Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
WithinReason
on April 4, 2022
|
parent
|
context
|
favorite
| on:
Pathways Language Model (PaLM): Scaling to 540B pa...
Based on their 3rd figure, it would take an approximately 100x larger model (and more data) to surpass the performance of the best humans
drusepth
on April 4, 2022
|
next
[–]
Its performance on answers to chained inference questions (on page 38 of
https://storage.googleapis.com/pathways-language-model/PaLM-...
) has already surpassed the performance of this human.
danuker
on April 4, 2022
|
prev
|
next
[–]
I placed a transparent plastic ruler on the screen to come to the same conclusion, then I saw your comment.
WithinReason
on April 4, 2022
|
parent
|
next
[–]
Your methodology is much more sophisticated than mine
sib
on April 4, 2022
|
root
|
parent
|
next
[–]
My dad, who worked on jet engine production many decades ago, would refer to MIL SPEC EYEBALL Mk I. (I
think
he was kidding...)
DantesKite
on April 5, 2022
|
prev
[–]
Is there even enough text for that? Wild. Although after a while, maybe you can generate some.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search:
