This Week's Most Read: Can Quantum Help AI Understand?

Here are this week's top trending stories:

1. AI Meets Quantum: Leap Forward for AI Models to ‘Understand’

AI systems that can understand the world instead of merely predicting the next word or code is the dream of many AI researchers.

A group of quantum computing scientists developed a new approach that brings the dream a step closer: They developed a framework that enables machines to learn the way humans do.

A new paper from the team at Quantinuum describes the framework, which lets AI systems learn concepts like shape and color. Not only can the machine look at an image and recognize it, but it also actually understands the meaning of the object.

They developed the Compositional Quantum Framework, which is designed to structure and learn concepts automatically from data through both classical and quantum computing approaches.

In simple terms, the researchers essentially merged insights from quantum computing with cognitive science concepts to create a framework that provides a mathematical structure to allow an AI system to visualize an action.

Quantinuum applied the concept to image recognition, demonstrating that concepts like shape, color, size and position can be taught to machines that are trained on images of shapes.

Read more

2. Air Canada Held Responsible for Chatbot’s Hallucinations

A Canadian small claims tribunal has ordered Air Canada to refund a traveler’s airfare as a result of its chatbot’s hallucinations.

Last November, Jake Moffatt purchased tickets from Vancouver to Toronto after his grandmother’s death. He researched bereavement policies on Air Canada’s website and was told by its chatbot that he can apply to get the discounted bereavement airfare within 90 days.

After his return, Moffatt applied to get the discount only to be told that he cannot ask retroactively for the discounted fare. Air Canada instead offered him a $200 coupon for his next flight.

Moffatt filed a case with the Civil Resolution Tribunal to get his money back.

The airline's defense was to say “it cannot be held liable for the information provided by the chatbot,” according to the tribunal’s ruling, which came down on Feb. 14. Moreover, Air Canada said the chatbot provided a link to the refund page of the website that correctly states the bereavement fare policy.

Air Canada was told to pay Moffatt $812.02, comprising his refund, interest and tribunal fees.

Read more

3. Hugging Face Debuts Open Source AI Assistant Builder

Popular code repository Hugging Face has taken the wraps off its own AI assistant builder – and it is free and open source.

The service, which is found in Hugging Face Chat, lets developers create their own custom AI assistants using open source language models. It is similar to OpenAI’s GPT builder that is accessed through ChatGPT Plus, which is available only to paying customers.

The Hugging Face assistant can use “any available open LLM, like Llama2 or Mixtral,” tweeted Philipp Schmid, technical lead at Hugging Face.

However, he warned that users should not input confidential information into the AI assistants as they are public. He also said that function calling – which OpenAI introduced last July for GPT − is not available at present.

Users can make their own private AI assistant and host it themselves. The Hugging Face Chat UI is available here. All assistants are public by default with the author's name.

Read more

4. Google Unveils Open Source Models to Rival Meta, Mistral

Google released a new family of open source language models today, in a salvo aimed at Meta, Mistral and other open source rivals.

The search giant took the wraps off Gemma, its line of open source language models that is based on Gemini, Google’s most powerful large multimodal model to date. However, Gemma is text and code only.

Google released two sizes of what it calls “state of the art” Gemma models: the 2-billion parameter model and 7-billion parameter model.

The most popular rival open source models in these weights are those from Meta’s Llama 2 and Mistral.

But Google said Gemma outperforms or holds its own against Llama 2 (both 7 billion and 13 billion parameters) and Mistral (7 billion parameters) in dialogue, reasoning, math and coding.

Read more

5. CISOs’ Most Common Concerns with Generative AI

CrowdStrike Chief Scientist Sven Krasser penned an opinion piece on the benefits and pitfalls of generative AI.

Krasser explores the top security questions today’s leaders are asking before allowing generative AI in their environments, either as a tool for staff or as a product component.

Read more

For the full roster of AI news, go to AI Business or sign up for our email newsletter.