RynnBrain
github.com60 points by jsemrau 5 days ago
60 points by jsemrau 5 days ago
I think "embodied" is a word with decades of past associations in cognitive science and AI and if you're claiming that your model is embodied you should try to back that up with specific evidence. This seems cool but I don't think they live up to their self-description
[dead]
(deleted)
It wasn’t that long ago that we were experimenting on with llama derived models. There is still a lot happening with open weight models, but you may have to look at those models meant for targeting edge devices. In my own opinion, small models that can do useful work are going to become a much bigger deal in domains where data privacy is a requirement. Most businesses were technology is not the core mission do not want to host their own frontier models.
Whether one needs a custom frontier model or tuned model or an existing one depends on the task at hand. Custom tasks that are not covered by existing models will inevitably require a new frontier model independent of whether the business is a technology firm or not. An LLM or a basic vision model isn't suitable for all tasks. For example, imagine a model that outputs coherent architecturally sound designs for bridges or electrical systems that a mere rule-based system cannot output.
Objectively?
NOBODY in "the West" has anything to contribute?
Get out of here with this.