OpenAI upgrades Operator with o3 model for enhanced reasoning, safety
OpenAI is updating the artificial intelligence (AI) model powering Operator, its AI agent that can autonomously browse the web and interact with certain software inside a cloud-hosted virtual machine to carry out user requests.
Operator will soon run on a model based on o3, one of the latest in OpenAI's o series of 'reasoning' models. Previously, Operator relied on a customised version of GPT-4o.
By several benchmarks, o3 is a more advanced model, particularly on tasks requiring mathematical ability and reasoning.
'We are replacing the existing GPT‑4o-based model for Operator with a version based on OpenAI o3,' OpenAI wrote in a blog post. 'The API version (of Operator) will remain based on 4o.' Operator is part of a growing set of agentic tools developed by AI firms as they compete to build agents capable of reliably performing digital tasks with minimal supervision.
Google offers a similar agent through its Gemini API, which can browse the web and take actions on users' behalf. It also offers a consumer-facing version called Mariner. Anthropic's models can perform various computer tasks as well, including opening files and navigating webpages.
According to OpenAI, the upgraded Operator model, dubbed o3 Operator, was 'fine-tuned with additional safety data for computer use,' using datasets designed to 'teach the model (OpenAI's) decision boundaries on confirmations and refusals.'The company has released a technical report detailing o3 Operator's performance in safety evaluations. Compared to the GPT-4o version, the new model is less likely to carry out illicit activities, search for sensitive personal data or fall prey to prompt injection, a common AI attack technique.'o3 Operator uses the same multi-layered approach to safety that we used for the 4o version of Operator,' OpenAI wrote in its blog post. 'Although o3 Operator inherits o3's coding capabilities, it does not have native access to a coding environment or terminal.'
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Time of India
34 minutes ago
- Time of India
Tech giant SAP asks US Supreme Court to reconsider rival's antitrust win
Europe's largest software maker SAP ( has asked the U.S. Supreme Court to review a decision that said the technology giant must face a lawsuit by U.S. data technology company Teradata accusing it of violating antitrust law . SAP in a petition made public on Tuesday said a decision by the 9th U.S. Circuit Court of Appeals in California that reinstated Teradata's lawsuit will threaten American tech innovation if it is left in place. Teradata accused SAP of violating antitrust law by "tying" sales of business-planning applications with the purchase of a key SAP database that can perform transactional and analytical functions. Teradata makes a rival analytics database. by Taboola by Taboola Sponsored Links Sponsored Links Promoted Links Promoted Links You May Like Alquilar coches de lujo nunca fue tan fácil Coches | Anuncios Publicitarios Más información Undo In its filing at the high court, SAP said the integration of software products can often benefit consumers and "represent an effort to 'compete effectively,' rather than to stifle competition." SAP declined to comment. Teradata did not immediately respond to a request for comment. Live Events San Diego-based Teradata filed its lawsuit against SAP in federal court in California in 2018. The two companies once had a joint venture, but SAP terminated it after developing its own analytics database. Discover the stories of your interest Blockchain 5 Stories Cyber-safety 7 Stories Fintech 9 Stories E-comm 9 Stories ML 8 Stories Edtech 6 Stories SAP won in the district court, but the 9th Circuit revived Teradata's case in December. The appeals court said there was material dispute between the companies that a jury could decide. If the Supreme Court takes the case, the justices could rule on which legal standard judges should use to weigh antitrust tying claims. Two key legal standards guide how judges resolve whether conduct restrains competition: the "per se rule," where alleged conduct is presumed illegal, and the "rule of reason," where judges balance between anticompetitive effects and a defendant's procompetitive justification. The 9th Circuit, using a version of the "per se rule," applied too stringent a standard in evaluating Teradata's claims, SAP told the justices. SAP said the appellate court's ruling clashed with how a Washington federal appeals court resolved a landmark antitrust case against Microsoft in the 1990s.
Mint
40 minutes ago
- Mint
Google Search now shows AI-generated weather snapshots for select users: Report
Google is reportedly testing a new artificial intelligence (AI) feature that provides users with brief, text-based summaries of current weather conditions directly within its mobile Search experience. The feature, which is currently limited in availability, has been spotted by some users in the United States when searching for city-specific weather information through the Google app on Android and iOS. According to a report by9to5Google, the AI-powered summary appears between the standard hourly and 10-day forecast sections and aims to offer an at-a-glance understanding of the weather. The brief narrative includes key details such as wind conditions, the likelihood of rain or thunderstorms, atmospheric pressure patterns, and trends in temperature fluctuations. At present, the feature seems to be in an early testing phase, with visibility restricted to certain regions. Searches for 'weather Los Angeles' and 'weather San Diego' reportedly triggered the new summary, whereas other locations did not yield the same result. Mint was also unable to access the summaries for the mentioned cities, indicating that the rollout is likely limited to a small user group. In addition to the text description, the AI-generated section includes a small icon linking to external sources for more comprehensive weather information. Beneath the summary, Google has added a disclaimer that reads: 'Generative AI is experimental,' reinforcing that the tool is still in a testing stage. Currently, the new feature appears only within the Google app and on the mobile web. It does not show up on desktop platforms or when users are not signed into their Google accounts. Google has not yet announced any plans for a wider rollout or confirmed when the feature might become broadly available. The addition comes as part of the tech giant's ongoing efforts to integrate generative AI into its suite of services, offering more contextual and user-friendly information across its platforms.
&w=3840&q=100)
Business Standard
an hour ago
- Business Standard
Hackers posed as IT staff to steal Salesforce data from firms: Google
A hacking group has been impersonating IT personnel to break into companies' Salesforce tools, using the access for data theft and extortion, according to a new report from Google's threat intelligence group. The hackers, which have links to a loosely affiliated group of hackers largely based in the US, UK and Western Europe called the Com, successfully breached the networks of at least 20 companies in the US and Europe, Google said. They operate by calling up employees and pretending to be IT support personnel, convincing them to provide sensitive credentials and using that to steal Salesforce data, Google said in the report published Wednesday. In some cases, the hacker was able to fool an employee into connecting a malicious app to their organisation's Salesforce portal, allowing the hacker to steal Salesforce data. Some victims didn't receive an extortion demand in exchange for the deletion of the data until months after it was stolen, according to the report. The hackers relied on manipulating its victims, not any vulnerability in Salesforce tools, Google said. 'There's no indication the issue described stems from any vulnerability inherent to our services,' a Salesforce spokesperson said in an email. 'Attacks like voice phishing are targeted social engineering scams designed to exploit gaps in individual users' cybersecurity awareness and best practices.' In a March blog post, the company noted that threat actors had been using social engineering techniques to break into its customers' Salesforce accounts, and it provided guidance to protect against such attacks. Google's report comes as a string of retailers have been hacked in recent months. Marks & Spencer Group Plc is facing a £300 million ($406 million) hit to operating profit this year due to a ransomware attack in April. Fellow British grocer Co-op Group disclosed shortly afterward that it too was the victim of a cyberattack. Adidas AG and Victoria's Secret & Co., Cartier and North Face have also disclosed cybersecurity incidents in recent weeks. Google's report didn't identify specific victims. 'While we've seen this group target retail, they have also targeted other industries and we do not have enough information to definitively link this group to the recent hacks in the US and UK more broadly,' said Austin Larsen, principal threat analyst at Google Threat Analyst Group. The hacking group used infrastructure and methods previously used by members of the Com, Google said. Members of the hacking group Scattered Spider, which was accused of a raft of high-profile attacks in recent years, many of which involved impersonating IT staff, have also been linked to the Com, made up mostly of young male SIM-swappers who organized on social media channels to steal cryptocurrency by taking control of victims' phone numbers. Google urged companies to remain vigilant against so-called social engineering attacks. (Only the headline and picture of this report may have been reworked by the Business Standard staff; the rest of the content is auto-generated from a syndicated feed.)
