You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Every app demo based on GPT-3 follows this pattern. You can build the demo in a weekend, but if you invest $20M and 3 years fleshing out the app, it's unlikely it will still be using GPT-3 at all, and it may ever meet customer requirements
The reason why is that parametric models trained with gradient descent make it easy to automate something, but have little ability to deviate from the patterns they've learned. Meanwhile, the real world is full of surprises, and handling it requires the ability to adapt.
Do you think the only solution is neurosymbolic integration? Or could there be feasible neural solutions to this too?
Neurosymbolic models, aka "writing software by hand to do the things DL can't" is already what everyone is doing to address this. It works to a large extent, but it is very labor intensive.
I think that this is the computer version of a general problem: linking depth to breadth in a coherent way is complex. There's not seeing the forest for the trees but there's also the ecosystem and the individual attributes of its inhabitants. Depth of field and field of view.
How much of this is due to the fact that machines are still binary? As complicated as the layers are that we’ve built on top, the gates we’re playing with at the lowest level only understand 1 or 0, on or off.
Approximately 0%. Deep learning models process floating point data. The fact that these floats are discretized (on 16 or 32 bits) does not account for any meaningful part of the generalization problem in deep learning.
The devil is in the detail. Most organisations are unprepared to go the last mile eliminating the gap between SOTA performance and economic feasibility. Usually going for some quick and dirty solution, then getting disappointed by the lack of progress and abandoning the project.
Agree. Many organisations are not set up to use ML effectively in terms of vintage technology, poor data management, offline processes, lack of the right skills. I wonder if any of the offline firms can really transition to the future.
Touché. I would attribute this to uncertainty in the real world. We can’t meaningfully model uncertainty and it will remain that way for some time.
Heh, it’s a very good point. But do you see any solution for this problem?
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
