Microsoft presents Copilot Pro, making AI easier for smaller businesses
Advertisement
Microsoft is expanding its Copilot offering, making it available to more users and across different contracts, including a new subscription model Copilot Pro. Microsoft’s AI assistant can now also create Copilot GPTs, special chatbots that can be offered in the OpenAI GPT store. Copilot for Microsoft 365 is now available to everyone, regardless of company size. Previously, companies had to purchase at least 300 licenses for $30 per month, which equates to $108,000 over a year.
The Microsoft 365 extension is now also available for $30, while private individuals or users with a family account can book the Copilot Pro for $20. The AI assistant accesses GPT-4 and DALL-E 3, the most powerful models, with the Copilot Pro even including an upgrade to GPT-4-Turbo. The reason for the lack of this upgrade in the enterprise versions is unclear.
Copilot is intended to help users write emails, evaluate Excel tables and create PowerPoint presentations from Word documents. The actual functionality of these features still needs to be tested in practice.
OpenAI implements security measures against election interference using AI tools
OpenAI announces new security measures and revised usage guidelines for its AI tools ChatGPT and DALL-E 3, to prevent their misuse in the 2024 US elections. The measures aim to stop misleading deepfakes, large-scale influence operations and deceptive chatbots.
From the beginning of 2024, images from DALL-E 3 will be provided with invisible watermarks of the C2PA standard. In addition, a tool is to be developed that can identify DALL-E images even after modifications, with an initial test expected shortly.
ChatGPT is intended to be more closely linked to real-time news reporting, including sources and links. The revised usage guidelines prohibit the use of ChatGPT and the API for political campaigning, lobbying, and impersonating chatbots as real people. A report button for chatbots will be introduced for the new GPTs.
The measures are intended to help protect the integrity of democratic processes and prevent the misuse of AI technologies in a political context. ChatGPT’s potential impact on future elections was previously discussed at a US Senate hearing with OpenAI CEO Sam Altman in May 2023.
IMF warns about artificial intelligence and job losses
The International Monetary Fund (IMF) warns in a study that the use of AI could affect up to 40 percent of global jobs. In advanced economies, as many as 60 percent of jobs could be affected. Because their employment structures are more focused on cognitively intensive tasks, advanced economies and more developed emerging markets will feel the benefits and risks of AI sooner than less developed countries.
The IMF recommends that these countries invest in AI innovation and integration and improve regulatory frameworks to maximize the benefits of AI. For less prepared emerging and developing countries, the development of basic infrastructure and the creation of a digitally qualified workforce is of utmost importance.
The report emphasizes that AI has the potential to increase productivity and growth, but its impact on the economy and society remains uncertain. IMF Managing Director Kristalina Georgieva warns that AI will worsen general inequality and calls on politicians to prevent further social tensions, for example through social safety nets and retraining programs.
Nvidia fixes critical security vulnerabilities in AI accelerators
Nvidia has released firmware updates for its DGX A100 and H100 systems to address critical security vulnerabilities. The affected systems, especially the H100 GPUs of the Hopper generation, are currently used by all major providers for AI calculations, for example in Microsoft’s Azure cloud services. In a security announcement, Nvidia lists eleven vulnerabilities that the update addresses: three are considered critical, four are considered high risk and four are considered medium threat.
Attackers could trigger a stack-based buffer overflow by sending crafted network packets without prior authentication. This makes it possible to inject malicious code, provoke a denial of service, gain unauthorized access to information or manipulate data. The critical gaps affect the KVM daemon of the DGX-A100 baseboard management controller (BMC) and partly also the DGX-H100 BMC.
Administrators should download and apply the provided updates as soon as possible as the vulnerabilities can be exploited directly from the network. The updates are available from Nvidia’s Enterprise Support portal.
How intelligent is artificial intelligence actually? What consequences does generative AI have for our work, our leisure time and society? In Heise’s “AI Update” we, together with The Decoder, bring you updates on the most important AI developments every weekday. On Fridays we examine the different aspects of the AI revolution with experts.
(olb)