OpenAI, Meta Ready Next-gen AI Models
- By Paul Mah
- April 10, 2024
Executives from OpenAI and Meta this week signaled that they are about to release new AI models that can reason and plan, according to a report on the Financial Times.
Meta had publicly confirmed that it will begin rolling out Llama 3 next month; a separate report on The Information noted that small versions of Llama 3 could be launched as early as next week.
In addition, OpenAI said it will release its next model, expected to be called GPT-5, “soon.” In an interview with the Financial Times, OpenAI chief operating officer Brad Lightcap says the next generation of GPT would show progress in solving hard problems such as reasoning.
“We’re going to start to see AI that can take on more complex tasks in a more sophisticated way. I think we’re just starting to scratch the surface on the capability that these models have to reason,” he said.
The pace of progress around generative AI is accelerating, as work at Meta, Google, and Anthropic picks up. Other AI nonprofits or technology firms such as Cohere and Databricks have also released their own LLMs, often incorporating new groundbreaking techniques.
The pace of AI research is also skyrocketing, with a tsunami of groundbreaking papers on AI released much quicker than can be read, according to Matt Johnson, the managing director for AI and data at Temasek-backed digital transformation firm Temus.
He did a study of AI-related scientific papers uploaded to arxiv.org and found that the compound annual growth rate (CAGR) of such papers uploaded shot up from 44% in the nine years before end-2022 to a staggering 299% since December 2022.
Johnson noted that there were more AI papers uploaded in March 2024 (11,363) than were uploaded in the three years from 2014, 2015, and 2016 combined (11,237).
“While in the past, there was time to absorb, reflect, and contemplate the intricate dance of algorithms and theories, that luxury has all but evaporated,” wrote Johnson on a LinkedIn post this week.
Determined to read all AI papers that are released? Someone who dedicates half a day to reading would have the luxury of an hour per paper in 2014. As of March 2024, they would have to finish each paper in 39 seconds to keep up.
Image credit: iStock/Maxger
Paul Mah
Paul Mah is the editor of DSAITrends, where he report on the latest developments in data science and AI. A former system administrator, programmer, and IT lecturer, he enjoys writing both code and prose.