Has the era of the AI boom already passed?

When generative AI products became accessible to the general public last year, it ignited a frenzy of excitement and trepidation.

People were astonished by the images and text these tools could craft based on a single prompt. Silicon Valley was enthusiastic about this potentially transformative technology, seeing it as a lucrative opportunity after a period of stagnation and the disappointments of cryptocurrency and the metaverse. Alongside the excitement, concerns emerged about the potential consequences of a world transformed by generative AI. There were fears of massive job displacement, the blurring of reality and AI-generated content, and even existential threats to humanity. The immense potential of this technology was both celebrated and feared.

However, several months later, the initial enthusiasm for AI-generated innovations is diminishing. Governments are intensifying efforts to regulate the technology, creators are filing lawsuits over alleged intellectual property and copyright violations, and people are expressing apprehension about privacy infringements facilitated by these products—both actual and perceived. Doubts have also emerged regarding the accuracy of AI-powered chatbots and the extent to which people should rely on them.

Assuming, of course, that people are still using them. Recent reports indicate a decline in consumer interest: The new AI-powered Bing search engine has not made a dent in Google's dominance, ChatGPT is experiencing a decline in user numbers for the first time, and these bots continue to exhibit fundamental errors that undermine their trustworthiness. In certain instances, they may even be less accurate now than they were previously. Is the novelty of this technological feat wearing off?

Generative AI remains a formidable technology that will persist, and chatbots employing this technology are among the most accessible tools for consumers to engage with directly. However, recent reports suggest that as the initial excitement and curiosity wane, public interest in chatbots may not be as robust as previously anticipated.

OpenAI, along with its ChatGPT chatbot, quickly rose to prominence as a prominent generative AI company and tool. This was likely aided by being one of the initial companies to offer tools to the general public, as well as a lucrative partnership with Microsoft worth billions of dollars. This collaboration resulted in Microsoft's significant announcement in February, wherein it integrated a customized chatbot powered by OpenAI's large language model (LLM) into Bing, its web search engine. Microsoft hailed this AI-infused search as the future of web search, combining information from various websites into a single response.

Amid the hype, Bing transformed from a mere punchline to a potential contender in a market almost entirely dominated by Google. Google swiftly introduced its own chatbot called Bard, and Meta, possibly still reeling from its challenging shift toward the metaverse, released not one but two open-source(ish) versions of its large language model. OpenAI licensed ChatGPT to other companies, attracting numerous adopters eager to incorporate it into their products.

However, the complete transformation suggested by the excitement a few months ago might be farther off than anticipated, or may not even materialize at all. According to a recent Wall Street Journal article, the new Bing has failed to captivate consumers, with two different analytics firms reporting Bing's market share as remaining relatively unchanged from the pre-AI days of January. (Microsoft disputed these figures, claiming they underestimated the numbers but did not share internal data.) Statcounter data indicated a modest increase in Microsoft's web browser, Edge, likely due to the need to use it to access Bing Chat. However, this increase was not substantial and is already receding, while Chrome's market share grew during the same period. Microsoft does still have potential avenues for growth. If Bing Chat becomes more accessible through different, more popular browsers, it could see increased usage. Microsoft has expressed intentions to make this possible.

Meanwhile, OpenAI's ChatGPT appears to be experiencing a decline as well. The Washington Post reported that, for the first time since its launch last year, traffic to the ChatGPT website decreased by nearly 10 percent in June. Downloads of its iPhone app have also dropped, although OpenAI declined to comment on specific numbers.

In contrast, Google has yet to fully integrate its chatbot into its search services, keeping it separate from the main search page and presenting it as an experimental technology that "may display inaccurate or offensive information." Google did not respond to inquiries about Bard's usage statistics.

Given the challenges some of these chatbots pose, Google's cautious approach might be appropriate. Instances of chatbots veering off-topic have multiplied, from crossing personal boundaries with users to disseminating outright inaccuracies. Biases inherent in technology also persist. While some companies have taken steps to mitigate these issues to a certain extent, the problems seem to be escalating rather than diminishing. The Federal Trade Commission is investigating ChatGPT's inaccuracies. A recent study indicated that OpenAI's GPT-4, the latest iteration of its LLM, experienced notable declines in accuracy in certain areas within a few months, suggesting that the model is evolving or undergoing changes over time, resulting in significant variations in output. Moreover, attempts by media outlets to populate pages with AI-generated content have led to numerous and glaring errors. OpenAI even had to retract its tool for detecting ChatGPT-generated text due to instances of misuse.

In a recent development, eight companies responsible for large language models, including OpenAI, Google, and Meta, brought their models to DEF CON, a major hacker convention. The goal was to subject the models to a "red teaming" stress test, where a diverse group of people would assess their accuracy and safety. This initiative was supported and endorsed by the Biden administration, which has been emphasizing the need for safe development and deployment of AI technology. Arati Prabhakar, President Biden's science adviser and the director of the White House Office of Science and Technology, explained that the event aimed to gauge how well these chatbots can perform and assess their susceptibility to issues.

The stress test aimed to provide the companies with essential data about how and where their models break, obtained from a diverse group of testers likely to explore ways that internal teams might not have anticipated. The companies' participation in this event is a positive sign, although the encouragement from the White House likely played a role.

In the interim, these models and the chatbots built upon them are already in widespread use by hundreds of millions of people, many of whom accept their responses at face value. This is particularly true when users may not even realize that the information comes from a chatbot (for instance, some articles from CNET were authored by bots without clear disclosure). As reports indicate diminishing interest in certain AI-powered tools among the public, it becomes crucial for these technologies to improve if they are to remain relevant. Furthermore, the technology's potential for improvement remains uncertain, as even their developers acknowledge that not all aspects of their inner workings are fully understood.

Generative AI has the potential for impressive accomplishments, and its appeal to Silicon Valley and the broader public is evident. What remains to be seen is whether this technology can transcend its current status as a mere novelty, especially considering its ongoing flaws that suggest it may be suited for that purpose for the time being.