AI News

[JB47394]

They are back at it again with another paper for a new architecture called Griffin, which they claim outperforms transformer architecture.

V13, here we come.

 

[DanCar]

What tasks can the Tesla Robot accomplish without being operated remotely?

More and more each day.

 

[DanCar]

An incredible $100 billion bet to get rid of Nvidia dependence — tech experts reckon Microsoft will build a million-server strong data center that will primarily use critical inhouse components

"Stargate" has a planned launch date of 2028

www.techradar.com

Microsoft has a 4 year plan and $100 billion to replace nVidia with in house chips. Google started that plan 10 years ago.

 

[DanCar]

AI wearable computer gets a big thumbs down. Thumbs up on potential.

 

[DanCar]

Quite amazing that to be released open weights model reaches gpt 4 scores.

https://twitter.com/x/status/1780997414071181370

 

[diplomat33]

Shashua (Mobileye) revealed Mentee, a humanoid robot, using "AI integration at every layer, from Sim2Real machine learning to NeRF-based algorithms and LLMs, we've achieved a complete end-to-end cycle for tasks":

https://twitter.com/x/status/1780611499133685889

 

[ThomasD]

If you throw an anomaly at the Robot can the Robot overcome and adapt to the change? For instance if you replaced the Vegetables with Eggs would the Robot know not to drop the Eggs in the container and place them gently in the Container without being programed to do so?

 

[diplomat33]

If you throw an anomaly at the Robot can the Robot overcome and adapt to the change? For instance if you replaced the Vegetables with Eggs would the Robot know not to drop the Eggs in the container and place them gently in the Container without being programed to do so?

Without any prior training, probably not. You would train the robot to know how to do that. First, you would train the robot to know the difference between eggs and vegetables. Then, you would train it not to drop the eggs.

 

[DanCar]

If you throw an anomaly at the Robot can the Robot overcome and adapt to the change? For instance if you replaced the Vegetables with Eggs would the Robot know not to drop the Eggs in the container and place them gently in the Container without being programed to do so?

Large language models (LLMs) are evaluated on their ability to generalize. So yes, in the future when LLM tech is built into robots, they will have a bit of common sense.

The same argument can be made for driverless vehicles. How do you train it for common sense? Obviously you don't want a car to hit anything. You can then program it not to move, but that defeats the purpose of the car. Another common sense approach is to get out of the way, rather than clogging things up. So yes, robots and robot cars will need to be programmed with the ability to generalize / common sense.

 

[kabin]

Shashua (Mobileye) revealed Mentee, a humanoid robot, using "AI integration at every layer, from Sim2Real machine learning to NeRF-based algorithms and LLMs, we've achieved a complete end-to-end cycle for tasks":

https://twitter.com/x/status/1780611499133685889

Nice work for a part time 2 year effort. I was looking for an entry - Think: Crap, this 5lbs plastic tote is heavy and I stumbled again.