The AI Overhypetrain: Why the Latest Safety Debate Misses the Point

The annual Dreamforce conference has come and gone, leaving behind a trail of high-profile discussions on AI safety and ethics. However, attendees were more focused on confronting a down-to-earth reality: their companies are struggling to keep up with the latest advancements in artificial intelligence.

Salesforce customers and partners reported that older and cheaper AI models are still plenty powerful for everyday sales and customer service work. In fact, they’re often sufficient without needing to upgrade to bleeding-edge frontier models everyone’s so excited about. This raises a crucial question: have we been sold a bill of goods by the AI hype machine?

Nvidia CEO Jensen Huang’s call to “run as fast as you can” in model development might sound like a clarion cry for innovation, but it also serves as a stark reminder that we’re chasing our own tails. The rapid pace of AI progress has left many companies gasping in its wake, struggling to keep up with the sheer volume of new features and capabilities being introduced.

A Reality Check on AI Adoption

Despite all the talk about the transformative power of AI, few companies are actually taking advantage of the latest models. According to attendees, older and cheaper AI models remain the norm in many organizations. This isn’t because they’re not seeing the value in AI – quite the opposite. They’re simply having trouble integrating these cutting-edge technologies into their existing workflows.

Alec Bronston, a senior Salesforce director at Chicago-based retail data company Spins, noted that “it’s already hard enough to keep up.” A potential slowdown would present “a lot of opportunity to just even catch up and get our feet wet.”

The Frontier Model Fallacy

The notion that frontier models are the only ones worth using is a fallacy perpetuated by those who stand to gain from selling these expensive, proprietary technologies. In reality, many companies are finding that older models – or even one generation behind – are still highly effective for their needs.

Tim Sanders, chief innovation officer at software reviewing company G2, pointed out that “the majority of agentic outcomes aren’t driven by frontier capabilities.” Instead, they’re driven by last year’s AI. As he noted, this is not particularly relevant to agentic providers or SaaS companies.

The Token Economy Conundrum

As companies navigate this new landscape, they also face a pressing concern: adapting to the shift towards token-based economies. In this model, users pay based on their use of AI and what they demand by way of outputs, rather than on a subscription basis.

Sanders noted that “this is going to require a fundamental shift in how we think about software delivery.” Companies need to figure out how to play in this new economy while keeping costs under control.

The Real Threat: AI Overpromising

The AI safety debate is an important one, but it’s only half the story. The real threat facing us isn’t the potential for AI to become too powerful – it’s the overpromising and underdelivering that comes with the hype surrounding these technologies.

As we move forward, it’s essential to separate the wheat from the chaff when it comes to AI. We need to stop chasing after the latest and greatest models and focus on what truly works for our businesses. Anything less is a recipe for disaster – and a whole lot of wasted resources.