The AI Daily Brief: Artificial Intelligence News and Analysis - Will AI Replace All Coders Within 5 Years?

Episode Date: July 5, 2023

Stability AI's CEO Emad Mostaque has predicted that there will be no human coders within 5 years. On today's episode, NLW explores this prediction in terms of what evidence there is for and against it. Before that on The Brief, he looks at why ChatGPT had its first decline in traffic ever last month, as well as CNET's push towards AI content.  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/ Quick LinksGet Embed PlayerShare on SocialDownload Audio File

Today on the AI breakdown, we're talking about whether coders are going to be entirely replaced by AI within the next five years. Before that on the brief, for the first time ever, chat GPT's monthly visits went down last month. 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. Hello, friends. Welcome back to another AI breakdown. I hope those of you in America had a great holiday yesterday. And those of you who weren't in America used our day off to kick ass and get a head. head. Now, quickly before we get into today's episode, two quick notes. One, I want to thank Supermanage

AI for sponsoring today's show. They're using AI to improve the one-on-one process, and you can check them out at supermanage.a.i slash breakdown. And two, if you're enjoying the AI breakdown podcast, go check out the YouTube channel. There is a lot of content that tries I might to translate it to audio is just going to be better suited for visual, and you can check it out at YouTube.com slash at symbol, the AI breakdown. With that, let's get to today's episode. Welcome back to the AI breakdown brief, all the AI headline news you need in five minutes or less. We kick off today with a statistic that has some people surprised. Last month, according to data from similar web, chat GPT traffic actually fell.

Ever since launching in November of 2022, chat GPT's traffic has just been up and to the right. The biggest jump came between February and March when chat GPT went from about a billion visits per month to more than 1.5 billion visits. Now, since March, it has significantly leveled off. April's growth was much more modest going up only to around 1.7 billion visits per month, and between April and May, the growth was even less. Now, in June, visits came down to around 1.6 billion per month. So the question, of course, is what's driving this? There was a temptation among some to see it as the ebbing of a hype cycle, and that's certainly possible.

However, another possible explanation has to do with what ChatsyBT is being used for. Francois Chalet from Google put together a chart that showed search interest over time, for ChatGPT on the one hand and Minecraft on the other. The chart shows that towards the middle of May there's a pretty strict divergence. ChatGPT starts heading down while Minecraft starts heading up. Francois says there's an obvious factor that might well underlie both trends, down for ChatGPT, up for Minecraft. Can you guess what it is? As it happens, the answer matters for both the future of AI and the future of education.

Francois goes on, the answer is of course summer break. Producing homework remains the number one application of LLMs, which is, remarkable because the point of homework is to do it yourself. The actual deliverable has zero value to anyone. LLMs are touted as a shortcut to economically valuable deliverables, but the market ends up using them for value-less or negative value deliverables such as homework and content farm filler. Not exclusively, but in very large part. What both homework and content farm filler have in common is that they critically need to sound like they were made by humans. That's the entire point. The content itself is worthless. And that's the fundamental value prop of LLMs, the appearance of human

communication. So holding aside for a second the question of whether producing content or doing homework with LLMs is actually a valuable use case, the argument that this is driven by seasonality is pretty compelling. That said, Joe Wisenthal from Bloomberg, who was the person who originally tweeted this chart, was a little more skeptical. He wrote, people are pointing out that there's a seasonality element, which, okay, students need it less in the summer. But I feel like world-changing products that are six months old don't necessarily produce such visible seasonality. Seasonality is for mature things. Now, for my part, I don't super agree with the way that Joe is looking at this. He's trying to pattern match chat GPT to other previous technologies, but I question whether that's

actually something that we can even do. You have to remember that chat GPT went from zero to 100 million users in five weeks. The previous fastest platform to 100 million users was TikTok, and it took nine months. Huge amounts of the U.S. population, something like half, have heard of it or actually used chat GPT. Those are mature product numbers, even if it's, not a mature product itself. Point being, I do think it's interesting that we have this decline in numbers, but I'm simply not sure that we can draw a lot of conclusions from how other platforms previously worked. Another possible explanation is that there's simply some amount of market saturation happening. ChatGPT may be down in June compared to April and May, but its use is still

astronomical compared to basically any other service ever compared to how old it is. Still, I do think it's a trend worth tracking. There is the possibility that there is a hype cycle ebbing, And I think to some extent, as a content creator, I'm certainly feeling a little bit of that. But then again, hype cycles ebbing is not necessarily a bad thing. And in fact, it's a natural part of any new technology settling into what is going to become its real long-term use cases. Now, staying on the theme of OpenAI for just one more moment, earlier this week they tweeted, we've learned that chat GPT's browse beta can occasionally display content in ways we don't want. E.g. if a user specifically asks for a URL's full text, it may inadvertently fulfill this request.

We are disabling browse while we fix this. We want to do right by content owners. Basically what happened here is that people were figuring out how to use ChatGPT's browse features to get around paywalls. That's exactly the type of thing that publishers and existing establishment companies are terrified of when it comes to AI. And so, of course, OpenAI is racing to close that loophole. Speaking of AI and content, however, there was a bunch of different news around publications

turning to AI to develop a new content strategy going forward. FutureSim is reporting that last one. week, CNET owner Red Ventures held an all-hands meeting where their CEO, Rick Elias, laid out a plan for how AI was going to shape their entire future. In that meeting, Elias said, today is day one of AI in our company. Today is the day that we will look back on, hopefully five to 10 years from now, and realize this was not just an opportunity for us to open up a new revenue source or our new business source, but to truly reinvent everything we do as a company. AI will change everything, and I believe, for the most part, in a good way. Now, this is not the first dust up with AI that

CNET has had. Earlier this year, after publishing an article that was absolutely riddled with errors written by AI, CNET staff decided to unionize, saying that the use of AI, quote, threatens our jobs and reputations. A representative said, quote, we're joining a lot of others in the media who are looking into how to address AI in relation to plagiarism, liability, and the impact to the workforce. Now, how then Red Ventures and CNET are thinking about how they're going to use AI in the wake of both those errors and the unionization is not exactly clear. In a statement that they emailed to Future Sim, a representative said, we believe it will impact every facet of modern work and help humans do uniquely human work better. We're embracing it. We are upskilling our teams. We're exploring how it can

help us better serve our customers and partners, but we're committed to doing it the right way, which, as you can probably tell, doesn't say much of anything at all. Another interesting story along the same lines comes from Gio, who owns Gizmodo, the Root, Courts, and other publications. Daily Beast reporter Corbin Bollies tweeted out, Geo Media Editorial Director Merrill Brown tells staff the company will begin testing AI-produced stories based around lists and data next week. The source was an email from Merrill Brown with the subject line innovating Geo Media. The email starts editorial team. I am writing to advise you of a new content feature that will be added into the mix at Geo Publications.

It shouldn't be a surprise that we've done a significant amount of thinking about artificial intelligence, just as everyone in the media business has been doing of late. We're convinced here that the changes AI will bring to the media, and journalism worlds will be very meaningful, if difficult to predict, with certainty. Now, Brown goes to pains to say that these new trials will not impact human written content, saying, these features aren't replacing work currently being done by writers and editors, and we hope that over time, if we get these forms of content right and produced at scale, AI will, via search and promotion, help us grow our audience.

Now, it might be hard to imagine that investing more in AI-produced content won't have implications for what journalists and writer these publications higher. and staying on the theme of work, the first law in America that regulates AI bias in hiring has taken effect in New York City just today. Basically, the law requires that employers who are using AI and other algorithmic tools to make hiring or promotion decisions have to disclose that they're doing so. These companies are also mandated to undergo annual audits for potential bias in the AI process. And while it seems like people who are concerned about AI bias might be excited about this law,

it's actually been critiqued for not being strong enough. And Winters, who leads AI and human rights at the Electronic Privacy Information Center, said, quote, there's a real concern that good governance tools like audits and impact assessments regarding AI programs become this administrative wand waving in front of your face. A quartz piece cites a federal trade commission complaint against Hireview, who hired independent auditors to investigate bias in its product and then issued a press release saying that those auditors had found no bias risks. However, if you actually looked at the audit apparently, the auditors didn't say that the tool is unbiased,

just that auditors didn't have enough information. Finally, today, one of the coolest AI uses that I've seen for a while, a group of archaeologists and computer scientists has teamed up to create AI software that can translate Cuneiform tablets from ancient Acadian, a dead language, into modern language, and to do so pretty instantaneously. Now, Acadian, the language of Sumeria, is one of the best preserved languages in the form of hundreds of thousands of Cuneiform tablets

that have been unearthed and discovered around the world over the past several centuries. However, the process of actually translating that into something that we understand is immensely difficult. One of the challenges, for example, is that Acadian is polyvalent. That means, as Big Think puts it, cuneiform signs can have several different readings depending on how each one functions in a sentence. Because of that, translating Acadian is actually a two-step process. First, cuneiform signs have to be transliterated.

In other words, they have to rewrite the cuneiform in similar-sounding phonetics of the target language. The second step is taking that transliteration and then translating it into a a modern language. The AI model that they developed has scored extremely well. Both scores Big Think says were above their target baseline and in the range of a high quality translation. In addition, Big Think writes there was a surprising result. The model was able to reproduce the nuances of each test sentences genre. The team writes, quote, in almost every instance, whether the translation is proper or not, the genre is recognizable. There's something quite cool to me about the idea of the most cutting-edge technology we have, helping preserve human knowledge and information from thousands

and thousands of years ago. That's going to do it for today's AI breakdown brief. If you're enjoying it, please share it with someone you think might be interested, 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 problems, 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.a.i slash breakdown today to start making the most of your one-on-ones.

Thanks again to Supermanage for sponsoring the AI Breakdown. One prominent AI entrepreneur has argued that within five years, there will be no human programmers left. Today on the AI breakdown, we're exploring his argument and what evidence there is both for and against it. Welcome back to the AI breakdown. Today, we are talking about the future of coding and development, and specifically we are building off of comments from Stability AI CEO Imod Mostock, who argued on a podcast recently that in five years there will be no human programmers left.

Now, the quote came from the Moonshots and Mindsets podcast with Peter Diamandis, And by way of evidence, Imad pointed to data from GitHub. He pointed to a couple trends. The first, he said, in three months, we overtook Bitcoin and Ethereum and GitHub and developed a popularity there. And second, he said, the data from GitHub suggests that 41% of all code right now is AI generated. Now, Amad is certainly in a position to know something about the trends. Stability AI is, of course, the company behind stable diffusion, which is the open source

image generator that underlies many, many different text-to-image platforms. At the same time, Stability AI has its fingers in a huge, huge array of different AI areas. New York Times columnist Farhad Manjou wrote something similar in an op-ed at the beginning of last month. The piece was titled, It's the end of computer programming as we know it, and I feel fine. In the piece, he remembers being a little kid, waiting for games to load, and reading manuals about basic, and coming to one very thorny question which he could never quite resolve. Manju writes, there seemed to be something fundamentally backward about programming a computer that I just couldn't get over. Wasn't it odd that the machines needed us humans to learn their maddeningly

precise secret languages to get the most out of them? If they're so smart, shouldn't they try to understand what we're saying rather than us learning how to talk to them? Now, Manju writes, that may finally be happening. In the kind of poetic irony, software engineering is looking like one of the fields that could be most thoroughly altered by the rise of artificial intelligence. Over the next few years, AI could transform computer programming from the rarefied, highly compensated occupation into a widely accessible skill that people can easily pick up and use as part of their jobs across a wide variety of fields. Manju points out that the gap between AI coders and human coders is getting smaller.

Last year, he says DeepMind reported in the journal Science that when Alpha Codes programs were evaluated against answers submitted by human participants in coding competitions, its performance approximately corresponds to a novice programmer with a few months to a year of training. Matt Welsh, a former engineer at Google and Apple, who now runs an AI startup, also thinks that programming will become obsolete. In an essay called The End of Programming from earlier this year, he wrote, I believe the conventional idea of writing a program is headed for extinction, and indeed, for all but very specialized applications, most software, as we know it, will be replaced by AI systems that are trained rather than programmed.

In situations where one needs a simple program, those programs will themselves be generated by an AI rather than coded by hand. Now, whether AI replaces coders entirely, it is certainly the case that developers are using AI probably more than any other industry. A recent GitHub report found that 92% of US-based developers are already using AI-powered coding tools in their work. GitHub partnered with Wakefield research to survey 500 US-based developers from companies with a thousand or more employees and found that AI is already solving some of their problems, such as repetitive tasks, including writing boilerplate code. In Balshani, the chief product officer at GitHub said, we found that developers spend most of their time writing code and tests than waiting for the code to be reviewed or for the builds to finish. We also found that

AI powered coding tools enable individual developer productivity and greater team collaboration. That means generative AI helps developers generate greater impact, increase satisfaction, and build more innovative solutions. Shani pointed out the developers who used AI tools reported 75% more fulfillment with their work and said that they're writing code more than 55% faster. Perhaps this study was evidence of the idea that AI tools won't replace developers but will simply augment them? Another similar study from coding Q&A site stack overflow found that 77% of its developer respondents felt favorably about using AI in their workflow and 70% are already using or planning to use AI coding tools this year. Still, these developers weren't totally trustful of AI coding tools.

5% were highly distrustful. 22% were somewhat distrustful. 31% were undecided. 39% said they somewhat trust these tools, and only 3% said that they highly trust AI coding tools. Now, part of that may be that there's some evidence that AI generated code produces more bugs. Wired recently wrote a piece called the huge power and potential danger of AI generated code. Programming can be faster when algorithms help out, but there is evidence AI coding assistants also make bugs more common. The piece points to a Stanford University study from last year that looked at how the quality of code from developers using an AI assistant compared to code that was created entirely by humans. As Wired writes, the researchers found that programmers getting

AI suggestions tended to include more bugs in their final code, yet those with access to the tool tended to believe their code was more secure. So the risk here seems to be that humans become more confident because of AI tools and miss bugs that they otherwise might have caught. Now, there are certainly some big potential upsides from AI-assisted coding. GitHub researchers estimate that AI-powered coding could add $1.5 trillion to global GDP. At the same time, it seems likely that there's going to be pretty significant employment disruption as well. Yesterday, CNN published a piece called AI is already linked to layoffs in the industry that created it. The article says, A small but growing number of tech firms have cited AI as a reason for laying off workers and

rethinking new hires in recent months, as Silicon Valley races to adapt advances in the technology being developed in its own backyard. For example, they point to Chegg, a public company that disclosed that it was cutting 4% of their workforce or about 80 employees to, quote, better position the company to execute against its AI strategy. IBM said that they expect to pause hiring

for roles they think could be replaced by AI. In April, Dropbox cut about 16% of its workforce around 500 people, saying that their skills didn't match up with the AI future that they were driving towards. And in May, Challenger Gray and Christmas said that 3,900 people were laid off due to AI. It was the first time they had broken

out job cuts based on that factor. All of those cuts came in the tech sector. Still developers who have skills in AI are getting even more rewarded than their peers. Salary data from Comprehensive. I.O. suggests that engineers who specialize in AI or machine learning see a 12% higher average salary than those who don't specialize in those areas. Now zooming out, the Wall Street Journal recently published a piece called What Will AI Do To Your Job? Take a look at what it's already doing to coders. The WSJ piece writes, want to know if artificial intelligence is going to eliminate millions of jobs, the first place to look is the industry that birthed the technology. AI seems set to do to computer programming and possibly other kinds of so-called knowledge work

what automation has done to other jobs, from the factory floor in the warehouse to the checkout aisle and the call center. In those industries, the end result of widespread automation has been the elimination of countless roles, and their replacement with ones that require either relatively little skill and knowledge, or a great deal more, with workers at either end of the spectrum being rewarded accordingly. In other words, software is eating the software industry. One of the places that this might hit hardest is actually in early career developers. The WSJ piece writes, Many experienced developers I spoke with expressed skepticism at the ability of AI coding tools to take over the most essential tasks of programming. That said, those who are already using such systems

think they could eliminate the need for certain tasks that are currently generally handed off to inexperienced in early career programmers. It certainly appears like change is coming. In a move that has made lots and lots of headlines, even if it's just for novelty sake, Harvard's popular introductory computer science 50 course next semester will be using an AI professor. All over the internet, there is content like this piece from ICCE, how coders can survive and thrive in a chat GPT world, or tips for programmers to stay ahead of generative AI. When it comes down to it, I'm not sure exactly what to think. It seems very unlikely to me that in five years, all programmers are going to be replaced by AI. It feels more likely to me that coding will be one of the industries that figures out what it looks like to

transition from a pre-AI paradigm to a post-AI paradigm the fastest. It seems likely that there's going to be fundamental changes in terms of what people do and how much their jobs are to check and debug code versus producing it in the first place. It seems likely that we're going to see a very strict bifurcation between elite coders who have the ability to develop cutting-edge code versus those who are just producing rote code that can be easily replaced by AI. But ultimately, I don't know. I can only say this for now. I'm still planning on getting my four-year-old and my two-year-old into code learning as soon as I can, although what they build and how they do it might look very different than it would have a few years ago. Anyways, guys, that is it for today's AI breakdown.

I appreciate you listening as always, and until next time, peace.