🧙🏼 AI in healthcare gone wrong

Happy Monday, wizards.

I still have a few sponsorship slots available in this newsletter and on whatplugin.ai for Q4! Leading AI startups are advertising through these channels because of the trust I’ve built with my audience.

Ready to reach thousands of early adopters and top professionals? Fill out this quick form, and I’ll be in touch with more details.

—

Let’s dive into what’s brewing in AI today (oh, and here's what you might’ve missed from last week)

DARIO’S PICKS

1. Transcription tools used by doctors make things up, occasionally

A recent study finds that Whisper, OpenAI’s model for speech-to-text transcription, occasionally hallucinates—inventing entire sentences during moments of silence in recordings. Whisper is widely used by several AI companies that do clinical note-taking, including Nabla, which is used by more than 30,000 clinicians, and has processed over 7 million medical conversations so far.

The study, conducted by researchers from Cornell and the University of Washington, found that Whisper added inaccurate or nonsensical phrases in about 1% of transcriptions—that’s 70k potential transcripts with hallucinations from Nabla alone. Some hallucinations included invented medical conditions or even phrases like “Thank you for watching!” (that’s right, OpenAI trained the model on a bunch of YouTube videos).

‎ Why it matters‎ ‎ AI making occasional errors isn’t just limited to medical transcriptions – it’s inherent in the nature of LLMs. One physician in the comment section of The Verge’s original article, noted that you can’t just copy-paste using these tools; proofreading and verifying the AI-generated notes is part of the process—similar to traditional dictation. The big win here is that these tools can reduce documentation time from 5-10 minutes to just 1–2 minutes per patient, allowing doctors to spend more time where it matters most: treating patients.

PS a bit off-topic but can we hear it for creative research titles like this 🎶

TOGETHER WITH WRITER

Writer RAG tool: build production-ready RAG apps in minutes

RAG in just a few lines of code? We’ve launched a predefined RAG tool on our developer platform, making it easy to bring your data into a Knowledge Graph and interact with it with AI. With a single API call, writer LLMs will intelligently call the RAG tool to chat with your data.

Integrated into Writer’s full-stack platform, it eliminates the need for complex vendor RAG setups, making it quick to build scalable, highly accurate AI workflows just by passing a graph ID of your data as a parameter to your RAG tool.

Learn more about our production ready RAG tooling here.

DARIO’S PICKS

2. How 4 major SaaS companies are using LLMs

I’ve taken a look at how some of the big tech companies are using AI models to power up their workflows and products.

Here’s 4 fresh examples on how AI is being used by software companies (and some results):

• Microsoft utilizes Copilot and the new autonomous agents internally across sales, customer support, marketing, and HR departments. Their sales team uses Copilot to power insights and automate routine tasks, achieving a 9.4% increase in revenue per seller and closed 20% more deals. Customer support resolves cases nearly 12% faster, marketing saw a 21.5% increase in conversion rates on Azure.com with a custom agent assisting buyers, and HR improved answer accuracy by 42% with an employee self-service agent.

• Notion uses Claude to improve its product, offering AI features like Q&A, autofill, and writing assistance. These features help businesses like Osaka Gas reduce search time by 35% and save Remote.com 10 minutes per search across 300 queries daily. Notion’s use of Claude even eliminates the need for additional AI tools for some companies like dbt Labs, who estimate they save over $35k annually.

• Gumroad, an e-commerce platform for digital creators, uses Claude 3.5 Sonnet to enable customer support teams to fix issues with code and contribute to product development. This has led to a 300% increase in feature shipping, faster feature deployment, and reduced context switching for engineers. Claude assists them in writing code, locating files, and resolving customer issues.

• Zoom collaborates with Perplexity to enhance its AI Companion by integrating multiple AI models, including Anthropic, OpenAI, and Meta, to provide richer insights during Zoom calls. The AI Companion 2.0 understands user workflows and tracking tasks, in addition to meeting summaries.

‎ Why it matters‎ ‎ From making their product more useful to customers to super-empowering their non-technical teams internally – SaaS companies are embracing AI across their operations. I’m especially impressed by Gumroad: Their customer support team is coding!

Jobs are certainly a-changing.

FROM OUR PARTNERS

Doing the same boring work again and again is exhausting.

What if you had a personal AI assistant who could do the job for you?

Register for this free 3 hour AI Mini Course & learn how to build it.

THAT’S ALL FOLKS!

Affiliate disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes link to products and other newsletters. Please assume these are affiliate links. If you choose to subscribe to a newsletter or buy a product through any of my links then THANK YOU – it will make it possible for me to continue to do this.