AI21 Labs Combats Generative AI Hallucinations With New Contextual Answers API

Generative AI developer AI21 Labs has released a new tool for minimizing hallucinations and incorrect and imaginary information composed by a large language model-powered AI text generator. The new Contextual Answers API restricts the text generator to data provided by AI21’s clients when answering queries instead of making up an answer without regard to its accuracy.

Generative Context

ChatGPT, Bard, and other AI text generators often avoid saying they can’t answer a question. They frequently make one up instead, offering a wrong, if occasionally plausible, imaginary answer. The Contextual Answers API improves the specificity and relevance of answers from generative AI chatbots through its reliance on the information uploaded by the organization running the software. That could include internal policy handbooks, technical specifications, fiscal data, and other resources that humans traditionally would have to consult for answers. The AI processes that data and acts like a librarian with perfect memory to answer questions with natural language. What makes AI21’s new tool different is that Contextual Answers only uses that data and won’t try to fill in any gaps with hallucinatory visions of what might be a good answer.

AI21 pointed to a recent KPMG survey finding that while about 70% of enterprises are experimenting with generative AI, 90% are worried about risks like hallucinated content. Contextual Answers serves as a “plug-and-play engine” to resolve that issue, fitting into existing generative AI applications and coming to AI21’s flagship Wordtune consumer-facing app soon. Contextual Answers improves generative AI  chatbots in a few other ways beyond hallucination filtering. Like search engines in recent years, it can deduce from context many misspelled or missing words, parsing the grammar that might be less than pristine.

“There has been a lot of hype – and fear – surrounding generative AI recently, with new exciting tools emerging daily and consumer demand through the roof. The initial fear around AI is starting to wane as the world realizes the benefits. McKinsey recently released a report estimating that AI will boost the world economy by trillions each year,” AI21 Labs CEO Ori Goshen said. “What we expect to happen over the second half of 2023 and into 2024 is that companies will get smarter about how to best implement AI into their respective workflows, and embrace the right tools, which is why we are so excited about the launch of Contextual Answers, further positioning us to serve as a key piece for any company’s AI strategy.”

The new API follows the March release of AI21’s Jurassic-2 large language model, which broke divided the generative AI tools into APIs that enterprise customers could pick from, reducing the overall size of the software package. AI21 has been pushed into the enterprise market since it released Wordtune Read, but the consumer-facing version of Wordtune and its rhetorical spice rack is clearly still a going concern. Wordtune arrived on Google Docs in April, while AI21 updated the iOS version of Wordtune in May by linking the app to Apple’s Siri voice assistant. AI21’s ambitions will likely see many more Wordtune upgrades thanks to a $64 million funding round last year.

Follow @voicebotaiFollow @erichschwartz

AI21 Adds Siri Voice Control and Text Creation to Wordtune Generative AI Writing Assistant App

AI21 Debuts New Generative AI Large Language Model Jurassic-2 and APIs

Got-It AI Debuts Compact Large Language Model With Hallucination Filters ELMAR