Skip to content
LiveNext 14:54:37
DevelopingResearch

OpenAI publishes its pretraining perplexity and urges shared limits

In a Nov. 22 post, OpenAI Chief Scientist Jakub Pachocki publishes the lab's internal log-perplexity figures (GPT-3.5 about 0.99, GPT-4 about 0.91) and says cybersecurity and math capabilities emerged below 0.8. He proposes shared perplexity-based limits for frontier labs, and says AI could sharply speed pretraining research as early as 2027.

HEAT
OUTLETS
0
FIRST SEEN
LAST SIGNAL

Primary source

OpenAIopenai.com/index/perplexity-and-intelligence/

Coverage · 1 article, oldest first

Share
Share card: OpenAI publishes its pretraining perplexity and urges shared limits

Share this article

Copying the image…

Open any post or message and paste it in. The link to TopFive is printed on the card.

Post it on

XLinkedInHacker NewsEmail
← The whole wire