armcat

And you can always for example swap out the LLM for GPT-5 or Claude.

There are a few caveats here, for those of you venturing in this, since I've spent considerable time looking at these voice agents. First is that a VAD->ASR->LLM->TTS pipeline can still feel real-time with sub-second RTT. For example, see my project https://github.com/acatovic/ova and also a few others here on HN (e.g. https://www.ntik.me/posts/voice-agent and https://github.com/Frikallo/parakeet.cpp).

Another aspect, after talking to peeps on PersonaPlex, is that this full duplex architecture is still a bit off in terms of giving you good accuracy/performance, and it's quite diffiult to train. On the other hand ASR->LLM->TTS gives you a composable pipeline where you can swap parts out and have a mixture of tiny and large LLMs, as well as local and API based endpoints.

1. Using the skills frontmatter to implement a more complex YAML structure, so e.g.

  requires:
     python:
       - pandas>=2.1
     skills:
       - huggingface-cli@1.x

  skills.lock

That's a bold statement! Of course I know the difference, in one case you are learning from correct/wrong answers, and in the other from a probability distribution. But in both cases you are using some X to move the weights. We can get down and gritty on KL divergence vs cross-entropy, but the whole topic is about "theft", which is perhaps in the eye of the beholder.

Then another lab comes, and "steals" from you - that beautiful, refined dataoil - by distilling your weights using inferior equipment but with a toolbox of ingenuity and low-level hacking tricks. They reach 90% of your performance at 20x cost reduction.

What happens when another lab distills from the distilled lab?

Who is the thief? How far will the Alice go?

So essentially it uses CLI to read all the x,y coordinates, speed, timing, it took screenshots, and combined those together.

My learning from this is - just let the agent do it. Actually trying to interfere with specific conditions and checks lowers the agent's performance. Simply give it a guide.

I don't have any professional experience but I did build a pacman clone in Turbo Pascal (on MS DOS) back in the 90s. Also I've been playing games all my life so have a feeling for what's important - collision detection, walkable/collidable areas, speed/timing, some basic NPC logic etc. So that seemed to have been sufficient in this case :-)

Essentially: `ollama launch claude`

That's true right now, yes. But things are changing rapidly, e.g. there is evroc [1], Mimer [2] and others are popping up too.

it took Google about 15 years of pouring money into Google Docs to be almost as good as the MS offering

I know, and I'm not saying that EU will do any different, but this is not necessarily an absolute gold-standard benchmark, things in principle can be done much faster if you are smaller, nimbler and more focused. The solution to EUs problems is less paperwork and meetings, and more smaller bespoke companies that are laser focused on solving a specific sub-problem. Can they do it? Probably not if they try to create their Google or Microsoft.

[1] https://evroc.com/ [2] https://mimer-ai.eu/

I played around with your avatars and one thing that it lacks is that it's "not patient", it's rushing the user, so maybe something to try and finetune there? Great work overall!

Also: there were 15 000 submissions that were rejected at NeurIPS; it would be very interesting to see what % of those rejected were partially or fully AI generated/hallucinated. Are the ratios comperable?