Hello, Starters!

As AI's capabilities expand into various fields, one of the most relevant is healthcare, medicine, and drug discovery. The prospect of this technology streamlining processes to develop life-saving methods is thrilling!

Here’s what you’ll find today:

• Google & Meta present LLMs for medicine

• Introducing: Multi-token prediction

• OpenAI’s Sora debuts on the music scene

• Fei-Fei Li plans to build a “spatial intelligence” startup

• Andrej Karpathy’s approach to AI in space

• And more.

🏥 Google & Meta present LLMs for medicine (1 min)

Google and Meta are pursuing innovation in the health field by presenting language models based on their most powerful LLMs, Gemini and Llama 3. 

Google's offering, Med-Gemini, can understand both images and text and has achieved the top spot in 10 out of 14 medical benchmarks.

Meta, on the other hand, has presented Meditron, an open-source model that the company expects to become a great asset for developing countries and humanitarian missions. The model excels in medical benchmarks, but it has yet to surpass some proprietary models.

📄 Introducing: Multi-token prediction (1 min)

A study conducted by Meta AI, CERMICS, and LISN introduces an approach that alters the functioning of traditional large models like GPT-4. According to the researchers, when AI models are trained to predict multiple tokens instead of just one, as they typically do, there's an improvement in performance, coherence, and reasoning.

This method, called "multi-token prediction," allows the model to focus on long-term dependencies rather than providing immediate predictions. As model sizes increase, further use of this technique offers the potential for faster execution speeds.

🎥 OpenAI’s Sora debuts on the music scene (3 min)

One of the creatives that OpenAI selected to test Sora's capabilities has presented the first music video made with the AI model. Paul Trillo is the mind behind the video for "The Hardest Part," a song by the indie musician Washed Out. The piece was created using 55 clips generated in Sora and then mixed with Adobe Premiere.

This showcases the possibilities we have yet to see in the creative field once Sora becomes available to the public. The question remains, though, how companies like OpenAI will overcome licensing challenges and how this will affect artists... But it seems some are eager to give the model a chance.

💡A leading figure in the AI industry, Fei-Fei Li, is reportedly building a startup that blends human-like processing of visual data to enhance AI's reasoning. This concept, called "spatial intelligence," is set to become a turning point for the technology.

🛸Andrej Karpathy has recently introduced a perspective on the future development of LLMs. Although it might sound far-fetched right now, there's a huge possibility of it happening. The researcher explains that certain LLMs can operate or be transmitted to space if modified accordingly, potentially enabling us to connect with extraterrestrial life.

⚡️Quick links

X launches Stories, delivering news summarised by Grok AI

Microsoft invests in humanoid robots with start-up Sanctuary AI

OpenAI: Reimagining secure infrastructure for advanced AI

Tesla's latest Optimus demo

AI discovers over 27,000 overlooked asteroids in old telescope images

Thank you for reading!