Say Hello/Sod Off to the World's first AI Centric Browser

[Lenny Solidus]

Oslo, Norway – May 28, 2025 – Opera, the Norwegian browser company [NASDAQ: OPRA], is announcing Opera Neon, a new agentic browser that rethinks the role of the browser in the coming generation of the AI agentic web. A result of years of development, Opera Neon is a browser that can understand users’ intent and perform tasks for them as well as bring their ideas and needs to life on the web. To achieve this, Opera Neon introduces agentic AI browsing capabilities that go beyond traditional browsing and turn user intent into action.

https://press.opera.com/2025/05/28/oper ... c-browser/

[Sly Boots]

Oh FUCK OFF

[Wrathbone]

Opera Neon is a browser that can understand users’ intent

Can it, though? Can it?


(No, probably not.)

[Rusty]

O sometimes wonder at my own intent when browsing. I head for specific websites, forums/reddit/facebook (sometimes)/Amazon.
That's it. Where would the AI come into play here?

Am I internetting wrong?

[Stormbringer]

I'll be adopting it, for sure. I use Opera for work, and am interested to see if this will do my job for me, so I can spend more time doing fun things (like running virtual text-based RPGs) in Chrome ¬_¬

[Mantis]

Very savvy of them to launch an entire browser with the AI gimmick as the main selling point.

Can't imagine it'll actually be all that impressive.

[Raid]

I'll be adopting it, for sure. I use Opera for work, and am interested to see if this will do my job for me, so I can spend more time doing fun things (like running virtual text-based RPGs) in Chrome ¬_¬

I'm interested to know what you get out of it, because the ad is yet another "look at the possibilities of AI!" rather than any actual real-world examples or workflows. You say it can book me a walking trip to Bergen? Ok, show me that happening, show me the results. Stop telling me the possibilities and just bloody demonstrate it already. I can't ever take these AI products seriously because there's no evidence that they're capable of the things they say they are, at least in any reliable sense. I'm willing to bet that the Bergen walking trip, 2 times out of 5, will recommend taking a helicopter because that's a possibility and it fits the brief rather than a sensible option.

I am not against AI, at least not in non-creative fields, but I swear that the entire industry is based around the concept of selling the development company for billions based on the possibilities of the technology, not developing an actual product.

[Wrathbone]

You say it can book me a walking trip to Bergen? Ok, show me that happening, show me the results. Stop telling me the possibilities and just bloody demonstrate it already. I can't ever take these AI products seriously because there's no evidence that they're capable of the things they say they are, at least in any reliable sense. I'm willing to bet that the Bergen walking trip, 2 times out of 5, will recommend taking a helicopter because that's a possibility and it fits the brief rather than a sensible option.

Your walking trip to Bergen will be just over 1500 miles and take approximately 21 days, including an exciting swimming expedition across the North Sea. Your default credit card has been charged £5000 to cover expected medical requirements upon arrival at your destination. Would you like to view the resignation letter I've sent to your employer to accommodate the trip?

[Stormbringer]

[Wrathbone]

I've just tried using Firefox's built-in ChatGPT toolbar on a work problem, which seemed like an ideal use case. This will likely be confusing and boring for anyone not versed in Power BI, but in short, I was having problems getting a chart to filter correctly to a table of relevant rows, because each month in the chart had two columns and it wasn't making a distinction between them. AI to the rescue, right? I gave it a detailed breakdown of the issue and intended goal, sent it screenshots that it analysed, etc, and it came up with a solution using an intermediary table. It seemed overly complex, but I decided to go with it and see what happened.

Immediately the DAX code it gave me for a measure didn't work, because it was referencing columns in a context where it couldn't do that. I pointed that out, so it fixed that with some updated code... which had the exact same issue in a different place. After going round in circles for a while, I fudged it into something that didn't error and achieved what ChatGPT was trying to do for that step. The measure we'd created was supposed to filter results on the table visual... except it didn't give any values to filter on. When I stated that to ChatGPT, it decided the solution wouldn't work and set about a whole other solution, which was even more complicated. And it turns out it didn't work, because it required manual filtering on results, rather than doing it automatically upon selecting a column.

Eventually, without the help of ChatGPT, I realised that it would work very simply by having a single column for the value counts per month and then a legend column to distinguish the values. Took a couple of minutes and it works perfectly.

This is why LLMs are such a mess right now. Yes, they're continually improving, but until they're capable of producing start-to-finish solutions that actually work, rather than providing fancy solutions that look impressive but have all the real-world applications of an arse on an elbow, they're not going to be replacing developers. They're an unreliable tool at the moment (which admittedly describes quite a few developers).

[Animalmother]

I understand fuck all about this. Will it eventually come to the stage where the AI is using samples of work/code that another AI created by copying another AI ....until the "gene pool" is exhausted and everything is just the same mush?

[Wrathbone]

I think that's kind of what's happened here. I tried the same query with Gemini and it went down the exact same route as ChatGPT, which suggests they're applying the same flawed logic or they're scouring the internet for potential solutions and using the same one. I'd wager there's an old tech forum post somewhere with a vaguely similar issue that the AI has latched onto, and it ticks enough boxes for the AI to decide it's good enough to try, even though the end result doesn't work.

That's a vast simplification of what's actually happening, I'm sure, but I believe the general way it works for problems like this is it says "X people marked Y solution as a success for problems matching keywords A, B, C, etc, therefore this is the best starting point."

[Alan]

I think opera and Firefox are struggling these days especially since they are losing out on that sweet Google default search engine money. This is a bit of a desperate move tbh

I’d pay for a browser that can filter ai out. I googled the differences between Werdsmith and Ulysses and the forced AI first result said that Werdsmith is an excellent word processor with many advanced features and Ulysses is a novel by James Joyce.

[Raid]

I saw a stat a little while back that an AI query uses five times the energy to process than an internet search would. Google, in their infinite wisdom, made AI summaries of their search results happen automatically, so presumably their search processing increased in energy usage sixfold. Given the rising costs of energy these last few years (and I'm assuming that the energy need of a data centre would vastly outstrip the energy generation of covering that data centre in solar panels), it feels a bit mad that they're pushing this rubbish as hard as they are.

[Animalmother]

Fresh water for data centres will be the new oil probably.