Palmier team ftw. Great execution. I love editing videos like this

I remember playing with claude to try and generate videos (or tools to generate videos) a year ago. I never reached something like this, and I think the models were already capable back then.

Let's go! We're at a point where models are smart enough and generation models are good enough where a product like things finally makes sense. Exciting times!

ok bro this is fire!!! I wish I would not be able to interact with the playback of the video, I think what's fun is knowing that we are all listening to the exactly the same

Ty!! That's the idea but it needs some work to get there. It's possible though.

Everything you guys write is absolute fire. I could not agree more :)

What are your digital 'employees' doing? Did they replace any humans or was there nobody before?

Looking into OpenClaw, I really do want to believe all the hype. However, it's frustrating that I can find very few, concrete examples of people showcasing their work with it.

Can you highlight what you've managed to do with it?

That ol' question. Reminds me of new cryptocurrency opportunities of 2019. "Few understand this" as they say.

set A1 = 3 set B1 = 4

set C1 = A1 + B1 = 7

now change C1 = 14 expected A1 = 6 expected B1 = 8

what it did A1 = 7 B1 = 7

great

Why do you think that 6+8 is a better solution than 7+7?

When C1 changes from 7 to 14, that’s a scalar change. The least-assumption, information-preserving update is to scale both inputs by the same factor.

It preserves an implicit relationship, ratio, between A and B.

Seeing the backend nodes generate feels like magic

Glad to hear that. We want to make it as logical and white box as possible. Have you tried adding custom behavior after the first generation?

It's completely not the same but the first thing that crossed my mind was the comic of "How standards proliferate"

[0] https://imgs.xkcd.com/comics/standards_2x.png

Also I think if I'd use your product, I'd like to be able to host the model elsewhere in case I don't like the platform anymore :)

That’s fair! It has been mentioned before so we‘ll likely build that into the platform. Would you like us to upload your model to your huggingface account, download the weights or choose an inference provider we then upload it to?

I (not GP) would like to be able to choose between the options. Inference provider isn't super necessary though (can do that through huggingface).

Thanks for stating your preference! This is something we can incorporate into the platform.

Seems like download the weights would be the most flexible option. HF and inference providers would be nice to have.

I love the idea of the product! I would trust your solution to be the best for very simple use cases but not for multistep or ReAct agents. Any thoughts / insights on that?

I think the demo could be more exciting, the voice of the person talking sounds like he's bored haha

Ha - here's the advice I give to YC startups about making demo videos for HN:

"What works well for HN is raw and direct, with zero production values. Skip any introductions and jump straight into showing your product doing what it does best. Voiceover is good, but no marketing slickness—no fancy logos or background music!"

I guess there's zero production values and zero production values...

Totally agree. Raw is great, but energy matters too. If the person sounds bored, it's hard to get excited about the product—even if it's amazing. Passion is contagious.

That's true, thanks for the feedback! In the end, it wasn't boredom, but the long work - put too much energy into the platform ;) Taking it to heart for the next one!

Well... we took the rawness to heart, that's clear!

Which was exactly correct!

Yes, great point. We are currently working on multistep RL. The big problem with the trivial approach (give a single reward to the entire (ReAct) trajectory) is that the model receives a weak learning signal per decision (called credit assignment problem in literature), i.e. the individual decisions are not properly taken into account, which will then make the training unstable. I guess this has been an unsolved problem for a long time; however was not really looked at since generalist “planning” agents were not a big thing in RL until o1/DeepSeek.

IMO, the most promising approach to this is something along the lines of MA-RLHF (https://arxiv.org/abs/2410.02743) but adapted to the real world, i.e., spitting up the reward model to grade individual actions inside the trajectory to reduce the “attention distance” between the reward and the decision.

This is super dope! Are you looking to hire? Your product made me super excited.

Thanks a lot.

Yes, we will be hiring for founding engineers pretty soon. Please reach out to founders@vocera.ai if you are interested