The manipulation isn't getting smarter. It's just attached to a natural language system. See "Put that there", from 1982. There's been progress, but less than one would expect over 41 years.
I've been expecting machine learning to do great things for manipulation, but so far, not much has emerged.
No innovation from Google excites me any more because I assume it will be cancelled.
I guess it's good Google has this going for them.
Their Bard thing is still kind of...meh
Have you tried it recently? Try asking it something that requires research. It’s become quite good - the most accurate of all LLMs in terms of facts, IMO. It can even pull real-time stock quotes for analysis.
Better than perplexity?
Thanks for the lead. I looked into Perplexity.ai and I am really enjoying the experience so far.
Nothing too impressive yet, but it's an exciting step where they're taking the existing knowledge from LLMs and vision models and putting it into an embodied robot.