Alberto - another great piece that soundly balances hope and hype. As someone in the enterprise tech space, I am intrigued by this potential product. It is a very clever way to layer NLP on top of billions of dollars of existing software investments. It will be interesting for Adept to release more details on the state of ACT-1. The demos are slick but there is no further evidence of capability. I’m also curious how the model was trained. I agree that there are challenges to creating an NLP that can handle every user intent for an Excel spreadsheet. But I could see where it could still be highly effective. I’m guessing that a lot of spreadsheet use follows a Pareto distribution. A small number of capabilities like highlighting, or creating a profit column, likely make up a disproportionate volume of user activities. Just solving for those use cases could provide significant lift.
Thanks Dan, I'll make sure to update this article (or write another one) once Adept discloses more details about ACT-1. I'm quite sure we'll start seeing more research in this direction--and eventually products and services.
Great article, Alberto, even though I read it only now, it's still relevant. Are there any updates? Have you come up with more possible solutions to build such AI agents?
Alberto - another great piece that soundly balances hope and hype. As someone in the enterprise tech space, I am intrigued by this potential product. It is a very clever way to layer NLP on top of billions of dollars of existing software investments. It will be interesting for Adept to release more details on the state of ACT-1. The demos are slick but there is no further evidence of capability. I’m also curious how the model was trained. I agree that there are challenges to creating an NLP that can handle every user intent for an Excel spreadsheet. But I could see where it could still be highly effective. I’m guessing that a lot of spreadsheet use follows a Pareto distribution. A small number of capabilities like highlighting, or creating a profit column, likely make up a disproportionate volume of user activities. Just solving for those use cases could provide significant lift.
Thanks Dan, I'll make sure to update this article (or write another one) once Adept discloses more details about ACT-1. I'm quite sure we'll start seeing more research in this direction--and eventually products and services.
Great article, Alberto, even though I read it only now, it's still relevant. Are there any updates? Have you come up with more possible solutions to build such AI agents?
Not much to update except perhaps this from OpenAI: https://www.theinformation.com/articles/openai-shifts-ai-battleground-to-software-that-operates-devices-automates-tasks?rc=j0xnsg