OpenAI introduced it has expanded entry to its Deep Analysis, its new AI agent designed to conduct complicated analysis. Deep Analysis debuted earlier this month to customers of ChatGPT Professional, the corporate’s $200/month subscription tier, however is now accessible to all paid customers.
“Deep Analysis is now rolling out to all ChatGPT Plus, Group, Edu, and Enterprise customers,” the corporate mentioned in a social media publish Tuesday.
Beforehand, Professional customers have been allowed as much as 100 queries per 30 days. Now, Professional customers are allotted 120 queries per 30 days, whereas Plus, Group, Edu, and Enterprise customers get 10 queries per 30 days.
OpenAI additionally introduced enhancements to Deep Analysis since its preliminary launch, together with the addition of citations for embedded photographs and enhanced capabilities for understanding and referencing uploaded information.
“Give it a immediate and ChatGPT will discover, analyze and synthesize tons of of on-line sources to create a complete report in tens of minutes vs what would take a human many hours,” the corporate mentioned of the brand new Deep Analysis functionality, accessible by the ChatGPT interface.
Deep Analysis is an AI agent designed for in-depth web analysis, able to conducting multi-step inquiries, analyzing textual content, photographs, and PDFs, and adapting its method primarily based on new info. Powered by an early model of OpenAI’s o3 mannequin, it additionally helps user-provided information and may execute Python code for knowledge evaluation.
(Supply: OpenAI)
OpenAI says Deep analysis is constructed for individuals who do intensive information work in areas like finance, science, coverage, and engineering and wish thorough and dependable analysis. It's additionally helpful for discerning customers in search of hyper-personalized suggestions on purchases that require cautious analysis, the corporate says.
In a tweet, OpenAI CEO Sam Altman mentioned, “Deep Analysis out for ChatGPT Plus customers!” and referred to as the function “One in all my favourite issues we now have ever shipped.”
OpenAI CPO Kevin Weil additionally mentioned it was his favourite product the corporate has launched and described how he has used it: “It may possibly do weeklong research-oriented duties in 15 minutes. I've used it to raised perceive muon colliders, the renewable vitality market, and AI publish coaching strategies—and to analysis/buy a basketball hoop for my children and new footwear for myself.”
Deep Analysis did nicely on a benchmark check referred to as “Humanity’s Final Examination,” made by researchers on the Middle for AI Security together with knowledge annotation agency Scale AI. The check is made up of roughly 3,000 multiple-choice and quick reply questions submitted and reviewed by consultants in tutorial fields like arithmetic and philosophy.
Deep Analysis scored 26.6% on the check, the present main rating, which can not sound spectacular at first look. However the earlier leaders, OpenAI’s o1 and DeepSeek’s R1 fashions, each solely scored 9%. This represents a big leap in intelligence for Deep Analysis, with the biggest positive aspects occurring on questions associated to chemistry, humanities and social sciences, and arithmetic, in keeping with OpenAI.
Nonetheless, all massive language fashions are vulnerable to hallucinations, or utterly fabricating details and occasions that may look and sound credible. These utilizing generative AI for analysis should fastidiously test mannequin outputs for accuracy, regardless of how nicely the fashions do on benchmarks.
That may be a actuality not too long ago confronted by Florida-based damage legislation agency Morgan and Morgan, who was fined by a Wyoming choose for submitting a case submitting with eight fictitious courtroom circumstances generated by a chatbot.
“This deeply regrettable submitting serves as a tough lesson for me and our agency as we enter a world through which synthetic intelligence turns into extra intertwined with on a regular basis observe,” the case’s lead legal professional Mike Morgan wrote in a declaratory submitting, including that “there aren’t any shortcuts in legislation.”
