AI Just Quietly Had One of It's Most Important Weeks in Recent History

0:00On today's AI Breakdown, we cover the most important stories from the past week, from big company announcements, to remarkable new products and research, to an incredible defection and a warning that I think makes this one of the most important. important weeks in recent AI history. And stay tuned at the end of this episode for a sample of the new single I'm Fine by Angel Baby, West Ghost, and AI Grimes. Let's start by talking about some of the new tools as well as product upgrades, and we have to kick it off with Midjourney version 5.1. This version came out at the beginning of the week and has had people delighted as it's more, as they call it, opinionated, bringing back some of the artistry from version 4, but also combining it with even greater photorealism than version 5.

0:47Lots of people are experimenting with this. There's a raw mode as well if you want it to be a little bit less opinionated. If you go just search mid-journey version 5.1 on Twitter, v5.1, you'll see how many amazing photos are out there. It's really unleashing creativity in a very cool way. Next, Inflection, one of the big buzzy companies in AI that has super well-known founders like Reid Hoffman from LinkedIn, announced their first product, which is Pi. Now, Pi stands for personal intelligence, and this is meant to be much more like her than ChatGPT. It's really meant to be something that learns about the person that's interacting with it, has more emotional conversations.

1:27I said earlier in the week that it's kind of more like the Call Annie app that I profiled on this channel than it is like ChatGPT. And at first, some people were kind of surprised because it didn't seem to perform that well on tasks, But it really isn't meant to be your coding helper or help you write ad copy for your marketing job or something like that. It's meant to be a very different slant on artificial intelligence. And when people started to grok that, they've had a much better experience, at least it seems so far. Next up, I have to shout out iBaby AGI. It's sort of like an auto GPT or baby AGI that lives on your iPhone.

2:02It just came out. It's from developer Nate Chan. And it's really quite an impressive product. I've been experimenting with it all week. I've had some success doing basic research with it. I've had some interesting success when it comes to planning an adventure in New York City. It's really a cool demonstration of what early instances of AutoGPT can do. Even if very specialized and specific, there is a lot of promise there. And the interface being on your iPhone is great. So go check out iBaby AGI on your iPhone. Now, of course, it wouldn't be a week in AI without some crazy new research. And the one that everyone has been talking about recently is a new AI method that converts brain signals to video.

2:44So you can see here their comparison of true video to predicted video. And the method that they're using to do this, they call SEBRA. They say SEBRA is a machine learning method that can be used to compress time series in a way that reveals otherwise hidden structures in the variability of the data. It excels on behavioral and neural data recorded simultaneously, and it can decode activity from the visual cortex of the mouse brain to reconstruct a viewed video. If you want to go learn more about Sebra, go to cebra.ai. Moving into the big companies in the space, we saw tons of companies move in.

3:18I didn't even put together a slide for Slack GPT, but Slack did their version of GPT. But a big one that many people did take note of was a set of updates to Bing Chat. Specifically, Microsoft has announced Bing Chat plugins, which follows OpenAI's ChatGPT plugins. It's also got multimodal answers. It's got persistent chat and chat history that follows you around the browsing experience. And most importantly for people who actually want to use it, there is no more waitlist. This also wasn't the only news from Microsoft. Earlier in the week, Bloomberg reported that Microsoft was working with AMD on their expansion into AI processors, which is a project that is codenamed Athena.

3:55Now, the goal of this would be to offer an alternative to NVIDIA, which absolutely dominates the markets for AI-ready chips. And while Microsoft has acknowledged Athena, it says that AMD is not involved in the project. Still, that didn't stop AMD's stock from rocketing up after this report. Not all big companies' experiences with AI this week were particularly good, though. Samsung has banned generative AI after a security breach from last month, in which they were concerned that sensitive information had made it to chat GPT. The ban isn't just for ChatGPT, it's for all generative AI tools, and you have to think that things like this are exactly why OpenAI announced that ChatGPT business, which was private by default, was going to be available in the next few months.

4:37Meta, meanwhile, warned that they are seeing tons of malicious imposters of these generative AI tools, such as ChatGPT imposters. If you've been on Facebook or Instagram in the last month, you've certainly seen ads for some use chat GPT on your phone type third party thing. And it's very difficult to tell what's real, what's not. Apparently, this is now a huge vector for scams. So everyone should be careful out there. Now, in the company disruption sphere, one of the big pieces of news was that IBM announced that 7 ,800 jobs that they were planning to hire for, they will no longer be hiring for because they believe they can automate them away with AI.

5:13This involves a number of back office roles, human resource roles, things that aren't customer facing. This, of course, really changes the nature of the conversation around how fast AI is going to disrupt the workplace in different industries. You still have folks on Twitter saying that it's unlikely to be as fast as people fear. But then again, you have IBM saying that about 8000 people that were going to get jobs are no longer going to. Meanwhile, education company Chegg saw an even more dramatic impact this week from ChatGPT. On their earnings call for the last quarter, they announced that ChatGPT was hurting their business by reducing signups.

5:48And you can see here on this chart that Chegg's stock absolutely cratered. It was off something like 50%. And on top of that, other education stocks like Pearson were down as well. Now, concerns around AI's disruption to business were certainly on the agenda for the biggest political meeting around AI this week, which was a group of AI-related CEOs from companies like Microsoft, OpenAI, and Anthropic, who all met with Vice President Kamala Harris, and Joe Biden stopped by as well for this photo op. And in addition to just the FaceTime, the White House also announced a couple of different initiatives.

6:20They announced$140 million for funding of AI research, and they also announced a new initiative by which these companies would open up their models for a process of public review. Of course, the U.S. is hardly the only sovereign who's getting more and more involved with AI. For example, OpenAI has been welcomed back to Italy after addressing a number of concerns around the GDPR regulations as well as age verification. And at the G7 last meeting, AI was one of the most important topics on the agenda. While there wasn't necessarily a lot of support for the idea of an AI pause, at least not right now, there was a sense that there needed to be a concerted and coordinated effort to figure out what the guardrails for AI really would be.

6:59Now, in some ways, I think that the biggest news of the week was the story about Jeffrey Hinton. Jeffrey Hinton is one of the fathers of AI. He's been working on neural networks for 40 or 50 years at this point. He's a Turing Award winner, and he's been at Google for the last 10 years after selling his company to them. He's just left Google and has spent the week subsequently warning about the dangers of AI. In his first post-Google interview that came out in the New York Times, he talked about how the arms race between all these companies was fundamentally changing the nature of the AI safety conversation.

7:31All of a sudden, you didn't have companies acting as he had seen Google act in the past as a good steward for this technology, trying to proceed with minimal harm, but as companies that had business objectives that were threatened and had to move full steam ahead regardless of the potential risks to society. Hinton has also talked quite a bit about how AI has surprised him and many others in terms of the speed with which it's evolved and the capacities that it already has. To get a sense of just how clear the warning is, Hinton says, I think it's quite conceivable that humanity is just a passing phase in the evolution of intelligence.

8:03Why this is so significant is that, frankly, the AI safety conversation is far, far behind the AI excitement conversation, the AI tools conversation. I'm sure I'm not alone in the fact that when I, as a content creator, create content around AI safety issues or AI ethics, it performs far less well than how to use ChatGPT for marketing. Now, on the one hand, this is completely reasonable, right? Individuals are trying to figure out how this technology is going to impact their lives. I don't fault anyone for being more focused on how it can change their job prospects or how they can get out ahead of it than on some of these big existential questions.

8:37But at the same time, as AI cascades right now, it is more important than ever to have exactly these sort of existential conversations. Hinton is doing something that folks like Eliezer and Paul Cristiano haven't been able to do, which is put this conversation into the mainstream media. Now, how fast that burns out or whether anything comes of it is a totally different question, but that's why I said that this week was quietly potentially one of the most important in recent AI history. And just to round that conversation out, one of the other really interesting things that happened this week that also came from Google was that a Google researcher had their memo leaked about how that company was getting out-competed by open source AI.

9:18The banner quote that everyone ran with was, we have no moat and neither does open AI. Now, I did a whole thread summing up their argument, but effectively, they said that in the last month or so, ever since the leaking of Meta's llama model, open source developers and individuals have been able to actually have access to the most powerful tools and the innovations that they're producing are astonishing to people at these big companies like Google and Meta. Now, this particular author is mostly concerned with what it means for Google from a business perspective, but there is, of course, a safety dimension to the open source conversation and AI as well.

9:48On the one hand, you have folks like Jeffrey Ladish here who writes, maximally open source development of AGI is one of the worst possible paths we could take. It's like a nuclear weapon in every household, a bioweapon production facility in every high school lab, chemical weapons too cheap to meter, but somehow worse than all of these combined. On the other hand, you have folks like Rune, the founder of MakerDAO, who says, if you want to use over-the-top analogies, then it's more like giving nuclear weapons to people after the megacorps have been building and pointing nukes at us for years already.

10:17Mutually assured destruction prevents nukes from launching. Balance of power is necessary to prevent domination. On a deep instinctual level, I'm more inclined towards the open source is a good thing argument, but I do think that when it comes to AI, everything that you thought and all your priors have to be re-examined constantly. Now to leave you on one really cool note, if you haven't seen it yet, go on Twitter and search Code Interpreter. This is the new plugin from OpenAI for ChatGPT that's one of their proprietary plugins, and it is doing amazing things with data analysis and visualization.

10:49Now right now it's in alpha and only a very small number of people have access to it. But if you go check out either my thread or other people's threads, you'll see things like incredible analysis on San Francisco crime data, an academic paper written just by feeding it census information, basic video editing in one case, and even a business strategy analysis that took music market data and turned it into a company strategy. Again, the name of this is Code Interpreter. It's super cool. And to the extent that this was an important week for discussing AI safety, I find myself constantly wanting to hold both the incredible excitement of things like Code Interpreter and what it might mean in terms of the democratization of data analysis and visualization on the one hand, with these X-risk and really important society questions on the other.

11:34Of course, in the middle, there's all this disruption around jobs and industries that is going to be so sweeping and profound. And I just kind of think you have to deal with it on all of these levels all at once. For now, that's the story that I'm seeing from here. I appreciate you watching the AI Breakdown this week. If you are interested in getting this via email, go to AIBreakdown.Beehive.com. Go check out the podcast version or subscribe here on the YouTube. Until next time, peace.

12:15Having conversations, but there's no one there. Just me and my tax side. I'm not scared. Trying to convince myself that I don't care. What's something I feel like? Living in a world full of robots. What a time to be alive. All my friends like AI. Signing out from this life. Yeah, I'm off like dying inside. But on the outside, I'm fine. In a world full of robots. What a time. Going dummy, going dust. Still, I'm out of sight. I've been running out of space. on a hard drive Dying in seven on the other side on five minutes and I'm gonna think it's like I don't know what it is you