At the 2024 Meta Connect conference earlier this week, CEO Mark Zuckerberg unveiled several products, including the Orion augmented reality glasses and its new affordable mixed reality headset, the Meta Quest 3S.
However, he didn’t talk much about Llama 3.2, the new multimodal version of the LLM that the company is trying to use to conquer this segment.
Although Zuckerberg keeps saying that his AI model is open source, it’s practically irrelevant at this point. Now, the curious thing is that Llama 3.2 offers not only chatbots that compete with ChatGPT and Claude but also “compact” versions designed for users’ smartphones.
Meta Wants to Get Into Your Smartphone With Llama
Llama 3.2 will have a huge 405 B model (405 billion parameters) and two more conventional ones (90 B and 11 B). Meta will also offer two lightweight options with 1 billion and 3 billion parameters (1 B and 3 B, respectively) that, as I said, it wants to get into users’ smartphones.
The performance of the “light” Llama 3.2 models is promising. It’s not at all clear that they’ll help much.
According to Meta’s internal analysis, the performance of these models is on par with competitors like Gemma 2 2.6 B or Phi-3.5-mini in tasks such as creating summaries, rewriting text, and following instructions.
Meta has worked with Qualcomm and MediaTek on developing the “light” Llama 3.2 models, which means they will run smoothly on both companies’ chips.
It’s clear that Meta wants to get Llama into users’ smartphones. However, the history of one software component should show Zuckerberg’s company how difficult this is.
The web browser.
The “Default” Browser Trap
That’s right. One of the great technology wars was over browsers. The rise of the Internet made Netscape the OpenAI of the late 1990s. Microsoft soon clipped that project’s wings and made Internet Explorer an integral part of Windows, but only for a while.
Then came Firefox, antitrust lawsuits, the famous browser choice screen, and Google Chrome, which crushed all its rivals. It also outperformed Safari, which barely maintains a reduced share (just 9.4%), even on Mac computers.
This happened with desktop browsers. With the advent of smartphones, Google and Apple wanted to impose their products on users. It was worth it to them, of course. Those browsers had search engines built in, and those search engines made a lot of money. With the advent of smartphones, Google and Apple wanted to impose their products on users.
Nowadays, smartphones allow users to install other browsers. However, the default browser, the pre-installed one, is Chrome on Android and Safari on iOS. Only a few people change the default browser, and Google and Apple know that.
Mobile browser market share worldwide. Chrome and Safari are on 9 out of 10 devices.
As a result, two out of three people use Chrome as their mobile browser (66.2%, according to Statcounter Global Stats). In addition, nearly one in four uses Safari (23.26%, same source), which is gaining ground here due to the immense popularity of the iPhone.
Are there other browsers on iOS and Android? Of course, there are. Do you know what their combined share is today? Barely 10%, reduced to just 6% excluding Samsung’s browser, which logically also uses the popularity of its devices to gain (a small) advantage.
Google, Apple, and Microsoft Won’t Make It Easy
The situation with mobile browsers could have implications for the future of AI on PCs and laptops. Moves in recent months confirm this:
- Google is integrating Gemini into the Android experience as much as possible, allowing users to go beyond what they’ve been able to do with Google Assistant.
- Apple’s introduction of Apple Intelligence as an integral part of iOS, iPadOS, and macOS raises the same issue for iPhones, iPads, and Mac computers.
- Microsoft is reaching out to users with Copilot and gradually making this AI model a native part of Windows 11.
The three primary players in the mobile and PC software experience are moving to standardize their AI experiences—as they’ve already done with their browsers—leaving competitors at a distinct disadvantage.
Meta, Anthropic, and even OpenAI face a complicated future. Unless they can develop differentiated features, it will be tough for users to prefer one or the other because “hey, we already have a chatbot preinstalled, and it’s not bad.”
This happened with browsers, and it’s likely to occur with AI platforms that want to get into users’ PCs and smartphones.
Image | Meta
Related | We Still Have No Idea What Constitutes ‘Open Source AI,’ But That’s About to Change
View 0 comments