Today’s language models are primarily trained to provide helpful and easy-to-understand responses. At the same time, it is possible for the AI to make up information that fits the text perfectly and looks factual, but is actually made up. Such errors are often called hallucinations. These can be avoided, for example, by using an optimized …
In his 1927 film “Metropolis,” Fritz Lang showed an artificial intelligence that frightened people. In the film, which describes a future with clear class distinctions, a robot known as a “Maschinenmensch” causes unrest. The robot, which is initially used as a worker, later takes the form of a young woman named Maria and provokes an …
SambaNova has set a new speed record for the Llama 3 language model, reaching 1,000 tokens per second. The company uses its own chips and software designed specifically for enterprise needs. The high speed enables faster response times, better hardware utilization and lower costs.
PwC, one of the largest accounting and consulting firms, has selected ChatGPT Enterprise to become OpenAI’s largest enterprise customer. PwC is deploying the AI software to more than 100,000 employees in the US and UK.
OpenAI continues to expand its partnerships with media companies: Following deals with News Corp. and other publishers, The Atlantic and Vox Media have signed agreements with the ChatGPT maker. OpenAI will gain access to the publishers’ content and data to train its AI models in exchange for financial compensation and access to its technology.
OpenAI is working on a new, powerful AI model that will outperform GPT-4. The new model is expected to be released in the next nine to twelve months.
Anthropic has published a new research paper that sheds light on exactly how large language models work. They did this by specifically activating certain neurons in the model, for example, for the concept of the Golden Gate Bridge. As a result, this modified version of Claude continuously weaved the Golden Gate Bridge into his responses, …
IBM unveils new generative AI technologies at its Think 2024 conference. It makes its powerful “Granite” models freely available. IBM wants to encourage the development and adoption of AI in the enterprise.
The recently introduced “AI Overviews” in Google Search have produced some strange results – some embarrassing, some ridiculous, some dangerous. This can be seen as an example for what various experts already know and preach: Don’t let your AI work unsupervised. For example, one of Google AI’s recommendations was that cheese sticks better to pizza …
OpenAI has received a lot of criticism in the last two weeks. The company apparently had disproportionately strict rules for employees leaving the company. For example, they are not allowed to criticize the company or else lose their valuable shares. CEO Sam Altman claims to have been unaware of this and has apologized. However, internal …