Microsoft's strategy for winning the AI race

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each weekday, we scour the web to aggregate the many stories related to artificial intelligence; we pass along the news, useful resources, tools or services, guides, technical analysis and exciting developments in open source.

In today’s Dispatch:

• (Long read) Fast Company explores Microsoft's strategy for winning the AI race and becoming a leader in the next major wave of computing. They trace how CEO Satya Nadella has reoriented Microsoft towards AI over the past decade through research investments, organizational changes, and product development. The article offers an insightful view of Nadella and an inside look at a legacy tech company's efforts to reinvent itself and compete in a rapidly shifting AI landscape.

• Just in time to support the above article, Microsoft and electronic health records provider Epic are deploying dozens of AI ‘copilot’ solutions across Epic's clinical profile. Epic oversees the most health records data in the US, and boasts over 3,000 healthcare organizations (including Cleveland Clinic and Johns Hopkins) as clients. The partnership claims these AI deployments will enhance patient care, increase operational efficiency, improve healthcare experiences, and ‘support the financial integrity of health systems globally’.

• YouTube is partnering with Universal Music Group to launch ‘Music AI Incubator’ - and they’re bringing major artists into the development process. Initial participating artists span genres and generations, including Anitta, Yo Gotti, and the Frank Sinatra estate.

Plus: Detecting Parkinson’s seven years before symptoms with AI, exploring ChatGPT’s hidden capabilities through chess, trending tools and more.

SeamlessM4T: Meta’s new open source AI translator

From Meta AI’s blog: Meta AI has just released SeamlessM4T, a new multimodal AI model that can translate speech and text across nearly 100 languages. Meta has been working on translation projects since last year, and this model incorporates elements from many previous projects. The AI represents another step towards the long-held goal of a universal translator.

More details:

• M4T supports speech recognition, speech-to-text translation, speech-to-speech translation, text-to-text translation, and text-to-speech translation for close to 100 languages. It’s the first single model capable of translating directly between speech and text for so many languages.

• It was trained on a huge multilingual dataset called SeamlessAlign, which contains over 270,000 hours of speech-text alignments across 443,000 hours of speech data.

• One of the interesting features of SeamlessM4T is its ability to recognize when a speaker is code-switching, or when someone moves between two or more languages in one sentence. For instance, Meta demonstrated in a video that the model immediately differentiates between Hindi, Telugu, and English.

Takeaways: This is an exciting step in Meta’s vision for connecting people across languages; the blog mentions they’ll be exploring how to use this foundational model for future capabilities. Open-sourcing the model encourages the research community to build on Meta’s foundational work and help carve out a quicker path to a Star Trek-like future. Meta acknowledges more progress is needed to ensure accuracy; but from our initial testing of the demo, translations of high-resource languages are excellent.

Download the code, model and data for from the GitHub repo here. Meta has noted that the model is not for commercial use.

Eye scans powered by AI could detect Parkinson's disease before people have symptoms

From The BBC: Researchers from London's Moorfields Eye Hospital and the UCL Institute of Ophthalmology have utilized AI-powered eye scans to identify potential early markers of Parkinson's disease in patients. These scans could provide a way to detect the disease even before patients manifest any symptoms.

More details:

• The AI was used to analyze data from OCT scans, which generate detailed images of retinal cross-sections (they illustrate the layers of cells in the eye). The dataset came from 154,830 patients aged 40 and over who had attended eye hospitals in London between 2008 to 2018.

• A repeat analysis was performed using data from 67,311 healthy volunteers aged between 40 and 69.

• The research revealed that individuals with Parkinson's exhibited a thinner ganglion cell-inner plexiform layer and inner nuclear layer in their eyes. These markers were identified, on average, seven years prior to any clinical presentation.

Takeaways: Detecting these conditions before symptoms manifest would create an invaluable window for preventative interventions and lifestyle adjustments. This technique shows promise for Parkinson's, but also importantly reinforces the concept that our eyes can provide profound insights into our overall health - data from eye scans has previously revealed signs of other neurodegenerative conditions. As the technology is scalable, non-invasive, cost-effective, and swift, it has potential to impact public health on a vast scale: 90,000 people are diagnosed with Parkinson’s each year in the US alone.

From our sponsors, Cerebrium:

Cerebrium: Serverless, Seamless, ML Deployment

Cerebrium offers seamless ML Model deployment: Less than 1s cold start, major frameworks (Pytorch, Onnx, XGBoost) supported, 18+ pre-built models, fine-tuning (FlantT5, GPT-Neo, Stable Diffusion), and opportunities for paid projects. Trusted by Twilio, Ramp, and Writesonic. Try for free.

Try Cerebrium For Free!

Social media/news/video/podcast:

• Crypto bot network powered by ChatGPT uncovered on X [Mashable]

• Prof. Jürgen Schmidhuber - the ‘father of AI’ on its dangers [Podcast]

• (Discussion) There's No Such Thing as Artificial Intelligence | The term breeds misunderstanding and helps its creators avoid culpability. [Reddit]

• Dolphins swimming in a Chinese subway? [X]

• Explained: The conspiracy to make AI seem harder than it is! By Gustav Söderström [YouTube]

Did you know? 

Popular AI text-to-image generator Midjourney has just incorporated a new ‘Vary Region’ editor into its Discord platform. This long-anticipated feature allows users to re-imagine individual sections of a generated image. From our experimenting, you could effectively highlight multiple areas within one image for simultaneous editing. The feature works surprising well and is a great addition to an already polished product.

Midjourney’s small team (44 employees providing service for 16+ million very global users - only 17% of Midjourney users are from the US) has managed to stay ahead of creative software giant Adobe (and open source competitor Stable Diffusion) at the top of generative AI text-to-image service quality.

Trending AI Tools & Services:

• Intentional AI: AI copilot for habits and goals

• Radiant 3.0: an AI DJ in your pocket

• Revoice: create a digital copy of your own voice.

• DocumentationLab: simplifies the documentation process for developers by providing an AI-powered software documentation tool

• (App) Dreamlife: allows users to transform any photo of a room into a completely new style