your123 - stock.adobe.com
LexisNexis launches commercial version of AI research tool
Nexis+AI enables users to research and analyze news, summarize documents and create first drafts. The platform is built on GPT from Microsoft Azure and Anthropic Claude from AWS.
LexisNexis on Wednesday commercially launched Nexis+AI, a platform that uses generative AI capabilities to enhance corporate research, intelligence gathering and business decision-making.
The release comes after LexisNexis introduced the beta version of the offering in April.
The commercial version uses more than 20,000 licensed media titles, with rights granted by large publishers including The Associated Press, Gannett and Benzinga.
Nexis+AI can research and analyze business, financial and legal information and news, summarize documents, create first drafts or outlines of intelligence reports, and derive actionable business insights, according to LexisNexis.
In addition to news data, to source Nexis+AI, the information and analytics provider used data from SEC filings and other worldwide registries.
LexisNexis’ approach with Nexis+AI comes after publishers criticized and brought legal action against several generative AI vendors for not getting permission to use their content.
For example, the New York Times is suing OpenAI, and Microsoft for allegedly infringing on its content.
Nexis+AI also comes as other media companies such as Getty Images are creating tools and offerings using generative AI. Getty previously sued Stable Diffusion creator Stability AI for copyright infringement and claimed the vendor infringed by using Getty images to train its image-generating model. Months later, Getty came out with its own image-generating tool.
The way Getty uses the numerous images and data it owns to train its AI tool is similar LexisNexis’ approach with Nexis+AI.
LexisNexis is using its array of different data to create a GenAI tool for the broad audience.
Therefore, the idea behind Nexis+AI – using an array of licensed data to create a generative AI tool for a broad audience is not strictly new, said Michael Bennett, AI Policy Adviser at Northeastern University.
The power of AI at scale
However, the scale at which LexisNexis is operating, with the range of publishers, content creators, and content licenses it has set up, is novel, Bennett said.
“The fact that they'll be able to presumably bring all of that understanding to bear suggests that their product should be pretty impressive,” Bennett said.
With this, LexisNexis provides subscribers access to a broad array of content, removing the threat of intellectual property infringement, he said.
“It seems like you will be getting a lot of the upside of powerful AI that's able to digest and provide insights on large, broad, diverse data sets. At the same time, you will get insurance against some content creator or licensee approaching you down the road and saying, ‘hey, you've infringed upon us,’” Bennett continued.
Solving for hallucinations
For legal experts and researchers, Nexis+AI is a time saver and is more mature than the earlier versions of tools released at the newborn stage of generative AI, said Katie Gardner, an intellectual property lawyer with the firm Gunderson Dettmer.
Those early models were not well vetted and were trained on any data found on the internet.
This has led to trouble for some legal experts. For example, one litigator infamously used one of those early tools to develop case references for a brief and then filed a motion in court. The cases had been made up by a hallucinating LLM.
“My guess is that Lexis has probably worked very hard to solve for that issue,” Gardner said.
For its part, LexisNexis has tried to reduce and eliminate hallucinations with its own guardrails.
Nexis+AI uses two LLMs: Anthropic Claude through AWS Bedrock and GPT models on Microsoft Azure.
The tool also uses retrieval augmented generation on licensed content.
“What that allows us to do is making sure that it is we're not asking the LLM a question and getting the answer from the LLM,” said Snehit Cherian, CTO for Global Nexis Solutions at LexisNexis. “We're basically saying out of our content repository, what are the pieces of text sections within articles that make sense for this question, then we're providing that as the context.”
The tool uses a faithfulness score and examines what part of the answer provided is contextually accurate and what part is inaccurate or concocted.
It also looks at how much context is used to compute an answer to a question.
The tool searches for answer relevance through the two LLMs.
This means that the answer one LLM provides is given to another LLM to see what question the previous LLM is answering. If the question is like the original, then that shows relevance.
Also, there is a human in the loop element. LexisNexis has a team that analyzes answers based on statistical sampling and computes their usefulness, accuracy, authoritativeness and hallucinations.
Finally, each answer provides citations to users so they can validate the answers themselves.
Due diligence and updating
Despite its time-saving potential and the work LexisNexis has done to mitigate hallucinations, users should still exercise due diligence, Gardner said.
“I'd expect that there still is a burden on the end user to make sure it's accurate,” she said. “Whenever you're not referring directly to a primary source, and you're going to rely on software that summarizes that source, I think you have to be aware of the risks involved with that.”
Another area of interest to those who will use this tool is the refresh rate on the material, or how fast the platform can output the most current information, Bennett said.
“With the material coming from so many different sources, I presume there would be different rates of refresh,” he said.
A rapid rate of refresh would be beneficial to organizations and teams that need to make decisions based on insights gained from the service, Bennett said.
Esther Ajao is a TechTarget Editorial news writer and podcast host covering artificial intelligence software and systems.
