mojosmojo

5 ms·

1. 1.▲
   by mojosmojo 29d ago

   extremely cool. I’ve had Claude walk model architectures to debug Loras and fine tunes, but this is delightful
2. 2.▲
   by mojosmojo 2mo ago

   Their enterprise customers pay via metered actual use.
3. 3.▲
   by mojosmojo 2mo ago

   making sure ML isn’t ever pigeonholed
4. 4.▲
   by mojosmojo 3mo ago

   Incredibly thoughtful. This essay gives that very rare sense of being well reasoned, gods at forest and trees, and sitting atop a shit ton of domain expertise.
5. 5.▲
   by mojosmojo 3mo ago

   As an Aubrey-Maturin lifelong re-reader, I cannot wait to play this.
6. 6.▲
   by mojosmojo 4mo ago

   They have all switched to usage plus cheap seats based costs for enterprise contracts. the seat costs are typically 20-35% of total spend.
7. 7.▲
   by mojosmojo 5mo ago

   iirc, there is a bunch of formal machinery you need to define probability distributions for situations such as infinite outcomes (eg what is the probability that a random real number between 0 and 10 is less than 3?)