Representative of OpenAI Japan, Nagasaki
The future AI model called 'GPT Next,' as mentioned by the representative of OpenAI Japan, is expected to evolve nearly 100 times based on past performance. The technology of AI grows exponentially unlike traditional software.
The slide clearly states 'GPT Next' in 2024.
This 100-fold is probably imagining the effective computation of +2OOMs, which includes improvements in architecture and learning efficiency, rather than just scaling computation resources.
GPT-4 NEXT, which will be released this year, is trained with a smaller version of Strawberry using approximately the same amount of computing resources as GPT-4, and the effective computation capacity is imagined to be 100 times higher.
Orion, which has been a hot topic recently, is expected to be released next year after being trained for several months with H100 10k-class, adding a 10-fold scaling of computing resources, and reaching +3OOMs compared to GPT-4.
The slide clearly states 'GPT Next' in 2024.
This 100-fold is probably imagining the effective computation of +2OOMs, which includes improvements in architecture and learning efficiency, rather than just scaling computation resources.
GPT-4 NEXT, which will be released this year, is trained with a smaller version of Strawberry using approximately the same amount of computing resources as GPT-4, and the effective computation capacity is imagined to be 100 times higher.
Orion, which has been a hot topic recently, is expected to be released next year after being trained for several months with H100 10k-class, adding a 10-fold scaling of computing resources, and reaching +3OOMs compared to GPT-4.
Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only.
Read more
Comment
Sign in to post a comment