TomGarden

Agreed! We desperately need an OSS framework of basic modular building blocks for the new generation of vibecoders, if nothing else just to stop the massive waste of resources of having all the models build the same thing hundreds of thousands of times.

Thanks for sharing, but did you even watch your example output[0]? The voiceover is cut off, the drums are out of sync with the rest of the music (rushing).

[0]https://www.youtube.com/watch?v=H6npd3uuqls

I disagree on the voice interaction point. I DO want voice modes to be glorified STT-TTS.

The OpenAI AVM is uncanny and over-eager to me. You can run Anthropic voice on Opus since a handful days and it's incredibly snappy without feeling weird.

This stands out to me too, it seems to be representative of physics the way a video game is representative of physics, likely less so given the non-determinism

Man, whenever I see nostaligic things like this I remember how instantly responsive computers used to feel. Good times.

I opened this with an open mind, but it felt dead and AI immediately to me. Neural Viz, however, had succeeded in making things I find watchable all by himself[0] by working around some of the current weaknesses and by having stellar human writing. Perhaps unsurprisingly, it seems he got burnt out from doing it all alone so it's been a while since the last upload. The Remo Green one is my favorite[1].

[0]https://www.wired.com/story/the-future-of-ai-media-parody-of...

[1]https://www.youtube.com/watch?v=5bYA2Rv2CQ8

That's a great point.

My experience is that if you're not trying to push the use cases very far or very long-running, the last few iterations of Claude work incredibly well for smaller, simpler things, which is already helpful

No, but it was very trivial (Claude did it in maybe a minute). It just made a new launcher that opens OBS, then refreshes/checks all my sources.

Answer this thread if you struggle and I can post when I get to the office

Love OBS! It's so far ahead of proprietary apps in so many areas, it's incredibly impressive

Easily fixable with a script that sets things up as you want it on launch via websocket

I made a little script that reconnects everything the way I want it on launch via WebSocket. Works great

I relate to that, but do you feel it's a fools errand to try and codify your preferences into a list of rules? You'd really only need a 'yes list' for things it can be allowed to do and have it not do anything else

I agree in principle, but the example doesn't hold. Anthropic is only more expensive over API, and they don't train on API usage by default.

https://privacy.claude.com/en/articles/7996868-is-my-data-us...

I've found the max 200 plans to be roughly equivalent, with Claude giving slightly more usage recently, especially after the OpenAI usage slash and upsell to the 500 plan

Max Subscription seems to be the HN assumption most of the time

Not OP, but I used to run out in a few days, for me it was a matter of suboptimal code review practices with many subagents looking at different aspects, often expanded to look at more than the diff.

I mean, same goes for OpenAI. Why point out only Anthropic?

For OpenAI they also have a separate opt-out for stopping training on your codex environment:

"Codex has a separate Include environments setting in Codex settings that controls whether additional context from your Codex environments can be used to train OpenAI models. Changing your ChatGPT setting or opting out through the Privacy Portal does not change that setting." https://help.openai.com/en/articles/7730893-data-controls-in...

I thought Opus 5.5 performed worse across a lot of benchmarks on Max, would love to see it one notch down

GLM 5.3 Flash on 128GB Ram would have to be 2-bit right?

I'm no expert but it seems the consensus is you get a suboptimal experience under 4-bit. Please correct me if I'm mistaken here

Well, that's one way to post your first HN comment

Yeah I'm more excited about the prospect of them explicitly saying it, which hopefully can lead to better places down the line.

I thought DS Flash needed 2bit quant to work on 128GB? I have an m4 max 128 GB and the most reasonable model to run still seems to be qwen's 32B A3B

I partly agree with the sentiment, but this is also the first time I've seen an actual lab announce they optimized for consumer hardware (very expensive, but still consumer).

It's kind of absurd that no non-quantized MOE model exists that's sweet spotted for 128GB Macs, for example. I hope more open labs follow this lead but also offer lower tier models, perhaps specialized on certain types of tasks

The exciting thing about this is in them optimizing for inference on Macs. This, and optimization for other consumer hardware, is where local AI needs to go

Had fun with this on my phone! Any chance of calculating velocity based on speed of pick movement? I'm sure some super basic ASDR and maybe a low-oass would get you to a very satisfying place

Same, gemini for writing is the strongest. Qwen was also surprisingly good last time I did a shootout

Pretty sure Opus 5.5 can do this with code only now. Seen a lot of strong motion design skills floating around

This service is the ground zero of AI psychosis and children talking to chatbots as if it's their friends/lovers.

I'm surprised Disney wants anything to do with it, then again, they wanted to integrate Sora into their streaming offering. Oh, how the mighty have fallen

Oh wow, strange.

I think it's invasive enough that they expose your usage of muse to your instagram mutuals without consent. Parts of my circle are very anti-AI and I would not want them to see me mentioned in a 'x, x and x are already on muse' ad on instagram. This example of yours is mind-boggling in an entirely different way

Brilliant. Laughed out loud

Someone steal this idea for OSS for vibe coders please:

Stop making full apps. Make modular, well optimized pieces of functionality taken from established apps others can pull into their vibe coded software.

I believe this is what Google wants to do eventually once they big data-crunch the most common vibe code cases, and the new OpenAI Intelligent UI seems to be this as well.

Let's end the endless churn of tokens for subpar todo-app back ends etc and make something streamlined for vibe coders that they can build on.

I see no value in specifically looking for vibe-coded software. The ones that are straight copies are going away soon anyway no doubt, once the lawsuits start hitting