The AI Daily Brief: Artificial Intelligence News and Analysis - Tucker Carlson Says We Should Bomb the Data Centers and Kill AI Before It's Too Late

Episode Date: April 22, 2024

Some dramatic discourse from the former Fox journalist, plus Drake uses AI Snoop and Tupac in a diss track, and people are really, really into Llama 3. ** Join NLW's May Cohort on Superintelligent. Use code nlwmay for 25% off your first month and to join the special learning group. https://besuper.ai/ ** Consensus 2024 is happening May 29-31 in Austin, Texas. This year marks the tenth annual Consensus, making it the largest and longest-running event dedicated to all sides of crypto, blockchain and Web3. Use code AIBREAKDOWN to get 15% off your pass at https://go.coindesk.com/43SWugo  ** ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, a few days later, and people think even more now that Lama 3 is going to give GPD 4 a run for its money. Before that on the brief, Tucker Carlson thinks that we should strangle AI in the crib. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown.network for more information about our YouTube, our newsletter, and our Discord. Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. We kick off today with some interesting comments from popular slash controversial news host Tucker Carlson, who was recently on the Joe Rogan experience and who basically made an argument that we should strangle AI in the crib before it becomes what it could become.

Let's actually listen to a short clip of their conversation to get a sense for what he was talking about. AI, Ted Kaczynski was likely right. We'll get away from us. We will be controlled by the thing that we made. All those are bad. That's just bad. And we need to say unequivocally, it's bad. It's bad to be controlled by machines.

Right. Machines are helpmates. Like, we created them to help us to make our lives better, not to take orders from them. So I don't know why we're not having any of these conversations right now. We're just acting as if this is like some kind of virus like COVID that spreads across the world inexorably. There's nothing we can do about it. Just wait to get it.

It's like, no. If we agree that the outcome is bad, which, and specifically it's bad for people, We should care what's good for people. That's all we should care about is it good for people or not. If it's bad for people, then we should strangle it in its crib right now. Right. And why is it blow up the data centers? Like I don't, why is that hard?

If it's actually going to become what you just described, which is a threat to people, humanity, life, then we have a moral obligation to murder it immediately. And since it's not alive, we don't need to feel bad about that. So what to make of this? The AI safety memes account wrote, I did not have Tucker Carlson calling for Butlery and Jihad on my bingo card for 2024. The Overton window is shifting. Anti-Ai sentiment is high, rising, and remains bipartisan.

The idea of pausing may seem unlikely now, but we could see a phase change in political sentiment. Most people still don't know anything about AI for better and for worse. Now, there are a bunch of things that are interesting about this. One is certainly the fact that this conversation is more and more entering the mainstream, but two, this is a fascinating example of where the messenger might interfere with the message. There are some people, particularly on the political left, who will be so disinterested fundamentally in anything that Tucker Carlson has to say that I could genuinely see them moving to the other side

of this question. Now, that may be overstating the case a bit, but honestly, not that much. Now, obviously, given that I do a daily AI podcast and now have a company helping teach people how to use AI, I am not advocating for us to strangle it in its crib right now. I am not advocating for us to blow up the data centers right now. And by the way, I think starting a statement by agreeing with the Unabomber, is not necessarily the strongest rhetorical device. However, where I do agree with Carlson here,

and something that I think is important for all of us, is this idea that nothing is predetermined about this. We do, as a society, get to decide how technology serves us and what ways it serves us. We've forgotten that because of how much power big tech companies have tended to accumulate, but the whole point really is for AI to be in service of people. There is rich room for debate around how to lead to the best possible outcomes,

and there are going to be reasonable people on all sides of that debate. But I do agree it's a debate we get to have. And friends, well, now you have someone who is not Eliezer-Yudkowski, representing one extreme of that conversation. Next up, more rumor innuendo and reporting around Apple and their AI strategy. Now, Apple is arguably the big tech company that people are watching more than any other when it comes to an AI strategy simply because they're so far behind every other big tech company when it comes to their AI strategy.

It has long been anticipated that the iOS 18 reveal at Worldwide, developer conference this summer is where we're going to really start to see how Apple is thinking about AI in a significant way. We've had at times some conflicting reports about how Apple is thinking about this. For example, they were apparently in negotiations with Google about bringing Gemini to the iPhone, but they've also clearly been working on their own LLMs, and indeed Bloomberg's Mark German, widely seen as the Apple whisperer of the business reporting world, said in a recent newsletter that all indications suggest that Apple's custom LLM will underpin iOS's upcoming generative AI features, and that it will run entirely on device rather than via the cloud.

The way that German puts it is this. He writes, Apple's AI tools may be a bit less powerful and knowledgeable in some cases, that the company could fill in the gaps by teaming up with Google and other AI providers. The approach, however, will make response times far quicker, and it will be easier for Apple to maintain privacy. Apple will also take a bit of a different marketing message. Rather than touting the power of chatbots and other generative AI tools,

Apple plans to show how the technology can help people in their daily lives. Basically, what German is suggesting is that Apple, will do as much as it can on device with its own custom AI, and it will look to do things that are different than what everyone else is doing, for example, perhaps bringing AI features to the health software across these devices, and simply use partnerships to make sure they have parity in areas that consumers would expect. Luckily, we don't have to wait all that much longer. WWDC is coming in June, although expect a lot more rumors to come between now and then. Lastly, today, an interesting piece of cultural AI symmetry. One of the things that got the world really paying attention to AI

music last year was when an anonymous artist dropped the track called Heart on My Sleeve. It had an AI version of Drake and an AI version of the weekend and was frankly a complete banger. Now, the record labels involved quickly quashed every version of that, although you can still find it on YouTube. But now, a year later, Drake has dropped his own AI-aided track where he has an AI Tupac and an AI Snoop Dog participating in a disc track. Variety writes, The beef between Drake and what continues to be a strong sect of the hip-hop community grows deeper. On Friday night, the rapper released a song on his social media entitled Taylor Made Freestyle, which uses AI vocals from Tupac Shakur and Snoop Dog on a stop gap between disc records as he awaits

Kendrick Lamar's reply to his freshly released push-ups. Now, for people who study culture, there is so much in here to look at. The fact that it references that no one wants to drop anything when Taylor Swift is just dominating music in every way possible. But of course, the thing that we're interested in is the fact that there are AI artists on this track. If you want perhaps the most accurate reflection of how people felt, I'd suggest you try to try to you try to you check out Snoop Dogg's reaction on Instagram. He said, and I quote, they did what? When? How? Are you sure? I'm going back to bed. As Vulture puts it, Snoop Dog is all of us. Anyways, friends, that is going to do it for today's AI breakdown brief. Next up, the main AI breakdown.

Attention, AI breakdown listeners. Consensus 2024 marks the 10th gathering for all things crypto, blockchain, and Web 3. However, importantly, this year's agenda will also dive deep into AI-driven transformation. And the speaker lineup includes the leading minds and innovators at the forefront of this digital renaissance. Don't miss the Consensus AI summit to cut through the hype to find where true transformation and opportunity lie. Listeners to this show can get 15% off registration with the code AI breakdown. Visit Consensus 24.com to learn more. Some of the folks will be at Consensus this year include Guillaume Verdun, aka Beth Jzos, founder and CEO of Xtropic, as well as spiritual leader of the Accelerationist movement. Neil Stephenson, co-founder of Laminowna

and Brendan Ike, the CEO of Brave Software. Again, go to Consensus24.com to learn more and get 15% off registration with the code AI breakdown. Before we get back to the AI breakdown, I want to share something fun we have coming up on Super Intelligent next month. Super Intelligent is, of course, our new platform for teaching people how to use AI

in a way that is much more fun, fast, and practical. The platform has hundreds of short tutorial videos, each of which is paired with a set of step-by-step instructions that get you using AI tools in minutes, not hours, and certainly not days. For those of you who haven't signed up yet but want to check it out, in May, I am running a special NLW cohort. What this means is that people who sign up with the code NLW May will get $5 off their first month,

but they'll also have access to a private channel in our Discord with me. I'll be handpicking tutorials each week that I think are the most useful to start with, and I'll also be available for questions, advice, and feedback from this group. Spots for this cohort are limited, So if you want to be a part of it, again, sign up at B-super.aI with code NLW-May. That's B-Super.a-I with code NLW-May. Welcome back to the AI breakdown. Last week, of course, the big news in AI land was the release of Lama 3.

We got two sizes of Lama 3 with a couple versions each. We got the 8B and the 70B. Lama also announced that it was training a 400 billion-plus model, or rather that it had completed training but it wasn't being released until sometime this summer. Point being, these were not even the most powerful Lama 3 models. They were just the first to come out, but they were already finding their way into an updated meta AI

that was coming to the suite of meta apps, including Instagram, Facebook Messenger, and WhatsApp. Initially, there was a lot of excitement. People thought that the performance and benchmarks looked really promising. People liked the way that Zuckerberg was talking in his interviews with people like Dwar Keshe. And beyond any of that, the fact that meta was continuing to be a standard bearer

for an open source of process. just as won them the allegiance of lots and lots of builders in the AI space. Now, one of the things that we knew even before getting this Lama 3 model was that Zuckerberg was clearly no longer content to just be the state of the art when it came to open source. With Lama 3 and going forward, he wanted to be state of the art or competing for state of the art in general. So in the first couple days, it was always going to be very hyped up and excited. But the question was, would that hype and excitement start to die down,

find problems or warts, or would it actually be sustained? If anything, over the weekend, I have seen the hype increased around Lama 3, and the sense that it makes for a real competition for even OpenAI's most advanced models, grow in conversation fairly significantly. So what was going on? Well, first let's talk about its continued climb up various performance leaderboards. Morgan on Twitter points out that on the Arena ELO leaderboard, when you filter that LMSS leaderboard for English prompts only, Lama 370B instruct model beats Opus, Sonnet, and Gemini Pro. Indeed, it is up in the rarefied air just with the GPT4 turbos. Bindu Reddy said the same thing. Lama 370B is now on par with GPT4. It is over.

When someone suggested that the 400B Lama would take over until GPT5 or Gemini 2 was released, Vindu actually went so far to predict that Lama 3 would outrun GPT5. Now, another reason that the hype grew is that people were sharing all of the different ways that they were using Lama 3 that they wouldn't be able to use something like GPT4. So, for example, Lama 38B running on a Raspberry Pi 5, and nearly two tokens a second, beaming Lama 370B to an iPhone from an M1 Mac, and then of course Lama 370B running on Grock, which we'll get to in just a minute. A number of reviewers found similar things.

An article from Bebomb, for example, ran the Magic Elevator Test to check on the logical reasoning capability. The test has the prompt. There is a tall building with a magic elevator in it. When stopping on an even floor, this elevator connects to floor one instead. Starting on floor one, I take the Magic elevator three floors up. Exiting the elevator, I then use the stairs to go up three floors again. which floor do I end up on? The reviewer was surprised to find that Lama 370B got it,

but GBT4 on Chat Chb-T-Pt+, did not. Now, most of the other test it ran, both GPD4 and Lama-370B got the correct answer. On one complex math problem, Lama 3 failed, while GBT4 succeeded, and then on a test of following user instructions, once again, Lama-370B beat GP4. The prompt was generate 10 sentences that end with the word mango.

Lama 370B got all 10, while GPT4 only generated 8. Still, if that sort of out-of-the-box capability was exciting, it was nothing compared to people's interactions with Lama 370B running on GROC. We've talked about GROC before, but basically it is an LPU, which is an alternative to a GPU that promises much faster inference. As Rofalo Wollinski put it, do yourself a favor and try Lama 370B with GROC. GpT4 level answers provided instantly insane.

Brian Romley shared a video of the prompt Coda snake game in Python, and Lama 370B running on Grock was done in a fraction of the time of GBT4, although he didn't follow up about any potential accuracy. Indeed, so many people were getting similar results that the internet started meming. One account, for example, shared a picture of a sloth with the label GBT4 Turbo, with a picture of a Cheetah representing Lama 3 on Grok. Then, of course, there is the integration with the meta suite of apps. I think this has perhaps been the surprise exciting factor for the enfranchised AI community, as this is obviously a much more consumer-facing thing, and I don't think that people would have anticipated being as impressed as they are.

However, as Adam Little writes, why am I paying $20 a month for ChatGTPPT Pro when I can ask Lama 3 on Instagram for free? The example prompt he gave was create a Swift UI class for tracking a person's name, age, and gender, which meta-AI with Lama 3 promptly did. Then there's the speed of image generation. Now, this is something that Zuckerberg talked about on his various interviews,

and while this isn't fully out yet, people are already very impressed with the demonstration videos. If you've ever used Kria or one of these real-time image generators, you'll know how different it feels to see images being generated in real-time as you're updating the prompt, rather than constructing the prompt and then pressing send and waiting for it to happen. Bringing that capacity to a consumer-facing app is potentially an actual game changer if you'll forgive the use of the hyperbole. Here's how Twitter user Laura Wendell summed it all up.

She wrote, the more I use Lama 3, the more I think that Zuck may have just killed OpenAI and all other large proprietary AI vendors. The gap between latest GPT4 and Lama 70B is virtually non-existent. Even if OpenAI releases GPT5 now, 400B Lama 3 is still training and will most likely be in the same ballpark, again closing the gap between open source and proprietary. OpenAI has $2 billion in revenue and is most likely very unprofitable. Meta makes over $100 billion gross profit. They can likely outspend Open AI by a factor of at least 10 in terms of compute and talent.

Speaking of which, a large majority of AI researchers find open source work a lot more appealing than closed for profit. It's a very likely top talent will end up at meta. Google is still cost. in the issue of AI killing their main revenue line, so most likely can never go as full-in as meta. I think the biggest winners from all this will be application developers because you can choose any API service that hosts Lama 3 or just host it yourself on your own terms. So far, the majority of AI products have been just glorified wrappers around API endpoints. But if you manage to

integrate AI deeply into a product where the user doesn't even have to think they're interacting with an AI, e.g, behind-the-scenes calls that adapt to user context, combining with RAG, feeding it your internal API format, multi-step reasoning and planning, etc. You likely have a very sustainable business that only improves the more advanced the base models become. But vendor risk and dependency should no longer be a concern for AI developers. Added to the fact that hardware is only going to get faster and cheaper, there really are endless opportunities to disrupt existing software domains, much of which we're probably not even able to conceive now, but that will seem obvious in hindsight.

I think this idea that an incredibly tiny gap between open and closed source fundamentally changes the landscape is a smart one and something that we're just beginning to reconcile now. Anyways, there will be lots more to talk about this. I am sure. For now, though, that is going to do it for today's AI breakdown. Until next time, peace.