Air Canada’s chatbot gave a customer incorrect information about the terms of a refund. In court, the airline argued that the chatbot itself was responsible for what it said, not Air Canada. The court disagreed, and the company had to pay up. Source: The Guardian
While Google’s Gemini offerings are proprietary, the company has now unveiled two open source models under the name Gemma that are free to use. Officials say they are based on the same techniques used for Gemini. With 2 billion and 7 billion parameters, the two Gemma models are small and have limited capabilities. However, they … Read more
If you use Slack in a large company and for a long period of time, it is not so easy to stay up to date and to find information again. New AI helpers are designed to change that, summarizing discussions and improving the search function, for example. Sources: VentureBeat, The Verge
The creators of Stable Diffusion give a preview of Stable Cascade, which, thanks to its particular architecture, can be used on PCs without specialized and expensive hardware. Source: Stability.ai
The nonprofit Cohere for AI has unveiled its large language model Aya, which can speak 101 languages – more than twice as many as other AI tools. Sources: VentureBeat, Cohere
Until now, the AI bot has not been able to learn. Each chat is completely separate from the others. OpenAI is now testing a feature that will allow ChatGPT to remember important details from other conversations it has had. It is currently available to a small number of users. The goal is to gain experience … Read more
Fine tuning refers to the subsequent optimization of a pre-trained AI model for a specific task or vocabulary. For the original training of these models, large amounts of data from the Internet were first processed to gain a general understanding. Fine tuning then adapts this general knowledge to a specific use case. For example, an … Read more
RAG (Retrieval-Augmented Generation) is a technique where AI language models are enhanced with additional, external sources of information. In this process, the AI system’s responses are not only generated from its original training but are also supplemented (augmented) with specifically retrieved current or specialized data. It can be compared to an expert who doesn’t just … Read more