4 ms·
Does MS think they have a big moat in these? It won't be long until we have an open version , and although they need a lot of compute , maybe in 1-2 years they
by seydor 4y ago
Does MS think they have a big moat in these? It won't be long until we have an open version , and although they need a lot of compute , maybe in 1-2 years they ll have come down in size enough to be usable. Is it the training data?
- galleywest200 4y agoFrom what I have read with these models it is in-fact the data that can be the most useful piece of the puzzle.
- seydor 4y agobut the data is public datasets, plus some human training that s not the hardest part
- HDThoreaun 4y agoMicrosoft can get just as much data as openAI has.
- cma 4y agoYeah, Microsoft doesnt even make something like Google's TPU, where's the moat? Datacenter design? Involuntary Bing users?
- peanuty1 4y agoIs the compute too big to run locally on a laptop with Nvidia RTX 3080?
- Der_Einzige 4y agoYa but how long will it be before I have 400GB of VRAM on a single GPU? Until then, I can't run these models at home even if the weights were published.
- seydor 4y agothere is already work to reduce the size of these LLMs and it can only get better from now on
- Der_Einzige 4y agoQuantization and pruning have dramatic impacts on model quality, no matter how often they say that it doesn't in papers.