Tech Brew Ride Home - Fri. 04/19 – All The News About Llama 3

Episode Date: April 19, 2024

Meta released Llama 3 yesterday, and some of the moves they made have made be consider if Zuck could win these first AI wars. Is Zuck also getting aggressive in VR? Why Apple had to take down the WhatsApp and Threads apps in China. And, of course, the weekend longreads suggestions. Links: Meta’s battle with ChatGPT begins now (The Verge) Meta, in Its Biggest A.I. Push, Places Smart Assistants Across Its Apps (NYTimes) Meta gives the Quest 2 its second permanent price cut in four months (Engadget) Cool or creepy? Microsoft's VASA-1 is a new AI model that turns photos into 'talking faces' (Tom's Guide) China Orders Apple to Remove Popular Messaging Apps (WSJ) Weekend Longreads Suggestions: How Perfectly Can Reality Be Simulated? (The New Yorker) The Life and Death of Hollywood (Harper's Magazine) How Neopets’ nostalgic revival tripled users in six months (The Guardian) Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to the tech meme right home for Friday, April 19th, 2020. I'm Brian McCullough today. Meta released Lama 3 yesterday, and some of the moves they made have made me consider

if Zuck could win these first AI wars. Is Zuck also getting aggressive in VR? Why Apple had to take down the WhatsApp and Threads apps in China, and of course the weekend long-range suggestions. Here's what you missed today in the world of tech. Big AI news for meta yesterday, but I'm going to break it up a bit. First, meta-detailed its new flagship open-source AI model Lama 3. Comes in 8 billion and 70 billion parameter models to different flavors,

and those are available now. They are also right now powering the meta-AI chatbot. But also interesting, they teased an upcoming model trained on 15 trillion tokens with more than 400 billion parameters. Lama 2 was trained on 2 trillion tokens, so the leap to 15 trillion is quite somewhat. something. Quoting the verge. A key focus for Lama 3 was meaningfully decreasing its false refusals or the number of times a model says it can't answer a prompt that is actually harmless.

An example Zuckerberg offers is asking it to make a killer margarita. Another is one I gave him during an interview last year when the earliest version of META AI wouldn't tell me how to break up with someone. Meta has yet to make the final call on whether to open source the 400 billion parameter version of Lama 3 since it's still being trained. Zuckerberg downplays the possibility of it not being open source for safety reasons. Quote, I don't think that anything at the level of what we or others in the field are working on in the next year is really in the ballpark of those type of risks, he says, so I believe that we will be able to open source it, end quote.

Before the most advanced version of Lama 3 comes out, Zuckerberg says to expect more iterative updates to the smaller models like longer context windows and more multimodality. He's coy on exactly how that multimodality will work, though it sounds like generating video akin to Open AISORA isn't in the cards yet. Meta wants its assistant to become more personalized, and that could mean eventually being able to generate images in your own likeness, end quote. Now, we mentioned that Lama 3 is powering the meta AI chatbot, and Meta is bringing that to Instagram, to WhatsApp, to Messenger, to Facebook in more than 12 countries. And when I say

bringing it to them, I mean it will be right there in the search bar in those apps. It'll be right there in the main Facebook feed. Quoting the New York Times. The AI software will become practically omnipresent inside the news feed, in search bars, and in chats with friends. People will be able to ask the assistant meta AI for help in completing tasks and getting information, such as what concerts might be occurring in San Francisco on a Saturday night or the best options for vegan enchiladas in New York, end quote. And if you want to see something right now that will wow you, go to AI.net.com.

it to draw a picture. Draw a picture of a beagle. A beagle riding a motorcycle, wearing goggles in the style of a 1950s black and white movie. As you are typing those words, step by step, auto-complete style, the picture changes almost instantly to conform to your prompts. If you go back in your sentence and change beagle to giraffe, the picture changes instantly too. If you're familiar with other image-generating models that can take half a minute or more every time you prompt it to do an image, you'll be wowed by the wizardry involved here. Okay, so on that last bit, the bit about meta incorporating Lama directly into their family of apps, lots of people are like, who cares?

What is the use of having an AI bot inside of, say, Instagram? Well, whisper it quietly, but I think this move is bold enough to suggest that Mark Zuckerberg is maybe the best positioned to win this first round of the AI wars. Why? Well, originally, I said that Meta was in a good place because they had access to a ton of data to train on that others don't. OpenAI can scan the web, can scrape YouTube videos, Google can do this too, so can others. But they cannot train their models on Instagram or 20 years of Facebook posts. Zuckerberg can. But also, the bigger advantage Zuck has here is distribution. Remember, meta's family of apps has three billion users. OpenAI has what, 100 million users of chat GPT? I'm going to invoke the, will my mother use this rubric here? Is my mother or hundreds of millions of other normies likely to sign up for a chat GPT account or a mistral account or anything like that to begin testing if AI is useful in their lives? Probably not, not yet at least. But what they will do is they will try AI out if it's right

there in front of them in their Facebook app. How did Zuck win over Snapchat? How is he battling TikTok? Sometimes by creating standalone apps, but mostly just by turning on stories and reels and all the other copycat features inside of the apps that people already use. It's a powerful lever. Not everybody was going to sign up for an Instagram account, but he got a lot of people to use stories on Facebook anyway because it just showed up. How many millions of people could potentially be educated about using AI and maybe becoming habituated to it inside a meta app they already use. If my mom can plan her next vacation inside of Instagram, well, then what does she need to try out some other new company if it just does largely the same thing? All of these companies integrating

AI and chatbots into Excel, into email clients and enterprise clients, the idea is just to make AI a part of the computing you already do. Not some new company or startup or new compute paradigm. There's been a lot of talk about how the big tech incumbents seem to be the early winners of the AI race. With meta, can you see the angle of how this could be that on steroids? Let's just take one narrow use case. For years, Zuck has dreamed of people doing their search inside an app like Facebook instead of Google. But search, you kind of still need the open web, as we've discussed with startups like the Ark browser. There's a new kind of search now with AI where you kind of don't need the open web at all. Well, the bots do, but you don't because you never see it.

It's all summarized for you. The web itself is abstracted away. Don't you think it's Mark Zuckerberg's fever dream to have the web abstracted away? For a lot of people, the internet is their experience on Facebook, on Instagram, on WhatsApp. Now, imagine they never have to leave a Zuckerberg production basically ever. A day when, after getting used to doing search and stuff, via his AI, you have your meta-AI assistant, do your shopping for you. Zuck doesn't need to make money on this AI stuff right now or for a very long time. If AI takes off inside his apps and people start spending more time inside his apps and not elsewhere, even incrementally at the scale of 3 billion people, he wins big time eventually. And with the whole Metaverse experiment, he's proven a willingness to

burn billions of dollars playing the long game. Open AI and the start- startups need to either hope their API calls, power the AI revolution in the background, or that they can get especially enterprises to sign up for subscription products. Google needs to hope they can train people to search using their AI in such a way as to not kill Google search overnight. Microsoft is sort of like meta in the sense that they hope AI happens inside of apps people already use and can just charge them extra money for that. Apple hopes you can do AI stuff on device to the degree that that is top of the funnel stuff for AI usage. But what if Meta and Zuckerberg suddenly have the leverage over the major hardware platforms that he's always dreamed of because of their scale?

Forget Asian-style super apps where you live your whole life inside the app. If Facebook or Instagram or Messenger, powered by AI, allow you to do everything you'd want to do on the web or on your phone inside of that app, then the hardware doesn't matter. The web doesn't matter. All that would matter would be Zuck's apps as the delivery method for that AI future. Speaking of Zuck and his long game, what do you think he's seen in the sales numbers of the Apple Vision Pro that prompted this? And I mean that either way. Either the Vision Pro is selling more than he thought or selling less. But meta has cut the price of the base 128-gigabyte Quest 2 VR headset from $250 to $199. the second permanent price cut in just four months after a drop from $299 to $2.49 in January.

Quoting in Gadget. Of course, the price cut comes in the wake of the company rolling out the far more powerful and capable Meta Quest 3, which starts at a far higher price of $4.99. The discount suggests that Meta is trying to get rid of its remaining inventory of the previous model. Still, the Quest 2 is our pick for the best cheap VR headset and the drop to $199 makes the device even more appealing. For a couple hundred bucks, you'll get a headset that has fast-switching LCD displays with a resolution of 1832 by 1920 per I and a 90-hertz refresh rate. It has a solid library of VR apps at this point, along with access to cloud gaming services like Xbox Cloud Gaming. However you slice it, the Quest 2 is a solid entry point into VR. In addition, Meta recently lowered the prices of Quest 2 accessories, end quote. One more new AI model here real quick. Microsoft researchers have introduced VASA 1, an AI model that creates a realistic talking head video from a portrait photo and an audio file. This has been released into research preview only, quoting Tom's guide. Vasa 1 takes a single portrait photo and an audio file and converts it into a hyper-realistic talking face video complete with lip sync, realistic facial features, and head movement. The model is currently only a research preview.

view and not available to anyone outside of the Microsoft research team to try, but the demo videos look impressive. Similar lip sync and head movement technology is already available from runway and Nvidia, but this seems to be of a much higher quality and realism reducing mouth artifacts. This approach to audio-driven animation is also similar to a recent vlogger AI model from Google Research. Microsoft says this is a new framework for the creation of lifelike talking faces and specifically for the purpose of animating virtual characters. All of the people in the examples were synthetic made using Dali, but if it can animate a realistic AI image, it can animate a real photo. In the demo, we see people talking as if they were being filmed with slightly jerky,

but otherwise natural-looking movement. The lip sync is very impressive with natural movement and artifacts around the top and bottom of the mouth seen in other tools. One of the most impressive things about Vasa 1 seems to be the fact that it doesn't require a face-forward portrait style image to make it work, end quote. Now, lots of people are rightfully concerned about the implications of this. Rightfully so, as I say, because just one photo and you could have absolutely anybody saying anything. That's why Microsoft is keeping this for researchers only at this point. But think of the lowly podcaster. If I could upload a photo of myself and the audio of the show each day, I could create a TikTok video of every segment of this show every day. Don't you imagine

I would want to do that? And real quick, Apple has removed WhatsApp and threads from its app store in China, saying it was ordered to do so by China's cyberspace officials, citing national security concerns, quoting the journal. The cyberspace administration of China asked Apple to remove WhatsApp and threads from the app store because both contain political content that includes problematic mentions of the Chinese president, according to a person familiar with the matter. And Apple spokesperson said that wasn't part of the reasoning. The move shrinks the number of foreign chat apps Chinese internet users can use to communicate with those outside of the country, a further tightening of internet controls by Beijing, which is sensitive to uncensored information circulating.

Tech tensions between the U.S. and China are already running high. Congress is fast-tracking a bipartisan effort to crack down on TikTok that could lead to passage of a law this month, forcing its Chinese parent to sell the popular video-sharing app in the U.S. or face a ban. Collectively, Instagram, X, Facebook, YouTube, and WhatsApp have been downloaded from Apple's App Store more than 170 million times in China over the past decade, according to estimates by market research firm Censor Tower. Apps such as X were key to disseminating information and videos of protests against COVID rules in China that erupted in late 2022, end quote. Time for the weekend long read suggestions. First up, I could not recommend this piece

from the New Yorker highly enough. It's called How Perfectly Can Reality Be Simulated? It has everything. Basically, the history of video games graphics, the history of game engines, the history of VFX, filmmaking techniques in Hollywood, thrown in for good measure, a history of epic games and Tim Sweeney, and even right at the top, how designers capture digital assets from the real world to put into games and graphics. Do they go out like Google does for Google Maps into the world and scan and photograph everything? You better believe it. One of the best articles I've read in ages. Conversely, this long piece from Harper's makes the provocative case that Hollywood itself, as we know, as an industry might be on the cusp of an extinction-level event.

Quote, the film and TV industry is now controlled by only four major companies, and it is shot through with incentives to devalue the actual production of film and television. What is to be done? The most direct solution would be government intervention. If it wanted to, a presidential administration could enforce existing antitrust law, break up the conglomerates, and begin to pull entertainment companies loose from asset management firms. It could regulate the use of financial tools. As D. Ward has suggested, it could rein in private equity. The government could also increase competition directly by funding more public film and television. It could establish a universal basic income for artists and writers.

None of this is likely to happen, end quote. And finally from The Guardian, a history of neopets, combined with the news that neopets is apparently having a resurgence. Quote, Neopets was launched in 1999 as an early internet universe where users could care for a variety of virtual pets and play mini-games. A crude social network that predated Facebook, the site allowed users to add friends, message one another, trade resources, or virtual currency, and battle. It peaked at more than 25 million active users in the mid-2000s before seeing a steep decline in popularity as competing gaming and social sites exploded across the internet. The platform has changed hands a number of times since its inception and amid the acquisitions, it languished. Reporting just 100,000 users when it was purchased

by the Chinese firm NetDragon 2017. Deepening its decay was the decline of Adobe's flash player, the software used to power much of the site, which browsers began to phase out around 2017 before the company officially killed it off in 2020. Neopets fell by the wayside, receiving little attention or updates to its interface. That was until the entrepreneur and investing consultant Dominic Law, who joined NetDragon in 2020 as director of new markets, launched an internal campaign to revive the Neopets brand, which he called a major leap of faith. You're going to hear an interview I did soon with a major figure in the tech industry who said Neopets was their entree into computing and even programming. No bonus episodes for you this weekend, but I've got something exciting to roll out next week,

something I've been working on from the beginning of the year. Can't wait to tell you about it. Talk to you on Monday.