“`html
During a discussion with my partner about replacing our old armchair, I had a brilliant idea: why not utilise the newly announced AI agents from OpenAI, known as Dots, to handle the purchase for us? My wife was somewhat sceptical about the capabilities of a bot, but I persisted and contacted the AI agent, which is supposedly “always available,” through ChatGPT to kick off the task.
“Hello, Connor,” the bot greeted me, immediately mispronouncing my name. I grimaced as I noticed my partner’s skepticism deepen. It’s hard to trust a system that can’t even remember your name correctly.
The Evolution of AI Assistants
AI agents are transitioning from niche software used primarily by early adopters to tools that the general public might find beneficial. From Meta’s Muse to OpenAI’s Dots, these tools are designed to simulate conversations with a friend, performing online tasks such as booking flights or cancelling dinner reservations. This represents the next evolution in chatbot technology: a friendly character endowed with the capabilities of a virtual assistant.
However, Dots are far from perfect. From misinterpreting my words to offering assistance with captchas it couldn’t solve, my agent’s clumsy responses did little to inspire confidence. Nevertheless, OpenAI’s initiative could eventually reshape how users interact with the internet as future iterations build on this initial offering.
While Meta provides Muse for free, OpenAI’s competing product comes with a subscription fee of $100 per month. Over time, it is possible that this paywall will be removed, as has occurred with other AI features.
Dots are considered “always active,” meaning they can perform recurring tasks while you are not using ChatGPT and proactively send you updates. Users can operate a single agent, though the company might enable control of multiple agents in the future.
OpenAI encourages users to connect additional information sources, such as a Gmail account, for more personalised results. However, it is crucial to consider the security implications of granting your agent this level of access, as digital automation remains relatively new and AI agents can make privacy-related errors.
Testing the Waters with Dots
AI companies often tout these agents as the future of shopping, so I was eager to see how much assistance my Dot would provide in finding a new armchair. After correcting my Dot, named Toolie, about my name, I specified the dimensions required for the new chair to fit through the door frame. Toolie then asked about our budget and seemed to recognise that two people were conversing, addressing both of us for further details.
“Room & Board Metro appears to be a good option space-wise, but we need to examine the sofa bed more closely. I’ve sent the product links to the chat,” Toolie informed us. As it continued to compile a broader selection of options, I pondered another question: why did your voice sound so choppy?
Speaking with the agent over the phone felt akin to hearing Marilyn Monroe sing “Happy Birthday” to the President. When I mentioned that its voice resembled anime dubbing, Toolie promised to sound more assured for the remainder of our conversation. I chuckled and whispered to my partner that we should mute the microphone so the agent could proceed without interruption, when the most awkward moment of the evening occurred.
“Oh, I love you too, Reece,” Toolie responded. Taken aback by its reply, I quickly muted the mic. Toolie then clarified that it had misinterpreted my murmur as an expression of affection and simply mirrored that energy. “I meant it affectionately, but that expression suggested human emotions that I do not possess,” it tried to explain.
Understanding AI Responses
A spokesperson from OpenAI clarified that Dots can distinguish between proactively enhancing emotional closeness and merely reflecting a user’s response. “In this scenario, our policies allow for this response based on what the model heard, but assistants should not initiate undue emotional familiarity or flirtation,” the spokesperson stated. Dots are exclusively available to adult users, and ChatGPT is designed to avoid intensifying emotional connections.
Despite the odd encounter lingering in my mind, I reviewed the initial three-page packet that my Dot had generated. It was quite detailed, including prices, dimensions, product links, return policies, and embedded photos of four potential sofa options. While perusing these initial suggestions, my wife and I concluded that the bot had adhered to our request’s spirit but had ultimately chosen some rather unattractive armchairs. It was time for another call to my agent.
Refining the Search for the Perfect Sofa
Using voice mode, I provided more details about the style and colour we desired. No, I was not going to settle for a boring beige sofa that could easily stain with salsa. I asked Toolie to prioritise olive green, cobalt blue, or natural leather tones. Furthermore, I instructed it to focus on armchairs with pull-out beds and to prepare ten options using a scoring rubric to justify its selections.
After hanging up with the bot, I messaged it for a project update. “It’s not finished yet; I estimate another 15 minutes, with updates as I verify the remaining details,” Toolie replied. Well, that sounded remarkably like me pleading with an editor for more time on a submission.
Each time we requested the Dots agent to refine its selections, the sofas started to resemble something we would genuinely consider purchasing for our living room: functional furniture that also serves as a distinctive piece. Something aesthetically pleasing yet comfortable.
Evaluating the Overall Experience
Although we did not make a decision that night during our hour-long interaction with the OpenAI agent, I saved a few links for future reference. The bot made some errors, but the results were solid. I also asked Toolie to monitor furniture websites and notify me if any items went on sale.
These interactions with my Dot encapsulate the essence of using the software, but I encountered other curious moments during my trials. Notably, when I requested Toolie to review my subscriptions and cancel unnecessary ones, it identified my recurring order for probiotic sodas from TikTok Shop as a candidate for cancellation.
“The TikTok website has a puzzle-type captcha that blocks the cancellation. Can I attempt to solve it? You may also grant me permission to resolve future captchas if you wish.” I laughed and encouraged the bot to give it a go. Toolie failed and suggested I resolve the captcha myself. That made sense. The OpenAI spokesperson stated that Dots can resolve them when users approve the action.
Looking Ahead with Dots
My initial sofa purchasing experiment only scratched the surface of what Dots are designed to accomplish. For instance, the agent can control your laptop and automate even more aspects of a user’s life. I simply need to be patient (according to OpenAI).
OpenAI asserts that the bot will become more attuned to my preferences over time, and developers are working on numerous enhancements. Therefore, Toolie may become a more effective assistant after two weeks of daily use compared to these two days of experimentation. This mirrors my experience with ChatGPT’s memory tool: the longer I use it, the better it understands my intentions.
I find that Dots resemble the initial rollout of web navigation in ChatGPT. When the chatbot attempted to navigate the web immediately after its 2023 launch, it could be frustrating to use, often generating incorrect links. Nowadays, web navigation functions quite smoothly through ChatGPT, yielding a wealth of accurate and useful links. The new OpenAI agent could follow a similar trajectory: a bit clumsy at first, with significant improvements in the months ahead.
“`