Company behind Arc browser teases a new browser called Dia Browser, an heavily AI focused browser (built on Chromium). Official website at: https://www.diabrowser.com/. Watch the video for a good laugh.
Invidious link to video: https://inv.nadeko.net/watch?v=C25g53PC5QQ
Youtube link to video: https://youtu.be/C25g53PC5QQ
For those not interested in a video, here is a TechCrunch article on the topic.
For those not interested in leaving Lemmy, here is that article -->
The Browser Company teases Dia, its new AI browser
The Browser Company, the company behind Arc Browser for both desktop and mobile, teased its new web browser Monday called Dia — and this time, it focuses on AI tools. In the last few years, the startup launched Arc on Mac and Windows and Arc Search on iOS and Android, but the company is beginning work on a new product with a broader appeal.
The browser is set to launch in early 2025. The startup has launched a new website that shows a video about the browser and lists different open roles in the company.
“AI won’t exist as an app. Or a button. We believe it’ll be an entirely new environment — built on top of a web browser,” the browser’s site reads.
In the video, the Browser Company CEO, Josh Miller, showed some early prototypes of some of its features. One demo showed a tool that works at the insertion cursor, which will help you write the next sentence or fetch facts from the internet when writing about a known subject, such as the original iPhone’s launch and specs. The tool also seems to understand your browser window and can fetch Amazon links that you have opened to insert them in an email with a basic description.
The second demo shows that users can type in commands in the address bar to perform various actions, like fetch a document based on the description, email it to someone based on your preferred email client that you use in the browser, and schedule a calendar meeting through a natural language prompt.
Some of these features sound like what any browser-based writing tools or calendar tools might already do, and we won’t know their usefulness or uniqueness until we actually get to use Dia.
The third demo is more ambitious: It shows the browser doing actions on your behalf, like adding items from an email to your Amazon cart. Dia does it by browsing Amazon on its own, finding these items, and adding them to your cart. In the demo, the list has “an all-purpose hammer,” and the auto-browsing function adds an Amazon listing with two hammers with a grip. I have no idea if that is the right choice, but it’s likely that it isn’t going to make the perfect decision every time right out of the gate — we have already seen that with the Rabbit R1.
Another example shows the browser looking at a Notion table filled with details of members for a video shoot. Dia can email each participant separately.
The Browser Company is not unique in thinking about building an AI assistant that will understand the interface and do tasks for you. Multiple startups have demos, concepts, and visions of AI models and tools that can control your screen.
In a video last month, Miller hinted about building new products for the masses, while assuring current users that it is not planning to meddle a lot with Arc’s design and workings. Miller admitted that while Arc has a passionate and growing user base, its complexity might not appeal to all users. The challenge for the company would be to produce a browser that has AI features that work seamlessly and that could possibly create revenue sources for the company.
Its not that big a deal if the models are local.
Models are not developed and trained locally on low-power devices, however.
And a small cluster like Alibaba used to train Qwen 2.5 is basically a drop in the bucket.
The hoard of GPUs Meta, Microsoft/OpenAI, and especially X have are apparently being used extremely inefficiently, or perhaps mostly not used to train AI at all, but do regular ad/engagement optimization stuff.
"Apparently" according to what source
Are you saying that the bulk of inefficiency comes from network traffic?
Just that bursts of inference for a small model on a phone or even a desktop is less power hungry than a huge model on A100s/H100s servers. The hardware is already spun up anyway, and (even with the efficiency advantage of batching) Nvidia runs their cloud GPUs in crazy inefficient voltages/power bands just to get more raw performance per chip and squeak out more interactive gains, while phones and such run at extremely efficient voltages.
There are also lots of tricks that can help "local" models like speculative decoding or (theoretically) bitnet models that aren't great for cloud usage.
Also... GPT-4 is very inefficient. Open 32B models are almost matching it at a fraction of the power usage and cost, even in servers. OpenAI kind of sucks now, but the larger public hasn't caught on yet.