Observations from Google I/O 2024
I had the opportunity to join Google I/O earlier this week.
To me, the macro theme was applications. How do we apply generative AI to our work and our day-to-day lives? In my view, for the first time, without any reservation, Google showed us the future of those who use the Google stack - from search to email communication to data storage to information retrieval. There was plenty of technical content but the big theme was how do we use these cool technologies to transform how we do things. From my perspective, Google did a fantastic job demonstrating those capabilities.
For example, summarizing emails and turning them into a list of action items and tasks were popular NLP data science projects that many of my Berkeley students had worked on but they had worked on these ideas before the arrival and availability of large language models. Google showed this capability for the future of Google Workspace. Google also showed an example of what I call "chaining tasks" automatically for us so we don't have to do them. Invoice in the email inbox directly goes into the right Google Drive folder and then information in the invoice is automatically extracted and entered into a Google Sheet. No one wants to do these tasks, and this demonstrates the power of generative AI where the generative models' prompting with correct logic and instructions can transform business processes. This is part of future of work, and I am convinced that we will get to this next phase of process automation aided by generative AI because if I were still a CFO today, I want to look at these tools to help improve productivity. As an individual user, I want to use these tools so I can do more and become even more productive and value-add.
Google launched Gemini 1.5 and variations of Gemini from heavy duty to lighter models for different use cases. Similarly, Google launched new Gemma modes, its open-source model series. It also announced Project Astra, effectively a competitor to GPT-4o (announced day before Google I/O) to grab the multimodal agent model. If we think AI model is race on, it is ON. And the race is among the biggest players in the world. This brings me to the last three points.
Future of data science: Data science has been focused on how to build models, how to fine tune models, and how to make these models better. Today, we have many powerful models at our disposal. So what will data science look like going forward? My view is that the most prized and valued skills of data scientists and machine learning engineers will be problem framing and solution evaluation. There will always be R&D roles to develop new models, but most roles will be applied - working with the organization to identify how to use AI. At the end of the day, everyone will have the same models so the differentiation lies in what we do with these models. Google showed what they will do with their models to add value to their users. I think that is absolutely the right strategy to rise above competition.
Where to compete for AI startups: Find your niche in big markets and solve meaningful problems. Be clear about where you can and cannot beat the big players. Before building and pouring a lot of money to build, I urge all startups to be students of the market.
Advice for current non-AI startups: Evaluate what you can do with generative AI capabilities to add value to your users! It doesn't have to be another product but can generative AI enhance features, reduce user frictions, improve customer success?
My writing reflects my own opinion and is not investment advice.