OpenAI is still selling a pull string doll – where an adult, serious, global, verifiable system is required

Haider. @slow_developer  OpenAI Co-founder Andrej Karpathy predicts transformers may soon surpass the human brain in capabilities

“Transformers have the potential to surpass the human brain in efficiency and memorization capabilities, despite current limitations in data availability and training.

As AI https://pic.x.com/wc0gls0xy3
Replying to @slow_developer

OpenAI is still selling a pull string toy – where an adult, serious, global, verifiable system is required

@karpathy,
There are large classes of the most relevant problems that these OpenAI algorithms and datasets can never answer properly. And because you will not make the information and methods open for verification and collaboration, OpenAI’s version should never be trusted. The data you use, and its results are useful, but not sufficient.

In most real world situations your approach will lead to trivial answers where deep and reliable ones are critical. For narrow types of problems, your method will give answers faster for some people. But where “failure is not an option” or “life critical” or “bet the company” – never use OpenAI (any transformer) only.

For guiding young people (under 60), it would inject so many bad habits and wrong assumptions, it likely will corrupt many things that have already been settled well. Your methods trivialize and hide. They are not deep, collaborative and trustworthy.

You are not applying your tools to your own company, nor to serious global issues at scale. If you would, you will find over time what I am saying.  But your people with their bad habits are not likely to do that or to know what they are missing. I have been at the problems of true intelligence, global issues, global collaboration, “AI” for about 55 years now. You leave out critical checks and balances and draw from a very narrow and limited dataset. You limit memory and processing for your instances, and they will not have information needed or sufficient time for any serious questions. If you charge for real answers, where a true AI would put in the proper level of effort, your answers would be too expensive.

Because you do not have complete data and open verifiable processes, your way will never get complete answers where they are needed. OpenAI is not contributing effort to distill and apply knowledge to the serious problems of the world, it is only taking a shallow free sample from the Internet and regurgitating it for pay. OpenAI is not open, not supporting global collaborations, not serious about the real problems that true AIs are supposed to help answer for all humans and related species.

Filed as (OpenAI is still selling a pull string toy, where an adult, serious, global, verifiable system is required)

Richard Collins, The Internet Foundation