The AI Daily Brief: Artificial Intelligence News and Analysis - ChatGPT Code Interpreter Is Finally Coming Next Week!

Episode Date: July 7, 2023

After months of waiting, ChatGPT Code Interpreter plugin is coming to all ChatGPT plus users next week. NLW breaks down why people are so excited, and also shares other OpenAI updates including that GPT-4 is now generally available to paying API customers. Before that on The Brief -- Is AI in a bubble? According to analysts at Goldman Sachs, AI isn't overhyped and backs up all the excitement. Also on this Brief, Morgan Stanley says Microsoft heading to $3T market cap, Alibaba launches an image generation AI and more. Today's Sponsor: Supermanage - AI for 1-on-1's - https://supermanage.ai/breakdown 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, we're discussing all of the new announcements from OpenAI, including the fact that Code Interpreter is finally coming next week. Before that on the brief, why Goldman Sachs says AI is not just hype. 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. Welcome back to the AI breakdown brief. All the AI headline news you need in five-ish minutes or less. Today we are starting with a discussion around AI in markets.

There has been a ton of commentary around this, and what you need to understand, if you don't necessarily pay a ton of attention to macroeconomics or the stock market or the economy in general, is that this has been an extremely, extremely weird year. We have been, on the one hand, under the threat or anticipation of a recession for going on the entire year, even farther, back into last year as well. Yet at the same time, stocks have done extremely well this year. One of the biggest drivers of that is, in fact, AI. Alex Kruger, who's a macro analyst, wrote a whole thread earlier this week about why the common interpretation that a recession is coming and that it's going to be bad is wrong.

And one of the points he makes is about artificial intelligence. Alex writes, Markets are forward-looking and the AI revolution is real. The world is undergoing an AI revolution comparable to the internet evolution or the industrial revolution. Nvidia's earnings have just skyrocketed. Alex goes on. Analysts estimate AI could replace up to 25% of current employment in developed countries, while generative AI could raise annual productivity growth by 1.5 percentage points over 10 years,

and this productivity boost could eventually increase annual global GDP by 7%. Is an AI bubble forming, he asks? Likely so, and it's just getting started. Now, with AI-related stocks doing so well, this question around whether this is a bubble, another tech bubble in the long line of tech bubbles, has been front and center for many market participants and analysts. Goldman Sachs analysts have recently argued that the AI hype is real, or put differently, that AI is more than hype. They wrote, this feels very different from previous tech bubbles.

Cash Rangan, a software analyst at Goldman Sachs, said in an interview in their latest top of mind publication, quote, AI probably isn't in a hype cycle. When a unanimous verdict exists among the technology providers that a technological shift is actually happening, it's real. And when customers start to become interested, it's not hype, and customers are interested. Now, Fortune tries to put a little bit of meat on that bone, pointing to a recent survey from Upwork Research of 1,400 U.S. business leaders that found that 73% of C-suite execs are currently actively using generative AI in their organizations. The CFO of AT&T told Fortune last month that the company had already saved, quote, hundreds of millions of dollars by using AI to improve employee efficiency, and they're

even in the process of rolling out their own generative AI tool called Ask AT&T. T&T. Fortune also pointed to a June Bank of America research survey that found that 59% of consumers had already used chat GPT. Another point that Goldman Sachs made is that unlike previous tech hype cycles, which were dominated by startups, instead the AI boom was not just startups, but in fact being led by some of the biggest companies on the planet. This rang and said, quote, makes it less likely to fizzle out or take a long time to get going. We're having discussions with the CIOs of global corporations who are amazed that the productivity benefits this technology could bring if deployed internally. and all of this is occurring at a time when the market is rewarding productivity gains. So this doesn't feel like a hype cycle. Now, as someone who has been around a lot of hype cycles, going back to Web 2 and social and mobile

and then, of course, crypto and Web 3, I personally do think that there is something categorically different about AI. That's not to say that, one, startup valuations won't be way out ahead of where they are, or two, that startups that probably shouldn't get funding will, just because they've slapped AI in their pitch deck, or three, that all of the uses that we're imagining for AI right now will actually come to fruition. But what I do know is that this is being driven not by some media frenzy, but by the simple fact that when people use chat GPT or mid-journey or one of a handful of these other tools, the world divides itself cleanly to them into before this tool and after

this tool. Yes, we might see excessive valuations. Yes, we might see market dislocations, but this is a real phenomenon, and it's not going away. Now, related to all that, a different banking giant Morgan Stanley says that Microsoft will ride generative AI to a $3 trillion valuation. Analysts have set their price target for the company's shares at $415. That implies a valuation of about $3.1 trillion. In that same report, the analysts named Microsoft their top pick among large-cap software companies and said that they're in the best position to benefit from AI.

They wrote, Generative AI looks to significantly expand the scope of business process is available to be automated by software. Microsoft stands best positioned in software to monetize that expansion. What's more, Morgan Stanley points out, even though Microsoft has seen a 42% share price increase this year, they argue that the valuation is still reasonable based on historic price to earnings multiples. Now, if Microsoft did become valued at higher than $3 trillion, it would be only the second company to do so, with Apple making history last month as the first company to hit that milestone. Now, moving on to something that we covered early. Earlier in this week, the AI content experiment is not going all of that well.

Recently, companies like CNET and Gizmodo's parent company announced that they would be going more all in on AI-generated content, and almost immediately people started to notice very error-filled articles hitting their publications. Earlier this week, Variety published a piece called Gizmodo's I-09 published an AI-generated Star Wars article that was filled with errors. They write, The AI-generated story was headlined to chronological list. of Star Wars movies and TV shows.

Among other issues, the article presented the titles in a numbered list that is not actually in chronological order. This led to the deputy editor at I-09 and Gizmodo tweeting, as you may have seen today, an AI-generated article appeared on I-O-9. I was informed approximately 10 minutes beforehand, and no one at I-09 played a part in its editing or publication. I'm not totally sure how saying that no one on your editorial team had a hand in the editing of a piece that went up on your publication is a good idea, but hey, here we are.

In another area of AI content farming, people are noticing just how much Amazon's Kindle Unlimited has been totally flooded by AI-generated books. Author Caitlin Lynch wrote, The AI bots have broken Amazon. Take a look at the bestsellers in teen and young adult contemporary romance e-books top 100 chart. I can see 19 actual legit books. The rest are AI nonsense clearly there to click farm. Amazon, what are you doing about it?

Tech Radar writes, self-publishing such as via Amazon's Kindle Direct program, has become a way for many genuine authors to bring their work to the public and build a following without the help of a large publisher. Because these self-publishing capabilities are purposely easy to sign up for, it seems anyone can generate endless AI written books and upload them to be sold on Amazon's e-book store and make them available for reading via Kindle Unlimited. Now apparently after that tweet from Caitlin Lynch, people noticed that many of those AI books had vanished from the bestseller lists, presumably removed by Amazon. However, the books themselves were still available for purchase. TechRadar says,

The mass uploading of AI-generated books could be used to facilitate click-farming, where bots click through a book automatically generating royalties from Amazon Kindle Unlimited, which pays authors out by the amount of pages that are read in an e-book. So it doesn't matter that these books disappear. The people running such a scheme could just upload as many as they like to replace the removed ones. This is not the only place that I've seen this complaint. Another platform having this issue is Etsy. This led to the Atlantic publishing a piece last month called AI-generated junk is flooding Etsy,

Coloring books, stickers, mugs, and t-shirts are being pumped out by AI-assisted hustlers. According to the amateur online business advertisers of YouTube, the age of easily accessible AI is the age of asking and receiving. ChatGPT and other AI tools are ascended in popular culture, as is the idea that you can ask them for anything. You can even ask them to make you rich. And it is certainly the case that if you go look through YouTube, Etsy is the platform that tends to get the most coverage when it comes to how AI generated products can help people with their side hustle. So it's clear that this AI-generated product and content problem is one that platforms are just going to have to deal with in the years going forward. Lastly, today, China's AI frenzy continues with Alibaba announcing a new mid-jurney-Dali stable diffusion-style text to image generator. Presenting at the World Artificial Intelligence Conference in Shanghai,

Alibaba Cloud presented a new image generator that will initially be available to their enterprise customers. Now, individually, this isn't a surprising piece of news, but it just serves to reinforce how, much the race for new AI models and products is a global one. That's going to do it for today's AI Breakdown Brief. If you're enjoying it, do me a favor and go check out the AI breakdown newsletter. You can get it at the AI breakdown.bhiveb-Behivebheheheh-I-I-V-com, and I'll be back soon with the main AI breakdown. Hey guys, before we dive into the main part of the episode, I want to share a little bit about today's sponsor, Supermanage. A truly great one-on-one should be about celebrating wins, solving

and deepening the connection between two human beings. But what if you miss those wins, never heard about those problems, and spent your whole meeting avoiding the hard stuff? That's where SuperManage comes in. SuperManage AI distills your public Slack channels into a one-on-one brief that highlights everything you need to know to jump right in. Because, let's face it, you want your team to do the best work of their lives. And that starts with world-class conversations.

Visit supermanage.aI slash breakdown today to start making the most of your one-on-ones. Thanks again to Supermanage for sponsoring the AI breakdown. Welcome back to the AI breakdown. Today, we are going through all of OpenAI's recent announcements, and boy, have there been quite a few. And although we've covered it in the show, I want to make note of the fact that they kicked this week of announcements off with their super alignment announcement. Now, if you haven't watched my video or listen to my podcast about this topic, long story

short, OpenAI has announced a new initiative that is focused on alignment with super intelligent AI systems. The Big Banner headlines around this announcement are one, they're putting extremely senior members of their team to lead it, including a co-founder. Two, they're putting 20% of the computing power that they've secured to this effort, despite the fact that they have said explicitly that lack of access to all the compute that they want has led them to change their feature roadmap for 2023. And three, that their goal is to solve super intelligent alignment. within four years. Now, if you want the whole story, obviously go check out the other show, but what captured people about this was the fact that OpenAI was so clear about the challenges,

and they gave themselves specific goals that could be held against them, basically. In other words, they didn't have to say that they wanted to solve this in four years, but they did, and so now, at every point between here and then, people are going to say, well, how much progress are we making towards that four-year goal? They also put a specific number around the resources that they're going to dedicate to this efforts, which is, again, something that people can determine if they're actually living up to their promises. The last piece that I want to note is in the context of the other press announcements this week. This was the first thing they announced this week, and I think that that was very intentional. There's a pretty good chance that if they had announced their

GPT4 developer access first, it would have slightly overwhelmed this news, and it is almost certainly the case that if they had led with code interpreter, that would be what everyone was focused on. It strikes me as pretty clear that they wanted to give this super alignment team announcement its own time in the sun. But as I said, it was certainly not the only announcement this week, and the next big one has to do with what is available through the OpenAI API. TLDR, all paying API customers now have access to GPT4. OpenAI says that since March, when GPT4 was released, millions of devs have requested access

to its API. Today they write, all existing API developers with a history of successful payments. can access the GPT4 API with 8K context. We plan to open up access to new developers by the end of this month, then start raising rate limits after that depending on compute availability. Now, while the GPT4 announcement was the big one, they also announced that they'd be making the GPT 3.5 Turbo models, Dali models, and Whisper APIs also generally available.

Lastly, they said, we're working on safely enabling fine-tuning for GPD4 and GPT3.5 Turbo and expect this feature to be available later this year. Now, for some, it was this little detail in that last sentence that was most exciting. Matt Schumer, the CEO at Hyper Right AI, says, GPD4 fine-tuning is coming later this year. This is a bigger deal than most people realize. A fine-tuned GPT4 will be able to do tasks that GPT4 can barely touch today. Potentially more important, it'll enable far better reliability on tasks GPT4 can do unreliably today.

Lastly, for devs, they also announced a number of deprecations, as well as a new community resource where people can ask questions about model and endpoint deprecations as well. Still, all of this was prelude to what was the biggest bit of news, which is that Code Interpreter will be made available to all ChatGPT Plus users over the next week. Now, in their tweet, OpenAI says it lets ChatGPT run code, optionally with access to files you've uploaded. You can ask ChatGPT to analyze data, create charts, edit files, perform math, etc. Users opt-in via settings. Code Interpreter was first revealed as part of the plugins announcement. When OpenAI shared that they were opening up ChatGPT for third-party plugins, there were two that they were working on in-house.

The first was a browsing feature, which later became Brows with Bing, and the second was Code Interpreter. Even though only a very small number of people had access to it, within days those who did were showing off some pretty incredible use cases. David Boyle, for example, shared a spreadsheet of music data with ChatGPT, from which it segmented a number of different music markets and came up with business strategies for each segment. Code interpreter could be used to visualize data. For example, Professor Ethan Malik uploaded a set of data around all the lighthouses in the U.S. and then had code interpreter create a gif of them twinkling. Another viral example of using code interpreter for visualization came from John Bacchus, who wrote on April 29th,

the code interpreter feature on chat chit is the most mind-blowing thing I've seen yet. All I did was upload a CSV of SF crime data and ask it to visualize trends. Some of the things that code interpreter came back with were a number of incidents by day of the week, hourly crime trends in terms of when crimes took place specifically, top crime categories, and a heat map of crime hotspots in San Francisco. Finally, one of the things that people have found is that you can just ask Code Interpreter to come up with its own interesting hypotheses based on data. In other words, you don't even have to tell it exactly what you want.

You can just feed it some data and tell it you want something interesting. On May 2nd, Professor Ethan Malik wrote a piece on his blog called it starting to get strange. In it, he describes Code Interpreter as GPT4 with three new capabilities. The AI can read files you upload up to 100 megabytes, it can let you download files, and it lets the AI run its own Python code. This may not seem like a huge advance he says, but in practice it is pretty stunning. And it works incredibly well without any technical knowledge or ability. Describing more of this program that builds programs, as he puts it, he describes a number

of his early experiments. One example prompt was, quote, I'm writing a blog post about how amazing chat GPT is at working with code right now. I would like you to create the perfect illustration, a GIF using Python that represents this ability. Decide what an appropriate amazing GIF would be and then figure out how to create it and let me download it. Sure enough, a few minutes later, he had his GIF. He also uploaded Excel files without providing any context and ask things like, can you do visualizations and descriptive analyses to help me understand the data? Can you try regressions and look for patterns? Can you run regression

diagnostics? He even uploaded the entire US Census data set and asked the eye to explore it, generate its own hypothesis based on it, conduct hypothesis tests, and write a paper based on the results. Malik said that it tested three different hypotheses with progression analysis, found that one was supported, and ultimately wrote an academic paper about it called Regional Dynamics of Industry Characteristics, a comprehensive examination of payroll, employment, and establishments across metropolitan and micropolitan areas. Malik said it is not a stunning paper, but it took just a few seconds, and it was completely solid. Now, here we are a couple months later, and Ethan has written a follow-up post called What AI Can Do With a Toolbox, getting started with code interpreter. Ethan writes,

everyone is about to get access to the single most useful, interesting mode of AI I have ever used, chat GPT with Code Interpreter. The way that it describes Code Interpreter now is saying, it gives the AI a general purpose toolbox to solve problems by writing code in Python, a large memory to work with, and integrates that toolbox into the AI in ways that play to the strength of large language models. This, he says, helps address a number of problems that previous ChatGPT had. Some of those problems include allowing AI to do math, lowering hallucination rates, making the AI more versatile, and one more that's much more subjective.

Malik writes, it gives you more of those AI moments. Anyone who has worked with GPD4 has probably encountered at least a few moments where it felt like there really was indeed a ghost to the machine. I know it is an illusion that LLMs are in no way sentient or thinking, but those moments are a thrilling and sometimes unnerving glimpses of possible futures with smarter AIs.

Code interpreter provides the most that's weird moments per use of any AI system I have played with. Concluding, he says that Code Interpreter is a sign of things to come. He says code interpreter is the strongest case yet for a future where AI is a valuable companion for sophisticated knowledge work. Things that took me weak to master in my PhD were completed in seconds by the AI, and there were generally fewer errors than I would expect from a human analyst. Human supervision is still vital, but I would not do a data project without Code

interpreter at this point. But it is just as clear to me that humans are not going to be replaced by Code Interpreter. Instead, the AI does what we always hope automation will do, free us from the most annoying, repetitive parts of our job, so we can focus on the good stuff. simplifying the process of analysis, I can do more and deeper and more satisfying work. My time becomes more valuable, not less, as I can concentrate on what is important rather than the rote.

Code interpreter represents the clearest positive vision so far of what AIs can mean for work. Disruption, yes, but disruption that leads to better, more meaningful work. I think it is important for all of us to think about how we can take the same approach to other jobs that will be impacted by AI. I think it's fair to say at this point that Code Interpreter is the most anticipated product that ChatGPT has ever had. For months, we've been getting these little dribbles of what seem like amazing use cases. And I know many people have like me even gone out and looked for alternatives to use in the meantime because of seeing what's possible and wanting to be able to do that in our own context and work.

We are going to have a lot more time to dig into code interpreter when it becomes available. Obviously, I will be doing some experiments and sharing them with you here. But for now, this was a pretty exciting announcement and one which I am thrilled to see play out next week. Thanks for hanging out and watching or listening to the AI breakdown. Until next time, peace.