The web browser Vivaldi is taking a firm stance against including machine learning tools to its browser.
So, as we have seen, LLMs are essentially confident-sounding lying machines with a penchant to occasionally disclose private data or plagiarise existing work. While they do this, they also use vast amounts of energy and are happy using all the GPUs you can throw at them which is a problem we’ve seen before in the field of cryptocurrencies.
As such, it does not feel right to bundle any such solution into Vivaldi. There is enough misinformation going around to risk adding more to the pile. We will not use an LLM to add a chatbot, a summarization solution or a suggestion engine to fill up forms for you until more rigorous ways to do those things are available.
↫ Julien Picalausa on the Vivaldi blog
I’m not a particular fan of Vivaldi personally – it doesn’t integrate with KDE well visually and its old-fashioned-Opera approach of throwing everything but the kitchen sink at itself is just too cluttered for me – but props to the Vivaldi team for taking such clear and firm stance. There’s a ton of pressure from big money interests to add machine learning to everything from your operating system to your nail scissors, and popular tech publishers are certainly going to publish articles decrying Vivaldi’s choice, so they’re not doing this without any risk.
With even Firefox adding machine learning tools to the browser, there’s very few – if any – browsers left, other than Vivaldi, it seems – that will be free of these tools. I can only hope we’re going to see a popular Firefox fork without this nonsense take off, and I’m definitely keeping my eye on the various options that already exist today.
It is a very good move by the vivaldi team. i tried to use vivaldi as my main pc browser, but as it uses tye blink engine i have some minor annoyances that made me go back to firefox. i still use it occationally on my work phine though.
>”there’s very few – if any – browsers left, other than Vivaldi, it seems – that will be free of these tools”
Pale Moon won’t have any AI added to it. That’s another one that will continue to be free of AI.
First of all, I still use Pale Moon for some specific tasks, and I will continue do so for the foreseeable future.
Having said that, unfortunately it’s rendered obsolete. Not by any fault on part of its developer or its community, but because of the way the web evolved in the past decade.
Pale moon and the old Firefox engine it runs on is failing on more and more sites. I know that’s because of the laziness of the website developers, but that’s the case.
And then the extensions… I guess only one tenth of the extensions on the Pale moon site got any update in the past 3 years. And that may be a generous estimate.
It was a noble attempt, and I have every praise to the developer for the huge work undertaken. Realistically however, it’s just part of the good old days we will always remember and cherish. I will always wish my browser didn’t clutter my task manager with 20 processes.
I have been with Firefox for more than a decade. I desperately want there to be an alternative to Chrome.
But the Mozilla foundation has completely lost me and I can’t stomach them any more. And with Firefox having an irrelevant share of the market, I have started having a lot of web sites that just don’t work with it any more. Therefore, I decided to move to a Chromium-based browser.
I switched to Vivaldi a couple months ago and have been quite happy with it. Yes, it has way too many features (please don’t include a half-baked email client). The good news is that the interface is highly customizable, so I just remove 75% of the controls to get a clean interface.
I wish the Android version allowed y0u use uBlock Origin, but I understand why they don’t.
I have been thrilled to see them state they won’t include AI.
The only complaint I have with them is that their bug-reporting and development process is opaque. If you submit a bug report, the only way to track it is to periodically send an email to them and ask the status of the bug. That’s ridiculous.
Vivaldi is currently the best browser solution in my opinion.
Good for Vivaldi! I have used Vivaldi as one of my preferred browsers for years, and while it does have its quirks, it does fine for what I need it for and I am very happy I won’t have to deal with any AI garbage while using it. I have absolutely NO interest in AI whatsoever, don’t want it and will never use it because of the gigantic privacy concerns primarily.
I checked, for Vivaldi to be free user privacy must be sold. So if people in general accept a concept, there is no more privacy, then i guess AI adoption is not something people will in general fight against. At least not from the privacy point of view.
I do not think the term “machine learning” is a good substitute for “AI.” It is too broad. For example, if Vivaldi has mouse gestures, then it uses machine learning. Instead, if you really want to avoid saying AI, I suggest simply “chatbot.”
Good post and I like and endorse Vivaldi’s move, but this blog post is from 6 months ago. The dateline is February.
It’s being promoted the past few days by Vivaldi on Mastodon after Mozilla’s recent announcements.
Also trending right now is Ladybird browser.
Somewhat relevant in this context:
https://link.springer.com/article/10.1007/s10676-024-09775-5 (Hicks, Humphries & Slater, 2024 june 8, ChatGPT is bullshit)
Basic argument of the paper is that using words like “hallucinations” and “lies” about LLMs is actually a misleading anthropomorphisation, because the way they function they’re incapable of considering what’s true or not. They’re purpose-built to generate human language and nothing more. They’re statistical next-token generators and there is no mechanism for building or considering a consistent model of the world. Thus, they’re always inherently bullshitting, even when they happen to generate true statements.
An interesting point made is that to make the models “creative” and generate variation in their output, you have to raise the “temperature” of the model, that is the randomness. But doing that also increases the chance that the statistics-based language it outputs will be factually incorrect.
And the lower the temperature of the model is, the higher the chance that it’s just going to output copyrighted material verbatim (because that’s where the statistical correlation is highest). That is, the answers might be more likely to be correct on a lower temperature, but then that would be because it’s effectively just copying text it’s seen.