All opinions are my own and aren't representative of Microsoft.
"the matrix background in dark blue with AI lightbulbs digital art" - DALL·E.
Two AI announcements today; the release of GPT-4, and generative AI in Google Workplace, met with a mix of excitement and skepticism.
Firstly, we've seen incredible progress. GPT has gone from failing the bar exam to passing with flying colors, bottom 10% to top 10% of results from GPT-3.5 to GPT-4. This is incredibly fast progress and they're just getting started. Friends excitedly tell me how they've drafted reports, generated scripts for side projects, and brainstormed business names. There's significant real world value and clearly there's plenty more growth in the S-curve. Which tasks come next? GPT is eating the world.
Secondly, there was a a healthy dose of skepticism to contrast the progress. Many comments on Hacker News made the point that AI will help us to generate long form content from prompts and then we'll use the same AI to summarize the content, so what's the point of using it in the first place? See what Google generates in response to a mail with the prompt "I'm on it". If you knew someone hadn't typed this reply themselves, would you still consider the reply to have a friendly conversational tone, or would it feel fake?
Finally, people are highlighting the second order effects. As a manager, you might want to automate writing an intro mail for a new hire to save yourself time, but what does it signal to your team? Does a three-page report hold the same weight now that it can be generated in minutes? The JTBD for intro mails is to build team connection. The JTBD for reports is to indicate the solutions have been well reasoned. We might think we're solving our problems with generative AI, but if we're not careful our solutions might come across as out of touch. We have a powerful new hammer, but we're just getting started with how to use it.