Beyond the Code: Deepmind's AI Comedian, LLM Tumor Detection, AI in Regulatory Compliance

Welcome back, readers! If you're new, this newsletter curates the top 4 AI innovations each week.

From cutting-edge research labs to promising startups, disruptive products to techniques like RAG and prompt engineering ― we cover the essential AI developments you need to know.

Let's dive into this week's lineup.

DeepMind's Challenge: Can AI Master the Art of Comedy?

DeepMind researchers explored whether LLMs could assist stand-up comedians in writing routines, publishing their findings on the arXiv preprint server.

A small team involved 20 professional comedians, who previously worked with LLMs, to write full routines using AI assistance.

• While LLMs demonstrated proficiency in generating joke content, the resulting jokes lacked humor, often being too generic and predictable.
• Comedians rated the AI-written material using a Likert scale, mainly finding the jokes to lack the necessary sharpness and originality for effective comedy.
• Despite the shortcomings in humor, some comedians found the LLMs helpful in establishing a basic framework for their routines, which could be enhanced with personal creativity.

The study noted the inherent cautiousness in LLM design, with filters to avoid offensive content, as a potential reason for the blandness of the jokes.

Full article here.

Mahmood Lab's PathChat: AI-Driven Diagnostic Breakthrough

PathChat is a new pathology-specific large language model developed by the Mahmood Lab that accurately identifies tumors and other conditions from medical images.

• It outperforms leading LLMs like ChatGPT-4V and LLaVA models in both image-only and clinically contextual assessments.
• The model uses a vision encoder and visual language instructions, and is trained across multiple pathology practices and diagnoses.
• PathChat is designed to support pathologists with differential diagnosis and tumor grading, enhancing diagnostic processes.

There is potential for broader medical applications, emphasizing continual learning and integration with digital health tools.

Full article here.

Sierra Technologies' ??-bench: Redefining AI Customer Service

Sierra Technologies Inc. has launched ??-bench, a benchmark that evaluates AI agents' ability to handle complex customer service tasks in realistic settings.

• The startup, founded by ex-executives from Salesforce and Google, has developed AI chatbots with enhanced contextual awareness to improve customer interactions.
• Sierra's AI agents can autonomously perform tasks such as processing returns and refunds, reducing the need for human customer service.
• Testing reveals most large language models, like OpenAI’s GPT-4o, struggle with complex tasks, highlighting the need for advancements in AI capabilities.

Sierra plans to share ??-bench with the AI community to foster development of more effective conversational agents and will continue refining the benchmark to improve AI performance metrics.

Full article here.

EKAI: Pioneering AI in Regulatory Compliance

Priya V. Misra is making strides in regulatory compliance by using LLMs to simplify the complex processes, helping firms adapt to new standards with cutting-edge AI.

• His latest tool, EKAI, is crafted for compliance managers and tackles fresh regulatory challenges like Operational Resilience and Consumer Duty. Its easy-to-use chat interface streamlines compliance tasks.
• As regulations demand more detailed compliance evidence, financial institutions are feeling the pressure to clearly show how they meet these tougher standards.
• In the financial services industry, GenAI and LLMs are becoming essential, particularly in areas like compliance and customer support. EKAI is pivotal here, processing vast amounts of unstructured data to boost efficiency and decision-making.

Misra encourages careful adoption of GenAI tools, especially with Europe's upcoming GenAI Act, which aims to set strong standards and foster broad, confident use of AI across various sectors.

Full article here.

Enjoyed this issue? Help grow our community by sharing with friends and colleagues who may find these AI insights valuable.

Thanks for reading!

Blake

Looking to integrate AI into your business workflows if you're a business owner? As the author, I'm offering free 30-minute strategy sessions. Book yours here.