AI21 Labs Ltd., a startup providing cloud-based broad language models for developers, announced that it has raised $53 million in new funding. The capital was provided as an extension of the Series C investment the company first announced in August. The new raise was led by the venture capital arms of Intel Corp. and Comcast Corp. at a valuation of $1.4 billion. Their investment brings AI21's total external financing to $336 million.
Tel Aviv-based AI21 was founded in 2017 by Co-Chief Executive Officers Ori Goshen and Yoav Shoham. The company's third co-founder is Amnon Shashua, CEO of vehicle chip maker Mobileye Global Inc. In 2020, AI21 launched its first product, an artificial intelligence writing assistant called Wordtune. After that, it expanded its focus to offer large language models.
Also See: CoinGecko Acquires Zash
AI21 Labs
The company provides language models through a cloud-based platform called AI21 Studio. The platform also makes the models accessible through application programming interfaces that allow developers to integrate them into their software. For customers, its APIs power thousands of manufacturing applications, according to AI21. AI21 Studio provides access to three general-purpose large language models, collectively known as the Jurassic-2 series. They can create text, rewrite existing text, summarise documents and perform other tasks. AI21 claims that Jurassic-2 is among the top performers in Stanford University's HELM evaluation, which assesses the reliability of language models.
Developers have the option to customise Jurassic-2 models using custom training data to further improve their accuracy. According to the company, specifying for a training dataset with as few as 50 to 100 learning examples. It can provide greater accuracy improvements than instantaneous engineering. In machine learning, instant engineering is the practice of also manually refining the inputs that users input into a neural network to improve the output. In addition, AI21 Studio provides access to a range of more customised language models. Each model in the series focuses on a single task, such as generating text or summarising documents. There is also a neural network model specifically optimised for powering chatbots.
No comments yet for this news, be the first one!...