I'm increasingly thinking the same as our spend on tokens goes up.
If you have HPC or Supercompute already, you have much of the expertise on staff already to expand models locally, and between Apple Silicon and Exo there are some amazingly solutions out there.
Now, if only the rumors about Exo expanding to Nvidia are true..
Revocations works great in theory, and in theory & practice particularly in DOD.
The problem is a ton of certificate authorities consciously chose not to produce validation data previously, created insecure CAs, chose not to cache validation data, had knee jerk reactions to potential exposures, and many industries chose not to invest in technical capability to make revocation data available, performant, resilient, failing-over, failing gracefully, etc.
MITM is now the default for half the enterprise security solutions operating with cert to website “suspected good whitelists” which makes new domains on HN nigh unreadable
Because, comparing vs GPUs
~16k–17k tokens/second per user
<1ms latency
10x power efficiency
20x cheaper production
Model to Si ~ 60 to 90 days
We have every reason to believe SW_to_Si will facilitate improving economics
reply