Alex Karp: Frontier Models Are Not Delivering Outcomes [video]

Discussion (4 Comments)Read Original on HackerNews

bfeynman•about 3 hours ago

ah yes, the company that spent 20 years losing money as basically embedded consultant building the equivalent of "nested if" intelligence and software thinks that AI models that can replicate same functionality now with little cost are not delivering outcomes.

toomuchtodo•about 3 hours ago

Grifter (Karp) game knows game.

dzonga•about 4 hours ago

TLDR:

Alex Karp: Dario & Sama are overselling the value of frontier LLM models. Enterprises & GVT are not getting value - LLMs labs are a potential vulnerability layer as valuable enterprise data can be shared & used for training.

adrian_b•10 minutes ago

His harsh criticism of OpenAI and Anthropic is intended to be a sales pitch for a new product of Palantir, "Sovereign AI OS", which is promoted under the slogan "The future of AI is on-prem."

This is claimed to be "a turnkey AI datacenter — from hardware procurement to application deployment".

Obviously, a product conceived for "on-prem AI" is the opposite of the cloud strategy of OpenAI and Anthropic, so there is no wonder that Alex Karp suddenly emerged as their biggest critic, who exposed that enterprises are tricked into "paying for tokens that create no value" and that the AI companies "are stealing" their customers data.

While my opinion about Palantir is very low, in this case I have to agree with how they promote their product, i.e. that the only way in which you can rely on AI for a business is to run an open-weights model on your own hardware.

Their product appears to be made in cooperation with NVIDIA, so it is based on NVIDIA hardware, and for the open weights LLM they suggest to choose models from the Nemotron family published by NVIDIA, with additional post-training to fine-tune them for the needs of the customer.

Alex Karp: Frontier Models Are Not Delivering Outcomes [video]

⚡ Community Insights

Discussion (4 Comments)Read Original on HackerNews