Email to Arc

I used to use Arc browser every day. But unfortunately, good things can't last, and the company behind some of my favorite design has decided that it was all a mistake. They're pivoting all of their work to a new browser where AI is integrated into every part of the experience. The CEO put his email in to the blog post asking for feedback, so this is what I sent. I'm not optimistic that this will ever be read by someone at the company, but I appreciated their work in the past and if there's a chance some real criticism can pierce the VC bubble, I'll take it.

Hey, I read your blog on the future of Arc.

I was an enthusiastic supporter of Arc, using it as my daily browser. I evangelized it to my friends and got them to give up chrome for it. I felt like the fact that someone was reimagining the browser from user focused principles was almost too good to be true.

Your announcement is disappointing to me because I think it fundamentally misunderstands what AI means to consumers. I am someone who approaches new technology with an open but skeptical approach. Fundamentally integrating generative AI, particularly large language models, into the fabric of our software is a huge mistake. To be as clear as possible, I will never use a so called 'AI Browser', or at least I will hold out until tech companies like yours destroy all alternatives.

Filtering the internet through an LLM has a lot of critical issues. The biggest one is the lack of context. AI optimists tend to think that 'hallucinations' are a problem that will go away soon. They won't. Such problems are fundamental the regression-based architecture of LLMs- adding more parameters and overfitting your model has exponentially diminishing performance returns. 95% or 99% success rates or accuracy sound really good- until they're part of critical software that you use every day. Imagine if a computer only had a 99.9% success rate at correctly evaluating an if statement in code. We simply would not trust that system to do important things. And LLMs are nowhere near that for any task and have no path to being that reliable.

Worse still, imagine that an LLM system had a 0% hallucination rate. These ways of presenting information are a disaster on the internet. False info, scams, and misinformation have been a problem on the internet since its creation. But internet users have built up a form of literacy for navigating that. We know what kinds of things feel off or sketchy. It's not perfect, but at least we have a fighting chance. LLMs surgically excise that context and sane-wash all information. The ability to employ skepticism is taken away. The only recourse a user has is to, without any particular evidence, decide if they trust the info that has been spat out at them. You have to gut check everything, which is more dangerous when you're not an expert. Or you give your critical thinking over to the model. Many people incorrectly think that ChatGPT is an intelligent system, capable of thought and critical thinking skills, and that its outputs have more credibility. This is a result of the diminution of internet literacy. We're actively acculturating large swathes of users to be more susceptible to AI slop, scams, and false information by validating that reading an LLM's output is the same thing as research. Unfortunately, Dia appears like it will be another sad chapter in that attack on our collective thinking.

I'm sure this will not dissuade you from going down this path. After all, I don't have any venture capital and my opinions can't make you rich! But you should know that you're participating in the normalization of a technology that is actively attacking the conditions that made the internet an interesting and worthwhile place, cannibalizing the creative output of a generation for a few brief shining quarters of growth. I'm glad you're not actively destroying a good product you made to jam AI into every interface, which makes this better than your average company, but unfortunately that's a low bar. I hope you and your team experience success, not with Dia, but with some other product that will actually be good for the internet and the world.

-Tim C