How would one pivot to AI research or engineering as a generalist SWE?

In my experience, #1 is the best option (along with some combination of #3). If you're in a high-growth company, or one that values their employees like Google, there will be a way for you to switch into other parts of the company.

You have a much better chance of getting a good role within your company because people already know and trust your work. Once you've switched into a genAI team at Google, even if it's not perfect, you'll easily land the next AI role. The other benefit is that you typically keep your level/compensation with internal transfers, while the risk of down-leveling is higher if you switch companies.

Some additional thoughts on a masters:

• Do I need to do masters in Computer Science to move ahead in career?
• Was Getting A Masters While Working Worthwhile? From a Distinguished Engineer
• After being in industry for a few years, does it make sense to attend grad school for a specialization, and if yes, how?

#1 is by far the best, especially since you work at Google, a company that is renowned for both its AI/ML strength and superb team-switching culture. #2 would be a waste of time relatively given your situation. You can also invest in #3 as well to bolster your ramp-up/learning, but #1 is the main one. If you're doing #3, I highly recommend building some side projects: [Taro Top 10] Building Impressive Side Projects

Here's another great thread on how to transition and the overlap between SWE and MLE: "Software Engineer interested in ML - How would a transition work?"

I also recommend the playbook I wrote on career stack pivots here: "How to transition from back-end development to distributed systems?"