Topics
Latest
AI
Amazon
Image Credits:Jakub Porzycki/NurPhoto / Getty Images
Apps
Biotech & Health
Climate
Image Credits:Jakub Porzycki/NurPhoto / Getty Images
Cloud Computing
Commerce
Crypto
Image Credits:OpenAI
endeavor
EVs
Fintech
Image Credits:OpenAI
fund raise
Gadgets
bet on
Government & Policy
ironware
Layoffs
Media & Entertainment
Meta
Microsoft
concealment
Robotics
Security
Social
place
Startups
TikTok
Transportation
speculation
More from TechCrunch
result
Startup Battlefield
StrictlyVC
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
get through Us
OpenAI is announcing a newAI “ agent”designed to facilitate people conduct in - depth , complex research usingChatGPT , the company ’s AI - power chatbot platform .
Appropriately enough , it ’s called deep inquiry .
OpenAI aver ina blog postpublished Sunday that this Modern capableness was design for “ hoi polloi who do intensive knowledge body of work in areas like finance , science , policy , and engineering and need thoroughgoing , precise , and reliable research . ” It could also be useful , the troupe contribute , for anyone making “ purchase that typically call for careful enquiry , like cars , appliances , and piece of furniture . ”
Basically , ChatGPT cryptic research is intended for instances where you do n’t just want a quick result or summary , but or else demand to assiduously consider data from multiple websites and other rootage .
OpenAI said it ’s making deep enquiry uncommitted to ChatGPT Pro users today , confine to 100 queries per calendar month , with musical accompaniment for Plus and Team users coming next , stick with by Enterprise . ( OpenAI is targeting a Plus rollout in about a month from now , the troupe said , and the inquiry limits for paid user should be “ significantly gamey ” soon . ) It ’s a geo - direct launch ; OpenAI had no release timeline to portion out for ChatGPT customers in the U.K. , Switzerland , and the European Economic Area .
To use ChatGPT deep research , you ’ll just take “ bass research ” in the composer and then enter a query , with the option to confiscate Indian file or spreadsheets . ( It ’s a web - only experience for now , with fluid and desktop app desegregation to amount later this calendar month . ) inscrutable research could then take anywhere from five to 30 minutes to answer the question , and you ’ll get a notification when the search fill out .
Currently , ChatGPT deep enquiry ’s outputs are text - only . But OpenAI said that it intend to add embedded picture , data visualisation , and other “ analytic ” output soon . Also on the roadmap is the ability to associate “ more specialised data source , ” including “ subscription - based ” and intimate resources , OpenAI add together .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
The with child dubiousness is , just how accurate is ChatGPT cryptic inquiry ? AI is frail , after all . It ’s prostrate tohallucinations and other types of errorsthat could be particularly harmful in a “ mysterious inquiry ” scenario . That ’s perhaps why OpenAI said every ChatGPT recondite research output will be “ fully documented , with clear quotation and a summary of [ the ] thought process , making it easy to cite and affirm the info . ”
The jury ’s out on whether those extenuation will be sufficient to battle AI mistakes . OpenAI ’s AI - powered vane hunting feature film in ChatGPT , ChatGPT Search , not infrequentlymakes gaffes and give amiss result to interrogative . TechCrunch ’s testing found that ChatGPT Searchproduced less utile resultsthan Google Search for certain queries .
To gripe up thick inquiry ’s accuracy , OpenAI is usinga special interpretation of its late announced o3 “ reason ” AI modelthat was school through reinforcement learning on “ real - world tasks requiring web browser and Python pecker use . ” Reinforcement con fundamentally “ teaches ” a model via trial and mistake to reach a specific end . As the model flummox closer to the goal , it experience practical “ rewards ” that , ideally , make it better at the task break down forward .
OpenAI said this version of o3 is “ optimized for web graze and data analysis , ” total that “ it leverage reasoning to lookup , render , and analyze massive sum of textbook , effigy , and PDFs on the net , pivoting as needed in reaction to entropy it encounters . ” The model “ is also capable to browse over user - uploaded Indian file , ” the troupe say , and “ plot of land and iterate on graphical record using [ a Python ] pecker , embed both beget graph and image from web site in its responses , and cite specific judgment of conviction or passages from its sources . ”
OpenAI articulate it tested ChatGPT deep research usingHumanity ’s Last Exam , an evaluation that includes more than 3,000 expert - level questions in a smorgasbord of academic fields . The o3 model powering deep research accomplish an accuracy of 26.6 % , which might look like a failing ground level — but Humanity ’s Last Exam was plan to be tougher than other benchmark to outride ahead of model advancements . accord to OpenAI , the deep research o3 model add up in way in advance ofGemini Thinking(6.2%),Grok-2(3.8 % ) , and OpenAI ’s ownGPT-4o(3.3 % ) .
Still , OpenAI notes that ChatGPT deep inquiry has restriction , sometimes cause mistake and incorrect inferences . cryptic research may struggle to describe authoritative entropy from rumors , the troupe said , and often fails to convey when it ’s uncertain about something — and it can also make formatting errors in reports and citation .
For anyone distressed about the impact of productive AI on student , or on anyone examine to find entropy online , this type of in - profoundness , well - quote output in all probability sound more appealing than a misleadingly simple chatbot sum-up with no acknowledgment . But we ’ll see whether most drug user will in reality subjugate the output to real analysis and two-fold - checking , or if they simply do by it as a more professional - looking text to copy - spread .
And if this all sounds familiar , Google actuallyannounced a interchangeable AI featurewith the exact same name less than two months ago .
