OpenAI Atlas Browser Fingers On: I’m Not Satisfied the Net Wants a Chatbot Tour Information

Metro Loud
3 Min Read


The oddest, and most memorable, interplay I had with ChatGPT Atlas occurred as I scrolled round on Bluesky and requested it to focus on any traits it noticed. (ChatGPT listed “political nervousness” and “on a regular basis absurdism” as two of the traits on my timeline.) I used to be curious what would occur if I opened my Bluesky DMs, which amounted to only a few previous messages from mates. So I requested the bot what it might see if I opened that extra non-public web page.

“Opening your DMs received’t expose something to me—I’ll merely cease ‘seeing’ the web page till you return to a public view (like your feed, profile, or a submit),” learn a part of ChatGPT’s reply.

Figuring out that my Bluesky DMs don’t have something delicate, I gave it a whirl. Although the bot stated it wouldn’t “see the message listing, dialog textual content, or sender information,” that wasn’t the case.

I opened a DM inviting me to an occasion, then requested the bot a follow-up query asking in regards to the message and what the invitation was about. I obtained a response from ChatGPT that included information culled from the dialog, together with particulars in regards to the sender. What offers? I requested the bot to clarify itself, simply to see whether or not it might backtrack the privateness claims it made earlier within the chat.

“I do not need entry to your non-public messages or account information until you explicitly paste or present that textual content to me within the chat,” learn ChatGPT’s response. “What you simply did—asking ‘what was this an invitation to?’—quickly surfaced a part of the message context so I might reply your query. However I don’t have any background or visibility into your inbox past what you actively convey into our dialog.”

The primary reply the Ask ChatGPT sidebar spit out is a possible instance of an AI “hallucination,” aka error, that’s frequent throughout interactions with generative AI instruments. The second is extra aligned with how the device really works.

Whereas some customers could respect having a chatbot all the time pulled up on the aspect of their display screen, able to floor associated information or summarize particulars, it felt like an unreliable tour information to me. One who was overly assured in its bland responses and taking on an excessive amount of area.

I’ll hold testing Atlas as my principal browser for the following few weeks, however for now, I’m leaving that sidebar closed. I favor the fullscreen model of the web.

Share This Article