From our founder
OpenAI just released Atlas, their new Chromium-based browser (for the non-technical readers out there, it’s the same codebase used by Google Chrome and several other browsers). This is yet another browser release, but this time, by THE top player in the industry.
I gave it a spin just hours after their first public release, and I'd like to share some early thoughts here, along with some predictions.
Surprise! OpenAI really IS trying to replace Google
We already knew that OpenAI wants to position itself as a replacement to traditional search engines such as Google. Clearly, this is corroborated by the fact that we are seeing a huge surge in interest in the so-called AI SEO, so much that notable tools such as SEMRush have implemented their own ways to help website owners improve in this area.
With Atlas, OpenAI takes it one step further by - of course - not only asking you to set it as a default browser (that’s a classic), but also by setting ChatGPT itself as default search engine.
It’s worth mentioning that the default results interface still relies on Google when it comes to web data, but I wonder how long this will last before OpenAI decides it doesn't need it anymore and go fully independent.
Agent mode works (kind of)
I admit I haven’t tried agent mode on other browsers as much as I should’ve, but I’m glad that I waited. It is now natively integrated into Atlas, which means ChatGPT can now literally take over tabs of your browser while you are minding your own business on other tabs.
I gave this a try by asking the Agent to visit The Guardian’s website, open their latest article, summarise the key points, then head over to my Google account, open a new Docs file, and paste them there with proper formatting. It started. Then it stopped because I wasn’t logged in to Google. I logged in and handed back control. It continued and completed the task.
Mission: success.
However, it's worth noting that the Agent completely failed when I asked it to listen to some recent WhatsApp messages and transcribe them for me, to download a PDF invoice from an email and upload it somewhere else, or open a chat with an Amazon associate (since it would require opening a new browser tab). But we'll certainly get there.
Considering that this feature is likely to be further improved over time, I can't imagine the massive computing power it takes for such complex tasks to be always available for everyone like it is today. Indeed, OpenAI has already limited the number of times you can use Agent Mode each month, so I expect them to either increase their pricing or sell it as an addon at some point.
I just can't see how this could be sustainable in the long term for the company if they want it to be a feature people regularly start using.
Sensitive mode makes sense... for now
Just like sitting in a self-driving car still requires you to stay alert and watch the road, when the Agent is working with sensitive data - such as emails or private messages - it requires you to keep the browser tab open and monitor what's going on, with the ability of stopping at any time.
I feel this greatly limits what we can do, but only for now. The more Artificial Intelligence advances, the more we will be able to rely on these tools to the point that we will feel comfortable authorising them to deal with our sensitive data. We already do now, though we feel we still have a certain degree of control (spoiler: we don't). I don't think there's any way around it, and it will soon become a requirement if we want to keep pace with others adopting these technologies.
Personally, I envision OpenAI implementing a prompt asking the user for their authorisation to deal with certain kinds of data without keeping the tab open, probably by the next 1 to 2 years.
For now, they still need to improve their models.
Google Chrome is still hard to dethrone
While surfing the web with Atlas on my Macbook Pro (Apple M2 Max) I've noticed that pages loaded in the browser have lost the "smoothness" and responsiveness that they normally would have on Chrome. There is also an incompatibility with certain browser extensions: for example, 1Password's button (which appears inside input boxes) doesn't seem to be working right now.
These factors, combined with a still immature support for the DevTools, which are essential for developers (e.g. no ability to stick them to the bottom or the side of the browser) means that, at least for now, Google Chrome is still hard to take over.
That said, I believe this is probably one of the most complete first public releases of a browser I’ve ever experienced, and it truly points to exciting times ahead.
We'll see what happens.