Artificial Intelligence (AI) is omnipresent today, yet its current state is chaotic. While AI can generate text in applications like Notion, it lacks seamless integration across platforms. For instance, to search the web, one needs Microsoft Copilot within Bing, which is essentially GPT-4 Turbo. Recapping a YouTube video requires Gemini Advanced, powered by Gemini 1.5, formerly known as Bard, which integrates with YouTube due to Google’s ownership.
AI today is plagued by intrusive pop-ups and random chatbots in every app, most of which are ineffective. Four key issues contribute to this chaotic state, but there is hope for a solution that doesn’t involve another AI gadget. Understanding why everything is becoming a chatbot is crucial. If an app lacks an AI chatbot, it seems outdated. Google is adding a chatbot to its search app, and Bing has re-entered the search game with a chatbot. New chatbots like Perplexity AI are emerging, and even tools like Photoshop and Shopify now have chatbots. Microsoft plans to replace the Windows key with a Copilot key, which also brings up a chatbot.
The proliferation of chatbots began with ChatGPT, which now has hundreds of millions of users. However, chatbots are often not the best way to interact with computers. For example, planning a trip involves visiting various websites and comparing prices, a task for which chatbots are ill-suited. They provide walls of text without images, maps, or easy ways to compare airfares. A study found that OpenAI’s GPT-4 model succeeded in only six out of a thousand travel itinerary queries.
Despite their limitations, chatbots are everywhere because they are easy to implement. When ChatGPT launched in December 2022, it quickly gained 100 million users, sparking a frenzy among tech companies to integrate AI into their products. The simplest way to do this is by using existing AI models like GPT-4 and creating chatbots. However, research shows that 85% of AI projects fail, and many AI features in apps go unused.
This frenzy is reminiscent of past tech booms, such as the dot-com bubble and the blockchain craze. Companies are now adding AI to everything, from toothbrushes to gaming laptops, often without practical benefits. AI has been around for years, used in Facebook ads and phone battery optimization, but the term “AI” has become synonymous with large language models like GPT.
The dream of creating an omniscient digital assistant, like Jarvis from Iron Man, has captivated the tech world. However, building such a chatbot requires access to vast amounts of personal data, which is impractical due to privacy concerns. Companies like Humane are attempting to create universal chatbots, but they face significant challenges in integrating with various platforms and services.
The solution lies in making AI invisible and integrating it seamlessly into existing user interfaces. For example, customizing trip options in Skyscanner without using a chatbot or using AI to smartly analyze and remind users of important messages. This approach focuses on solving real problems without highlighting the AI aspect, making the technology more user-friendly and effective.
Some optimistic researchers and economists claim that AI has the potential to supercharge growth to 20% a year, a significant bump from the 3.2% growth average the US has experienced since World War II. If that supercharged growth materializes even slightly, that’s an amazing change in society. A Northwestern economist found that the typical American is 100 times richer than in 1870 when growth was stalled at 0%. With growth at 20%, the typical American would be a thousand times richer than they are today in a relatively short time window. This is the promise of exponential technology meeting exponential times, but there’s a lot of things that would have to go right to meet that growth dream, and AI is entering a turbulent phase in the hype cycle.
The head of Amazon Web Services, Adam Selipsky, who has overseen much of the company’s massive investment in generative AI, has made connections to the dot-com bubble of the early 2000s. The internet in the long-term was underhyped, but a slew of companies trying to cash in were gravely overhyped, and those companies led to a bubble that popped, hurting everyone else. We may be at the point where the needle is starting to pierce the balloon with three big examples to investigate.
Stability AI
Imad Mostaque started Stability AI in late 2020, and based on research from a project called latent diffusion, Stability released a model called Stable Diffusion in August 2022. The model used a denoising process in its architecture that enabled unique image generation. The result was a really effective open-source model capable of creating great images and enabling users to train the model on their own dataset. Stability AI raised $101 million at a billion-dollar valuation in 2022, just days after launching Stable Diffusion, buoyed by AI interest and a growing base that featured tens of millions of active users. 2023 promised to be a big year for the company, and Imad had an engaging vision for what AI could do for the world. But if we do it right, then we can really uplift the entirety of humanity, which I think is pretty crazy, pretty awesome, and it’s the first time we’ve ever had the tools to do that. He wanted to create AI models that could help governments, poor and underserved communities, and aspiring entrepreneurs across the globe, and he took Zoom selfies with Microsoft CEOs. Imad was ambitious and creative but also unfocused and unprepared.
Stability has had serious issues across business operations, funding, and talent retention. The company was burning through cash at an alarming rate thanks to the compute needed to power its models and overwhelming research spend. In 2023 alone, Stability spent almost $100 million on cloud services, and the company generated a grand total of $1 million. Unpaid debts started to accelerate. Stability was $1 million short of its AWS bill in July 2023 and didn’t even have a plan to pay the $7 million it owed for its usage in August. They’re kind of lucky Amazon doesn’t run on mafia rules. It also had an outstanding bill of $1 million to Google Cloud and $600,000 to GPU data center CoreWeave. Despite this foundational financial issue, Imad continued to make promises he couldn’t keep. He told prospective customers, including the Singaporean government, that he could deliver custom national AI models within 60 days. All of us are engaged with governments; we’re trying to help them. It’s a bit difficult when they’re just catching up to the internet right now. The timeline was untenable and confused the company’s leading researchers. Stability was unable to develop a model for Singapore, which never became a customer. A viable business model was failing to materialize. They tried to monetize through an API, then a managed service arm, both of which have struggled to gain traction, and the service may have infringed on their contract with AWS.
Imad scrambled to come up with funding that would pay off debts and retain talent. In July 2023, he shared a plan to raise $500 million in cash and $750 million in computing credits from marquee investors like Nvidia, Google, Intel, and the World Bank. However, most of these investments did not materialize, with only a small fraction of the targeted funding being secured. The debts and empty promises led to an even bigger issue. Employees at Stability AI grew increasingly unhappy with the company and its leadership. The whiplash shifts in priorities and resources, often based on Imad’s changing whims, demoralized and infuriated workers. Projects were abruptly reallocated, and researchers were reassigned. One former executive described Imad as the most disorganized leader I have ever worked with in my career. Concerns over the company’s financial situation and Imad’s leadership led to a steady stream of executive departures. Some departed over worries about cash flow and potential legal liabilities, such as Imad’s reportedly lax approach to preventing the generation of sexually abusive imagery.
The final blow came in March 2024 when the star research team behind Stable Diffusion, including lead researcher Robin Rombach, turned in their resignations. Other senior leaders issued an ultimatum to Imad: resign, or they would walk as well. The departure of the researchers, widely viewed as the company’s crown jewels, underscored the depth of dysfunction and discontent within Stability AI under Imad’s leadership. Just days later, Imad announced that he would step down as CEO, claiming he wanted to decentralize power in AI. However, sources say he fought to maintain control until the very end despite mounting pressure to leave. Imad also resigned from the board, which has initiated a search for a permanent replacement. The temporary leaders put in the role have inherited a company in crisis. Stability continues to burn cash much faster than it generates revenue, making only $5.4 million in February 2023 against $8 million in costs. It faces ongoing concerns about making payroll for its 150 remaining employees. Most existentially, Stability has been hit with a trio of copyright lawsuits alleging its AI models were trained on unlicensed art and photography. The litigation could drag on for years and threaten the entire generative AI industry if Stability puts up an insufficient fight. Once flush with over $100 million in funding, Stability is now in a deep hole, needing not just more cash but a viable business model in a fast-moving competitive space. Some employees have expressed optimism that Imad’s exit could make the company more appealing to investors or acquirers, but after his disastrous tenure, any savior will have their work cut out for them to right the ship before the money runs out.
Inflection AI
Inflection AI came out of nowhere to raise $1.5 billion and launch a ChatGPT competitor. At the time, the $1.5 billion was the third-largest round in the space, right behind OpenAI and Anthropic. Now it’s stripped to the bone, and most of its key talent went to Microsoft. This is the story of an unexpected acquisition that technically didn’t happen. Mustafa Suleyman is one of the most influential figures in AI. Much of his mass influence comes from a great book called “The Coming Wave,” which details the convergence of exponential technologies like AI, robotics, biotech, and more, and what effect their fusion will have on the world. Based on his experience forming DeepMind, which eventually sold to Google in 2014 and gave the company an AI boost, Suleyman co-founded Inflection AI in 2022. The company quickly raised over $1.5 billion from big-name investors like Microsoft, Bill Gates, Eric Schmidt, and Nvidia at a valuation of $4 billion. Inflection’s flagship project was Pi, an AI chatbot focused on providing emotional guidance and companionship for users. Pi, a personal intelligence to navigate whatever your challenge is, could be. They also poured money into infrastructure, amassing 22,000 expensive Nvidia H100 GPU chips, which recently just got outdated with the announcement of the new Blackwell chips. Pi wasn’t the original product that Suleyman imagined and pitched at the start of Inflection’s funding. His big idea was creating an AI chief of staff for businesses that could carry out many of the tasks across an organization that require adept staff planning, HR prowess, and a certain level of emotional intelligence. Pi was the first iteration of what that could look like, but its use case diverged quite aggressively from something that would be integrated into a business. It struggled to gain traction with consumers in the crowded AI chatbot market. Then, within a year of its bloated funding round, Suleyman, his co-founder, and many key team members left the company to join Microsoft, and Suleyman would be named the head of their consumer AI. This move was highly unusual. Typically, if a company wants all of the key talent from a company, they’ll buy it and integrate the people or make it a subsidiary. But Microsoft didn’t have to do that and were able to bring on everyone that they wanted. As for Inflection, Microsoft agreed to pay the company $650 million to license the company’s technology as long as they agreed not to sue over Microsoft’s hiring of its top talent.
In conclusion, while the current state of AI is chaotic, there is potential for improvement by focusing on practical applications and seamless integration. As the hype around chatbots subsides, designers and engineers will develop AI in ways that are truly useful, enhancing user experiences without the need for intrusive chatbots.