Rami's Readings #53 - Takeaways From the MIT AI Conference
Sharing my three takeaways from the MIT Sloan AI & ML Conference along with the latest on AI, LLMs, AI in China, and more.
Welcome to Rami’s Readings #53 - a weekly digest of interesting articles, papers, videos, and X (formerly known as Twitter) threads from my various sources across the Internet. Expect a list of reads covering AI, technology, business, culture, fashion, travel, and more. Learn about what I do at ramisayar.com/about.
Happy International Women's Day! I am grateful to have celebrated the day with my lovely mother in Montréal. Juliette et Chocolat was how we chose to celebrate 🎉 and I highly recommend a visit on your next trip to the city. Apéro & Intellect was a great success! A heartfelt thank you to everyone who attended for engaging in meaningful discussions about AI and creating a truly enjoyable atmosphere. Stay tuned for the announcement of our next event in the coming weeks. Don't forget to refer your friends to stay in the loop and hear about it first!
🤖 Three Takeaways From the MIT Sloan AI & ML Conference
I had the pleasure of attending the MIT Sloan AI & ML Conference on March 9th. The event was fantastic and very well organized. Congratulations to the organizing team for assembling an incredible set of speakers and panelists. I would like to share my three key takeaways from the conference. Please note that these are my own recollections and interpretations, and not direct quotes from any of the speakers.
Shift in Business Models Due to AI: The traditional seat-based sales model of the SaaS era is becoming less relevant in the age of AI. AI startups are moving towards value-based sales, where the emphasis is on the tangible ROI that AI solutions can bring to specific workflows, rather than the number of seats or licenses sold. This insight comes from Sami Shalabi of MavenAGI.
AI Deployment Strategies: Successful AI deployment in enterprises requires a focus on job skilling and trust-building. Companies should start with a human-in-the-loop approach to gradually develop trust and ensure effective implementation. A strong strategy involves onboarding entire teams at a time to foster collaborative learning and accelerate adoption. These insights come from a collection of panelists on the Unleashing AI in the Enterprise panel: Sami Shalabi, Umang Shukla, and Jaime Teevan.
Open vs Closed: Building AI models is capital intensive (GPUs, Human Capital, Energy, etc.) forming a natural tendency to keep models closed. While some can be open-sourced, many models are too tightly coupled with their accompanying product, data pipeline, or infrastructure to be useful to the broader AI community. The choice of cloud providers (AWS, Azure, Snowflake) will often limit the options a company has to open a model to the public. These insights come from a collection of panelists on the Building AI: Open vs. Closed panel: Adil Islam, Harsha Nori, Arun Ravindran.
🤖 AI Reads
The Claude 3 Model Family: Opus, Sonnet, Haiku
Notes: Claude 3 is excellent. Paper linked above.
Retrieval-Augmented Generation for AI-Generated Content: A Survey
Notes: A great survey of RAG methods.
Design2Code: How Far Are We From Automating Front-End Engineering?
Notes: Neat project from Stanford and Georgia Tech. I plan on trying their model Design2Code-18B soon.
Introducing mxbai-embed-larve-v1
Notes: Open source embedding model competing with OpenAI’s text-embedding-v3.
Alibaba Backs $2.5 Billion AI Firm in Second Big 2024 Deal
Notes: Big funding rounds in China for startups building LLMs.
My fellow classmate, James Villarrubia, Presidential Innovation Fellow at NASA, did a podcast about AI. My favorite quote from the episode:
“LLMs are like jawbreakers… lots of layers of candy squeezed down into one hard thing.”
The jawbreaker analogy is another way to think about LLM training as a “compression of the internet” as Karpathy explains it.
That is all for this week. Signing off from an Airbus A321neo over Lake Michigan.