OpenAI says publishing ChatGPT code was a mistake

Contrary to its developer’s name, GPT-4 code will be closed. According to OpenAI principal researcher and co-founder Ilya Sutskever, opening the language documentation was a mistake from the start, given the market potential of the platform, the “brain” of ChatGPT.

In a recent interview, Sutskever stated that the future of Artificial Intelligence is very promising, both in terms of application possibilities and job opportunities. Initially, the main goal of the company was to “create value for everyone, not just shareholders”.

Developers criticize new approach

So far, the developer community hasn’t been quite happy with OpenAI’s new approach. answer BorderThe lack of access to an AI’s training process “makes it difficult to determine where a system can be used safely,” said Ben Schmidt, vice president of Nomic AI.

I guess we can turn off ‘Open’ AI: The 98-page article promoting GPT-4 proudly declares that they don’t reveal *nothing* about the content of the training sets. pic.twitter.com/dyI4Vf0uL3

— Ben Schmidt / @benmschmidt@vis.social (@benmschmidt) 14 March 2023

“To make informed decisions about where this model won’t work, you need to have a broader idea of ​​what assumptions are being fed into the algorithm. I wouldn’t trust a driverless car whose training didn’t involve snowy weather,” Schmidt continued.

Without access to detailed documentation, it becomes more complex to develop fixes in case of errors or critical failures. Another sensitive issue is Unauthorized use of copyrighted materials.

A significant portion of the content provided to the algorithm came directly from the internet to train many AIs, including early versions of ChatGPT. Without strict control of the educational process, indiscriminate use of copyrighted material is possible, which may constitute a crime.