@actualinc
actual.inc Actual Computer › Case Study : GB300 and Nemotron 3.5 Super VL Department-scale AI from one NVIDIA DGX Station: Nemotron 3.5 Super VL at 8,000+ tokens per second, shared by your whole team through api.actual.inc. 
actual.inc Actual Computer › Introducing toks Actual Computer's first public software release: toks, the best tokenizer on earth. The exact same ids as Hugging Face tokenizers, 13x to 151x faster on one CPU core. 
actual.inc Actual Computer › The Fat Long Tail of Neoclouds There are maybe 50 companies on earth that can fill a GPU datacenter, and maybe 500,000 that can fill a rack. The neoclouds buy the pile and rent it back. Actual arms everyone else: one install turns... Actual Computer CEO @Tom_A_Lynch explains why Fable's restrictions made people realize they shouldn't rent intelligence they can own:
"The bottom line is always the thing that pushes people the most. But I think a lot of people actually got really into local models when they saw the barriers put on models like Fable."
"If you asked for really strong kernels, or you asked for certain types of software, or you asked for how to reverse engineer an open source model, Fable would say no and throw you to Opus."
"Having certain spikes of capability completely shut off of people in their face, it opened a lot of people's eyes that you wanna own this capability."
@actualincView quoted post on X
