📦️ OpenAI loves data

🗞️ Your Weekly Round-Up

1️⃣ Google’s latest healthcare-focused generative AI model

Google recently unveiled MedLM, a set of generative AI models aiming to assist medical professionals in their day-to-day tasks. 

So far, the models have been tested by HCA Healthcare and BenchSci for drafting patient notes and biomarker identification.

But as with all AI technology, caution is urged, as historical challenges in healthcare AI include the possibility of generating incorrect answers, spreading misinformation, and revealing sensitive health data.

Read more: TechCrunch

2️⃣ Snap’s newest AI adventure

Snapchat is back at it. 

Snapchat+ subscribers have now been gifted new AI-powered features that allow users to generate and send AI images based on text prompts.

The AI image generator offers creative possibilities, allowing edits, downloads, and sharing. 

Plus, the “Dream selfie” feature gets an upgrade, enabling users to include friends in “fantastical scenarios.”

Let’s hope that this model doesn’t hallucinate like their MyAI Bot 🤒 

Read more: TechCrunch

3️⃣ OpenAI loves data  

After all the drama, OpenAI is back to its regularly scheduled programming and is looking to ramp up its data quantity. 

Most recently, the AI company partnered with German giant Axel Springer, home to brands like Business Insider and Politico.

The deal? OpenAI will be shelling out tens of millions of euros over the next three years to leverage Axel Springer's news articles for training its cutting-edge AI models. They will also get archives from all Axel Springer's brands to feed OpenAI's LLMs.

The goal? To transform how people access real-time news through AI tools while providing new revenue sources for publishers.

Read more: Engadget

4️⃣ Microsoft’s Phi-2 Model

Following the success of Phi-1 and Phi-1.5, Microsoft's research team has launched Phi-2.

Here’s the TL;DR:

• Improved performance: Phi-2 excels in common sense reasoning, language understanding, math, and coding, surpassing models with significantly larger parameters.

• Compact Size: With only 2.7 billion parameters, Phi-2 outshines larger models like Llama-2 (up to 70B parameters) and Gemini Nano 2 (3.2B parameters) on various benchmarks.

• Training Innovations: The model leverages high-quality training data, including synthetic datasets for common sense reasoning and implementing scalable knowledge transfer techniques. 

Read more: Microsoft Blog

5️⃣ Image to 3D

Stability AI recently launched Stable Zero123, their latest image model that promises a giant leap in 3D model creation.

Here’s the TL;DR: 

• Unlike its predecessors, Stable Zero123 isn't a direct 3D model generator. Triggered by a simple text prompt, the model transforms an object's image into a series of panoramic views from different angles—a crucial step in the 3D creation process.

• Stable Zero123 surpasses its forerunner, Zero123-XL, thanks to a superior training dataset curated from the Objaverse dataset. 

• The model boasts a 40x increase in training efficiency compared to Zero123-XL.

Read more: Stability Blog

6️⃣ Nuclear energy x AI?

Microsoft is experimenting with generative AI to streamline the lengthy nuclear regulatory process, a significant hurdle for nuclear power projects. 

The process involves training a large language model with nuclear regulatory documents to expedite the complex approval paperwork, which can generally take years and cost millions.

It’s also interesting to note that the partnership between data centers and nuclear power is symbiotic: AI helps speed up the regulatory process, while more nuclear power can offer constant power generation crucial for AI processes - a win-win situation.

Microsoft is even seeking a nuclear power expert to integrate smaller modular reactors (SMRs) into its energy planning that can help support the company’s AI operations.

Read more: WSJ

7️⃣ UK courts allow AI to help with rulings

The UK Judicial Office has given judges the green light to use AI when crafting legal rulings and various tasks. 

While praising AI's utility, the guidance urges caution, emphasizing the need for judges to verify the accuracy of AI responses before making life-altering rulings. 

Thoughts: This particular use case is quite surprising, especially given the instances of imaginary quotes and citations generated by ChatGPT that led to fines for lawyers in New York.

It will be interesting to see how this impacts legal rulings.

Read more: Gizmodo

8️⃣ AI’s impact on climate change research

The U.S. Department of Energy (DOE) has established the Office of Critical and Emerging Technology to understand how the US can exploit cutting-edge technologies, including AI, biotech, quantum computing, and semiconductors.

Their mission? To combat climate change, address pandemics, and fortify national security. 

Their newly-appointed Chief Artificial Intelligence Officer, Helena Fu, plans to:

• Leverage DOE's extensive resources, including grant funding for AI startups and access to cutting-edge supercomputers. 

• Prioritize the security of AI models, ensuring ethical use and facilitating broader access to DOE's wealth of data.

Read more: Semafor

9️⃣ How will AI impact news production?

In an interesting move, the NY Times created a new newsroom position: Editorial Director of AI initiatives.

The role will be focused on the following:

• AI Integration Strategy: NYT wants to develop a plan for strategic AI integration, emphasizing principles for the ethical use of generative AI. 

• Navigating the Future with AI: NYT wants to understand AI's role in journalism and how they can remain committed to human-driven journalism. 

• AI as a Collaborative Tool: NYT views AI as a collaborative tool rather than a sole content creator. That means understanding AI's potential to assist human journalists in enhancing tasks such as headline and image generation.

Read more: CNN

🔟 Can AI be a diagnostician?

Physician-investigators at Beth Israel Deaconess Medical Center shared interesting insights into the synergy between AI and medical decision-making. 

Published in JAMA Network Open, their study compared human clinicians against Chat GPT-4 in probabilistic reasoning - a key aspect of medical diagnosis - across five medical cases. 

The outcome? 

• When the results were negative, the chatbot was more accurate in making diagnoses than humans in all five cases.

• When the results were positive, the diagnoses remained comparable in two instances, with the bot outperforming humans in two and lagging in one.

Read more: Mirage News

🎙️ BWAI: Season 1 Wrapped!

Hey Team! For the last couple of months, we've had the privilege of hosting some phenomenal guests who've generously shared the secrets of what it truly means to Build with AI.

In case you've missed any of these episodes, don’t worry! You can catch up with us on YouTube, Spotify, or Apple Podcasts today!

Your continued support means the world to us, and we can't wait to bring you more exciting content. Stay tuned, and keep building with AI!

🤑 AI Fundraising News

• Citrusx raises $4.5M in Seed funding to create a platform that offers model validation, governance, and monitoring to help stakeholders make better decisions about machine learning in regulated industries.

• Solvento raises a $50M debt facility to help supply chain managers automate their accounts payable process, enabling them to reduce the time and cost of manual invoice processing. The firm's software uses artificial intelligence to extract invoice data and automate approval workflows.

• Relevance AI raises $10M in funding round to offer a low-code platform for custom AI agent creation.

• Tacto raises $54M/€50M in funding to offer AI software to optimize supply chains and identify cost savings.

• Chalk raises $10M in Seed funding to create a platform that enables developers to declare features and their dependencies with Python, allowing for efficient and consistent training sets for data scientists and live feature values for models.

• Essential AI raises $56.5M Series A funding to enhance human-computer collaboration through its full-stack AI products, including LLMs that automate tasks and scale user capabilities.

• Kapital raises $40m in Series B funding and $125m in debt financing to provide financial visibility to LatAm SMBs.

• Arcane raises $5M in Seed funding to develop its AI assistant, automating day-to-day tasks and integrating various marketing channels, grouping all disconnected data and tasks into one platform.

🛠️ AI Tools to Check Out

• Watto AI:  Ultimate writing companion for professionals.
  Seamlessly integrate your data from multiple platforms & generate documents in a single click. Check them out!

• JazzUp AI: A cutting-edge solution designed to enhance customer segmentation and targeting for e-commerce businesses. Check them out!

🐦 Tweet Post of the Day

The AI x Robot combo is going to be interesting to follow next year 🤖