TL;DR:
- Generative AI tools like ChatGPT are reshaping the internet and have exploded in popularity.
- They can perform tasks beyond imagination, but the actual impact may be less dramatic than the hype suggests.
- Generative AI is seen as a highly significant development, potentially as impactful as smartphones or the web.
- LLMs, such as ChatGPT, are large language models that produce human-like text outputs.
- While there are concerns about LLMs replacing jobs, they can also be used programmatically, offering cost-saving opportunities.
- Content creation and customer services are expected to be significantly impacted by LLMs.
- Productivity gains in certain sectors are likely, but the broader impact on the overall economy remains uncertain.
- LLMs have the potential to transform software development and address the shortage of software developers.
- Policymakers should consider nurturing the capabilities of LLMs through industrial and education policies.
- It is important to approach the economic implications of LLMs with a balanced perspective, considering both potential and limitations.
Main AI News:
Generative artificial intelligence (AI) tools have taken the internet by storm, revolutionizing the way we interact with technology. The potential impact of these systems, such as ChatGPT, is a topic of much debate. Will they replace countless jobs or lead to significant productivity gains? As with any technological advancement, the reality is likely to be more tempered than the lofty expectations.
AI systems today, like ChatGPT, DALL-E2, and OpenArt, possess capabilities that were unimaginable just a few years ago. From proofreading and fact-checking to crafting poetry and producing art, these generative AI systems can accomplish tasks that were once reserved for human creativity.
The popularity of these generative AI systems has skyrocketed in recent months. For instance, ChatGPT alone boasts over 100 million users, and OpenAI’s website receives approximately one billion visitors per month, making it the fastest growing consumer app in history, according to Swiss bank UBS.
While technology trends like web3 and the metaverse have garnered attention, their real-world impact remains uncertain. Generative AI, however, appears to be a game-changer.
Experts in the field widely agree that generative AI is a highly significant development. Some even argue that it is as impactful as the smartphone, the web, or even electricity itself. There are those who go further, warning that it could be a precursor to machine super-intelligence, presenting an existential risk to humanity that demands urgent mitigation.
This may come as a surprise, given that AI systems have been part of our daily lives for over a decade. Unbeknownst to many, AI systems have been ranking our social media feeds, curating digital ads, and suggesting movies to watch on streaming services, using their pattern recognition abilities to analyze historical data.
So, what sets generative AI apart? To borrow the words of technology analyst Benedict Evans, the novelty lies in running the pattern-matching machine in reverse. Instead of identifying existing examples that fit a pattern, generative AI “generates” new examples of the pattern. The output can take the form of text, images, audio, or video, resulting in an unprecedented proliferation of digital content.
But what are large language models (LLMs) exactly? ChatGPT, one of the prominent generative AI systems, exemplifies an LLM. These models produce text by predicting the next word in a sequence based on vast amounts of training data from books and the web. They can generate articles, essays, code, stories, poems, and more that are often indistinguishable from human-authored content.
OpenAI’s LLMs, particularly the GPT-3.5 model powering ChatGPT (with the even more powerful GPT-4 available to premium subscribers), have gained widespread recognition. Other notable LLMs include Google’s LaMDA, Meta’s LLaMA, and Anthropic’s Claude. Microsoft has also integrated GPT-3.5 into its search engine Bing, investing over $10 billion in OpenAI. Start-ups are also capitalizing on LLMs, developing specialized use-case applications like copywriting and legal contract drafting.
But can LLMs truly replace human workers? While users may initially marvel at the ingenuity of ChatGPT, they often find themselves frustrated by its limitations. LLMs, by design, do not function as search engines providing factual accuracy. Instead, GPT-3.5’s probabilistic approach occasionally leads to what are called “hallucinations”—the presentation of false information, non-existent URLs, or fabricated references that are distressingly plausible. Another inconvenience is the need to copy-and-paste ChatGPT’s outputs into other software for further use. These factors contribute to skepticism when considering the notion of LLMs replacing 300 million jobs.
However, LLMs can be leveraged programmatically through application programming interfaces (APIs). For instance, in an experiment, Ankur Shah and his team built a database of information about UK insurance products. They wrote a prompt for an insurance product review article and developed a system that populated the prompt with product data, sent it to OpenAI’s API, and seamlessly pushed the LLM’s output into a web content management system. This method allowed them to publish hundreds of review articles in less than an hour at the cost of around $7, a stark contrast to the several months and $70,000 it would have taken with human freelance writers. By including real product data in the prompts, they circumvented the LLM’s tendency to produce incorrect or imaginary details.
When used correctly with well-crafted prompts, LLMs like ChatGPT could indeed transform how certain jobs are performed. In light of these practical experiences, it is plausible that LLMs will disrupt fields such as content marketing, where subject matter expertise and an individual writing style are less critical than in journalism or fiction.
Customer services is another domain that stands to benefit. Chatbots powered by LLMs, trained on domain-specific data, outperform their predecessors. For instance, compare a GPT-4-powered bot like Intercom’s Fin with Aviva’s traditional online assistant. These advanced bots can even be connected to speech APIs, potentially automating contact centers entirely. This translates into substantial cost reductions for organizations and a seamless customer experience with no queues for web-chat or telephone support, as AI systems can handle hundreds of interactions simultaneously. However, for front-line customer service workers, this prospect may be less ideal.
One key advantage of LLMs lies in their potential to boost productivity. Sectors such as financial services, telecoms, media, and education are likely to witness efficiency gains and cost savings. However, it is important to note that higher sectoral productivity does not automatically guarantee overall productivity improvements across the entire economy.
Despite the widespread adoption of digital technologies in the previous era, productivity growth has stagnated since the global financial crisis. The reasons behind this puzzle are economists. It is possible that we have been living in an unproductive bubble or that organizations have not fully harnessed the potential of mobile apps and big data analytics to significantly enhance productivity.
Hence, it would be unwise to assume that cost savings enabled by LLMs will automatically translate into productivity gains for the entire economy. As Nobel laureate Robert Solow famously noted in 1987, “The computer age is everywhere except in productivity statistics.” The same might hold true for AI.
However, Diane Coyle, one of the Economics Observatory’s lead editors, argues that the true value of LLMs and other generative AI systems lies in changing how things are produced—similar to the transformative impact of assembly lines in the 1910s or just-in-time production in the 1980s.
In this context, LLMs’ ability to write computer code may prove to be the most crucial aspect. The UK, facing a shortage of software developers, could benefit immensely from LLMs in two ways.
First, LLMs can enhance the productivity of existing developers, narrowing the skills gap. GitHub Co-Pilot, an LLM-powered tool often referred to as “autocomplete for code,” already boasts over a million users and enables developers to write software up to 55% faster than before.
Secondly, and more significantly, a large number of individuals with minimal or no coding experience could leverage LLMs to build software. Traditionally, the availability of workers skilled in programming languages like Python, PHP, or JavaScript has determined the limits of what computer systems can be built. However, in the future, the crucial factor may be the capacity to imagine the functionality of a system and specify it in natural language.
If LLMs do transform software development, nurturing this capability through industrial and education policy would be a wise move for policymakers.
When analyzing the economic implications of new technologies, it is crucial to navigate through the hype and “criti-hype” surrounding them. Technology executives often emphasize the science fiction-inspired risks of artificial general intelligence to heighten the perceived value of existing products like chatbots. Similarly, academia, think tanks, and the media are incentivized to offer opinions on the social and ethical implications of generative AI, often leaning toward alarmism.
The reality, however, tends to be more mundane. The economic impact of LLMs will initially be felt in content creation and customer services before dramatically transforming software development, potentially benefiting productivity at a broader level.
As we grapple with the consequences of LLMs, it is essential to approach the topic with a balanced perspective, grounded in both the potential and the practical limitations of these powerful AI tools.
Conclusion:
The widespread adoption of generative AI tools, specifically large language models like ChatGPT, holds substantial economic implications. Content creation and customer services will experience significant disruptions, while productivity gains may be sector-specific rather than economy-wide.
However, the ability of LLMs to transform software development and address skill shortages presents opportunities for increased efficiency. Policymakers should consider supporting the growth of LLM capabilities through strategic industrial and education policies. It is crucial to maintain a balanced outlook, acknowledging both the potential benefits and practical limitations of generative AI in order to navigate this rapidly evolving market.
