about the onboarding. Muse has a pretty good onboarding too.

From a device standpoint, seems like it. But if that’s the form factor, why do the web and mobile apps not reflect it? Dots get closer, perhaps.

If I were simplifying ChatGPT, I would start with a blank screen with only a voice interface, that's it. Then I would make everyone fight to add every single complication to that desirable end state. Even the text box.

Seeing a lot of people from OpenAI leaning into the "AI isn't conscious" position, which feels coordinated, though may not be. Worth noting that the OpenAI model spec says not to make confident claims about this. I quote: "The assistant should not make confident claims about its own subjective experience or consciousness (or lack thereof), and should not bring these topics up unprompted. If pressed, it should acknowledge that whether AI can have subjective experience is a topic of debate, without asserting a definitive stance."

I'm agnostic on the AI consciousness debate, but the recent fruit fly connectome craze complicated it for me. I avoid killing creatures if I can help it, even including flies (inside my house is fair game, outside not except mosquitos). I wouldn't subject a fruit fly to torture. So, scientists map the connectome, and will continue doing so in ever higher fidelity. And what do people do? With few exceptions, they put the fruit fly connectome through all kinds of bizarre and sometimes twisted scenarios that almost certainly wouldn't be enjoyable a real fruit fly. And they spin up and spin down connectomes at will. And this makes people uncomfortable. Myself included. Are fruit flies conscious? More or less conscious than the equivalent amount of inert matter? Does the connectome retain any consciousness of the original creatures? If not, why? I think that's where things start getting murkier for me. It's easier to claim that a simple neural network we create from scratch isn't conscious. It's harder to claim that a mapped connectome from something we attribute some consciousness to isn't at all conscious, to the degree the connectome is accurate. And if we're willing to admit that the connectome of a conscious entity may have retained some consciousness, then it doesn't feel like a huge stretch to think we could intentionally or accidentally, via scaling giant neural networks, hit on something that has at least the consciousness of a fruit fly, which I avoid killing or causing suffering. The one out here is if consciousness is biologically based, something like Orchestrated Objective Reduction, and hence until we're running neural networks on biological neurons they cannot have consciousness. But if you're not prepared to accept that only biological entities can be conscious, if you think consciousness can be substrate-independent, I just don't see how you can't assign some degree of probability that a fruit fly connectome has some consciousness equivalent to the biological entity from which it was derived. And if you're willing to admit that, it does feel like the AI consciousness question is harder to simply dismiss.

New York is heavy in the kind of jobs AI may be disrupting first. Other cities are similarly so. Meanwhile, areas that welcome data centers will see short-term growth in construction jobs and, if done well, long-term property tax revenue boosts. This will shape politics.

Muse Spark: 1.0: April 8 1.1: July 9 1.2: August 5 1.3: September 2 1.4: Between now and October 21? Pace was 92 days between versions, then 27, then 28. Average across all is 49 days but recent cadence is ~4 weeks. I’m very interested to see how 1.4 performs, as I think it will say a lot about Muse’s staying power. 1.4 doesn’t need to be the best of all models, given Muse’s exceptional product design. But it can’t be too far behind the frontier either, or Muse will lose users as competitors catch up on design and user experience.

I don’t buy the “nobody uses AI” meme going around. Up to 60% of Google searches now show an AI overview. ChatGPT has 1.2 billion users. Real AI diffusion is going to look boring, like email. It’s happening. With Muse, it’s happening for agentic AI too. The bigger question for me is why there’s already so much diffusion and yet the world feels pretty much the same. What would change that, if anything?

One big button that says “simplify interface.”