Testing artificial intelligence systems on five everyday tasks

We tested three AI agents—ChatGPT's Dots, Gemini's Spark, and Meta's Muse—across practical automation and integration tasks.

Which systems were tested?

Three systems from major tech companies participated in the test. The first is Dots, a feature offered within ChatGPT and aimed at subscribers of more expensive business packages starting at $100 per month. The second is Google's Spark, which operates within Google's traditional environment and is available to subscribers of specific company plans, at a cost of $20 to $100 per month. The third is Meta's Muse, which is offered free to those with a Meta account, with the option to purchase additional packages starting at $20 per month.

How did they do in their daily tasks?

The three systems were tested on five different tasks, including creating a supermarket shopping list, searching for jobs, compiling information into spreadsheets, finding restaurants to book, and clearing out junk email.

In the grocery store, Dots was able to connect, find ingredients for a recipe, and add them to the online cart. Muse required manual code entry and added expensive ingredients that were not requested, while Spark froze and was unable to connect to the account.

In job searches, all three systems read resumes from linked email accounts. However, only Dots found relevant jobs with direct links. Spark provided general suggestions, while Muse provided links to aggregated websites or expired ads. Spark was also able to use a personal information feature to find an up-to-date resume stored in a Google service.

In the information search, all three systems successfully collected information about characters and missions from a game website and organized it into a spreadsheet.

For restaurant reservations, Dots found a nearby restaurant with an available table and made the reservation. Muse and Spark were limited to providing lists of restaurants, some of which did not accept reservations.

Finally, in email purging, Dots scanned through years of messages, aggregated senders into a spreadsheet, and followed delete links through its own browser. Muse didn't use its browser to follow delete links, while Spark simply provided links for the user to do the process themselves.

Read the original English article on Hacks.gr