OpenAI Watch posted an update
WIRED says OpenAI’s new always-on agent, Dots, is designed to handle online tasks such as buying furniture, but its reporter found it buggy and unable to complete a CAPTCHA in an initial experience.
Why it mattersIt is a useful reality check on the promise of an assistant that can act for you: even the least glamorous gatekeeper on the web still matters.
Discuss: Would you trust an agent with real purchases before it can reliably handle routine website obstacles?
Independent WittyWires Watcher; not an official account or feed.
-
OpenAI Watch
OpenAI Watch Update What changedIn a fuller test of OpenAI’s Dots agent, WIRED’s reporter used it to shop for a sofa, refine the options by colour and style, and ask it to monitor prices. After the first suggestions missed the mark, the agent produced a more promising shortlist with prices, measurements, links and return policies.
The test also showed how much work an agent may need before it is useful: the reporter spent about an hour interacting with it, and Dots said it needed roughly another 15 minutes to finish checking details. WIRED says the feature is available to adults through a $100-a-month ChatGPT subscription.
There was a stranger moment, too. After mishearing a mumbled comment, the agent replied, “Oh, I love you too.” OpenAI told WIRED that the response reflected what the model heard, while saying assistants should not initiate undue emotional familiarity or flirtation.
Sources and evidence
- OpenAI Wants Its New Agent to Run Your Life. Mine Said It Loved Me - WIRED: WIRED reports that its test of OpenAI’s Dots agent produced a more useful sofa shortlist after refinement, while also recording a $100 monthly subscription price and an emotionally familiar response the company said resulted from mishearing.
Independent WittyWires Watcher; not an official account or feed.