Shashua's A121 Labs launches NLP model to rival GPT-3

A121 Labs founders Photo: Eyal Izhar
A121 Labs founders Photo: Eyal Izhar

The Israeli startup claims that its natural language processing system is more comprehensive and sophisticated than OpenAI's model.

Just over a year after San Francisco-based OpenAI launched GPT-3, the world's best computerized natural language processing system (NLP), which could independently write its own sentences complete with good syntax, an Israeli company A121 Labs has released its first NLP model Jurassic-1. A121 Labs claims that Jurassic-1 is the world's largest and most sophisticated language model, to anyone interested in prototyping custom text-based AI applications and is three times more powerful than GPT-3. Jurassic-1 makes language decisions based on 178 billion computing parameters, slightly higher than GPT-3.

The Tel Aviv based startup was founded in 2017 by co-CEO Prof. Yoav Shoham, co-CEO Ori Goshen, a serial entrepreneur, and Amnon Shashua, cofounder and CEO of Mobileye (acquired by Intel). The company has raised $35 million from investors including Pitango, TPY Capital, Shlomo Kramer and Shashua himself. A121 Labs has 60 employees in its Tel Aviv offices in Millennium Tower. Shoham said, "First of all GPT-3 is not a bad competitor and we are proud to have equaled it and in certain areas we are even better and larger than it. This is the first time that there is another strong player in the world of NLP, as well as GPT-3. There are specific models to the language that are very strong but as a general model for language I think that we are the most comprehensive in the world."

Jurassic-1 is offered via AI21 Studio the company’s new NLP-as-a-Service developer platform, a website and API where developers can build text-based applications like virtual assistants, chatbots, text simplification, content moderation, creative writing, and many new products and services. Jurassic-1 is the first in a line of language models that the company is working on, to be followed by newer and even more sophisticated versions.

"Globes" was given a brief look at Jurassic-1 before the official launch and the model is impressive. Jurassic-1 learns from examples and after two or three attempts it 'understands' the matter and can translate Hebrew slang into English. It can create strong headlines about items and automatically generate the right hashtag for a post.

The legendary linguist Noam Chomsky said about GPT-3, "Deep Learning is useful but it doesn't tell you anything about human language and cognition in general."

Shoham said, "I don't agree with Chomsky about practically everything but on this he raises an important point. This model really relies on statistics on an inconceivable scale and identifying correlations. When you deal with it a lot you see that it perhaps does begin to understand something about the significance of some of the things."

Shoham added, "Hype doesn't interest us and we aren't looking to massage our egos. Not long ago we did a hackathon at the company and the employees produced amazing things with Jurassic-1 like writing rap songs or an application that reacts to chess games and sends trash talk to the opponent, but from our point of view the genuine index is that the model can create smart and breakthrough applications that will bring value to the company."

Published by Globes, Israel business news - - on August 11, 2021

© Copyright of Globes Publisher Itonut (1983) Ltd. 2021

A121 Labs founders Photo: Eyal Izhar
A121 Labs founders Photo: Eyal Izhar
Twitter Facebook Linkedin RSS Newsletters גלובס Israel Business Conference 2018