Large language models are outdated for another 5 years

Ian Lacan, a senior scientist of meta artificial intelligence and one of the artificial intelligence godfathers, has made an interesting comment on large linguistic models (LLM), saying these models will be outdated for another five years.

Lacan is one of those who has played a significant role in the development of artificial intelligence and even won the Turing award, but when asked about the future of large language models, such as ChatGPT, Jina and anthropic models, they said these models would be outdated in the next five years.

Large language models have no place in the future of artificial intelligence

Lacan and his colleagues work on a new generation of artificial intelligence models called JOINT EMBEDING Predictive Architecture (JEPA). “My colleagues and I have taken a path to the Facebook and New York University Laboratory that if we can do it within 2 to 5 years, we will have a much better model for systems that can reason and plan,” he said in a recent interview with Newsweek.

Lacan hopes this new pattern will be able to obsolete large -scale language models in artificial intelligence; Because these new systems will include better display of the world. Lacan said these systems are “controllable; That is, you can give them goals, and according to their structure, they can only achieve those goals. “

The senior researcher believes in its new systems so much that it was advised developers not to work on the development of large language models last year at a conference. “Don’t work on LLMs; These are in the hands of the big companies and you can’t put anything on the table. “You have to work on the next -generation artificial intelligence systems that will eliminate LLM limitations.”

Yan Lacan’s claims about the obsolete of large language models for the next five years have been his previous talk, which said that a major revolution in the field of artificial intelligence will take place in the next five years. Lacan said the big drawback of current models is that they cannot analyze and predict real -world behaviors well.