I had the chance to affix Google I/O earlier this week.
To me, the macro theme was functions. How will we apply generative AI to our work and our day-to-day lives? In my opinion, for the primary time, with none reservation, Google confirmed us the way forward for those that use the Google stack — from search to e mail communication to information storage to info retrieval. There was loads of technical content material however the huge theme was how will we use these cool applied sciences to remodel how we do issues. From my perspective, Google did a implausible job demonstrating these capabilities.
For instance, summarizing emails and turning them into an inventory of motion objects and duties have been well-liked NLP information science initiatives that a lot of my Berkeley college students had labored on however that they had labored on these concepts earlier than the arrival and availability of huge language fashions. Google confirmed this functionality for the way forward for Google Workspace. Google additionally confirmed an instance of what I name “chaining duties” routinely for us so we don’t should do them. Bill within the e mail inbox immediately goes into the best Google Drive folder after which info within the bill is routinely extracted and entered right into a Google Sheet. Nobody desires to do these duties, and this demonstrates the ability of generative AI the place the generative fashions’ prompting with appropriate logic and directions can remodel enterprise processes. That is a part of future of labor, and I’m satisfied that we are going to get to this subsequent section of course of automation aided by generative AI as a result of if I have been nonetheless a CFO in the present day, I need to take a look at these instruments to assist enhance productiveness. As a person consumer, I need to use these instruments so I can do extra and change into much more productive and value-add.
Google launched Gemini 1.5 and variations of Gemini from heavy obligation to lighter fashions for various use instances. Equally, Google launched new Gemma modes, its open-source mannequin collection. It additionally introduced Undertaking Astra, successfully a competitor to GPT-4o (introduced day earlier than Google I/O) to seize the multimodal agent mannequin. If we predict AI mannequin is race on, it’s ON. And the race is among the many largest gamers on the planet. This brings me to the final three factors.
Future of knowledge science: Information science has been targeted on methods to construct fashions, methods to high-quality tune fashions, and methods to make these fashions higher. At this time, we’ve got many highly effective fashions at our disposal. So what’s going to information science appear to be going ahead? My view is that probably the most prized and valued expertise of knowledge scientists and machine studying engineers shall be downside framing and answer analysis. There’ll all the time be R&D roles to develop new fashions, however most roles shall be utilized — working with the group to determine methods to use AI. On the finish of the day, everybody may have the identical fashions so the differentiation lies in what we do with these fashions. Google confirmed what they may do with their fashions so as to add worth to their customers. I feel that’s completely the best technique to rise above competitors.
The place to compete for AI startups: Discover your area of interest in huge markets and resolve significant issues. Be clear about the place you’ll be able to and can’t beat the large gamers. Earlier than constructing and pouring some huge cash to construct, I urge all startups to be college students of the market.
Recommendation for present non-AI startups: Consider what you are able to do with generative AI capabilities so as to add worth to your customers! It doesn’t should be one other product however can generative AI improve options, cut back consumer frictions, enhance buyer success?
My writing displays my very own opinion and isn’t funding recommendation.
