-
Compact KI-Update: Video-KI Sora, OpenGPT-X, Anthropic, KI-Ampel
Artists share access to video AI Sora for everyone
Advertisement
A group of early access testers of OpenAI’s Sora AI video model have written an open letter and made the tool publicly available in protest against unpaid development work and exploitation for PR purposes. The artists criticize that they troubleshoot, give feedback and experiment for free, while OpenAI skims off the PR and marketing value. In return, the company simply offers a competition in which selected films are shown.
The group is calling on OpenAI to provide more transparency and support for artists beyond PR campaigns. As an alternative, they point to open source video tools that enable free experimentation. OpenAI rejects the artist group’s allegations. According to the company, participation was voluntary and without obligations. The only condition was to use the tool responsibly and not to reveal confidential details. OpenAI emphasizes that it will continue to support artists with free access, scholarships and events.
European AI language model released by OpenGPT-X
The European research project OpenGPT-X has published a large language model for AI applications. The model, called “Teuken-7B,” was made available for download on the “Hugging Face” platform. To date, almost all relevant AI language models in the Western world come from the USA. OpenGPT-X, on the other hand, is a European research and development project that was launched in early 2022. The aim was to develop a large AI language model that meets the requirements of European values, data protection standards and linguistic diversity.
The project is led by the two Fraunhofer Institutes for Intelligent Analysis and Information Systems (IAIS) and for Integrated Circuits (IIS). The TU Dresden, the Research Center Jülich and companies such as Aleph Alpha and IONOS SE are also involved. “Teuken-7B” was trained from scratch with the 24 official languages of the EU and includes seven billion parameters. It should now be freely available worldwide and thus offer an alternative for science and companies based on public research. Research and corporate actors can also use the open source model in commercial projects and use the code in their own artificial intelligence applications.
New image model Runway Frames brings many fresh looks
The AI company Runway has with Frames a new basic model for image generation presented. Frames features improved control over the visual style of the generated images as well as a high level of detail. The model is intended to enable users to design coherent visual worlds with specific aesthetic characteristics. Frames can establish a consistent look for projects and create image variations that stay true to that style. Runway is currently phasing out access to Frames for select users of the Gen-3 Alpha video model, which launched in June.
Since July, the video model supports entering a start image for video generation, which is often used to influence the result. Runway presents some examples of the different visual styles that can be created with frames. These include, but are not limited to, photorealistic portraits, creatures in the style of handmade special effects from the 80s, record cover designs and retro anime in the style of the 70s. By integrating a high-quality image generator directly into Runway, the start-up is making itself more independent of the current market leader Midjourney. This in turn could face more competition with the publication of its planned video model Runway.
Claude can now supposedly imitate writing styles better
Anthropic’s Claude.ai is now said to be able to imitate writing styles better. In a demo published on This method is probably less reliable than so-called many-shot prompting (i.e. prompting with multiple examples), but is more accessible to users.
Anthropic wants to bring AI chatbots out of isolation
Anthropic presents the “Model Context Protocol” (MCP), a new open source standard that enables AI chatbots to interact with external data sources. The protocol allows data to be read out as well as fed in. A demonstration showed how the chatbot creates an HTML page and saves it directly to a Github repository.
The MCP is based on two main components: MCP servers that provide data and manage access, and MCP clients that access these servers. Anthropic has already deployed servers for popular services such as Slack, Google Maps, Fetch, Google Drive and GitHub. The protocol can be used for both local and Internet-based databases, although initially only local servers are supported. Claude Enterprise customers can already connect their internal systems via MCP servers.
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 look at the different aspects of the AI revolution with experts.
Intel rakes in billions in subsidies for chip factories
The current US government under President Joe Biden is investing billions of US dollars to promote chip production in the USA. Intel reports that it has now been able to conclude agreements totaling $7.86 billion from the CHIPS Act. The money will be invested in chip fabs under construction in Arizona, New Mexico, Ohio and Oregon. In addition, Intel receives tax breaks depending on the actual investment amount. Separately, Intel was able to secure up to $3 billion from the CHIPS Act in September for the US Department of Defense’s secret “Secure Enclave Program”. Intel is currently in difficult waters, making high losses, laying off thousands of people and has already written off parts of its existing chip production facilities. Intel is currently concentrating its investments primarily on the USA. Intel has already put the construction of the planned fabs in Magdeburg and the packaging plant in Breslau on hold. A project in Israel is also on hold.
Facebook Messenger finally learns Siri
Users of Facebook Messenger on the iPhone can look forward to a number of new functions that will be updated to the devices in the coming weeks. For the first time, the application will directly support Apple’s voice assistant Siri. Meta apparently uses Apple’s SiriKit for this. It should be possible to directly ask the voice assistant to start calls via Messenger or send messages via Messenger. The only important thing is that the messenger app is mentioned as a means of communication, as is also the case with other third-party applications integrated into Siri. This also improves support for people with disabilities, as they can communicate completely without having to use their iPhone. Meta also took inspiration from FaceTime and added the option of leaving audio and film messages to other people.
This makes Facebook Messenger “more of a fully-featured telephone replacement,” writes Meta. If a person doesn’t answer, you can record a message at the push of a button, which will then be displayed to the other person. In later versions coming in the coming weeks, Facebook Messenger will also support personalized AI backgrounds in video calls. These can be accessed via the effects icon and generated using Meta-AI. In addition to Facebook Messenger, Meta also operates WhatsApp. Both applications already contain the company’s AI features.
AI traffic light in Hamm should work without AI during rush hour
In the spring, the city of Hamm was largely satisfied with its first traffic light controlled by artificial intelligence, which has been regulating traffic at the Heßlerstrasse/Marker Allee intersection since July 2023. Several cameras are intended to detect cyclists and pedestrians at an early stage. The system then uses AI to calculate when they arrive at the traffic light to give them priority green. But for a few weeks now, the automated circuit has been causing long queues of cars in rush hour traffic and complaints from drivers, local media report. The AI should therefore initially take a break from 3:30 p.m. to 5:30 p.m. During this period, according to the administration’s new plan, the traffic lights will be switched conventionally, the reports say. The aim is to allow car traffic to flow more quickly during rush hour. In the morning the problem is not so big, the Westfälischer Anzeiger refers to a meeting of the city’s district council. One assumption is that the traffic light traffic jams also have to do with several inner-city construction sites nearby and evasive traffic. If these are lifted, the AI should run again in the afternoon. But before that, new tests were carried out.
That was the AI update from heise online from November 27, 2024. A new episode is available every working day from 3 p.m.
(igr)
