what is the business model here
? tapeout for every new model? hehe
maybe the demand for inference is so much that this strategy has a way forward

Sally Ward-Foxton
@sallywf
4d
AI chip startup Taalas @taalas_inc is showing off a chip that can do 16,000 tps/user on Llama3.1-8B, many multiples of its nearest competitor. The catch? The chip ONLY runs Llama3.1-8B, and a model like DeepSeekR1-671B would need 30 separate tapeouts:
https://t.co/IJuprQZqaE
1
0
1
I guess Sam has never seen an Indian political rally, lol
1
0
0
“once told a story of how one researcher (presumably from his lab or from one of his coworker's lab) intentionally mispelled citations so that the author's citation count does not go up on google-scholar, citeseer etc.”
comment on hackernews
academia is wonderful place they said when I was young 🤣
0
0
1















