What will these *generic* research labs do in the world where Meta, Snowflake, Databricks and others release their very competitive models to the world completely free of charge? Why would someone use their models, and presumably pay extra for royalties? What's Anthropic's durable advantage? I thought it's the people who are very smart (many from OpenAI) and/or know the secrets of the compute multipliers, but seeing that Meta released a model of a comparative intelligence, it appears that either secrets are much easier to steal than data, or a lot of people are coming to the same insights ~at the same time. Is the durable advantage *the data*? It would explain why snowflake, a data company, #tech intends to release everything, including theory, but except data. #ai Is the durable advantage *specialization*? Domain specific application of AI?
Generic do-it-all LLMs will fade. Waste of computing resources to host a model that can answer history questions, help one to cook a dinner or solve a leetcode problem. The future are smaller, domain-specific LLMs. From the performance perspective it doesn’t matter much if you’ll run it through GPT4, DBRX or snowflake model. The biggest issue is where to get a high quality data from to train your LLM.
That and scale of compute available
They have no moat
OpenAI has no moat and Nvidia’s revenue trajectory will not continue. Just my 0.02
That was the point of googles memo a while ago.
Thanks, we’ll take their 80B valuation