Today at #GoogleIO @sundarpichai showed some examples of the capabilities of the PaLM 540B language model. For example, you can prompt the model with:
"I will ask a question in Bengali and get English and Bengali answers"
And then give it two examples of this behavior.
(cont)
"I will ask a question in Bengali and get English and Bengali answers"
And then give it two examples of this behavior.
(cont)
You can then ask novel questions in Bengali, and get surprisingly good answers on both English and Bengali:
Why is this surprising and powerful?
Model has never seen parallel sentences between Bengali and English (or any other explicit parallel training data)
Model has never been taught to explicitly answer questions
(cont)
Model has never seen parallel sentences between Bengali and English (or any other explicit parallel training data)
Model has never been taught to explicitly answer questions
(cont)
Model has never been taught to take a Bengali question and answer it in English
Model has never been taught to take a Bengali question+English answer and produce translated Bengali answer
Model can compose all of these skills together to accomplish complex task.
Model has never been taught to take a Bengali question+English answer and produce translated Bengali answer
Model can compose all of these skills together to accomplish complex task.
The same model can perform various coding tasks, logical reasoning tasks, other translation tasks, explain jokes, and various other tasks for which it was never explicitly trained to perform.
You can read more about PaLM here:
https://twitter.com/JeffDean/status/1511023169003405313?t=rvqQubGL4vZeubyAhDnGIQ&s=19
• • •
Missing some Tweet in this thread? You can try to
force a refresh
