• Be Datable
  • Posts
  • GPT-5 Exposes the Gap Between AI Nerds and Everyone Else

GPT-5 Exposes the Gap Between AI Nerds and Everyone Else

OpenAI's automatic model switching reveals the massive divide between early adopters and the mainstream users they desperately need

OpenAI rolled out GPT-5 with automatic model switching last week.

The backlash came immediately.

OpenAI doubled GPT-5 rate limits for ChatGPT Plus users in response, partly because users complained about insufficient bandwidth. The AI nerds who track performance differences between Sonnet 3.7, Opus 4, Opus 4.1, GPT-4o, o1-mini, o1, o3, and GPT-5 complained about losing control.

Sam Altman responded within hours, promising transparency about which model answers queries and easier manual triggering options.

This incident reveals something bigger than feature complaints. It exposes the gap between AI's early adopters (5M+ Power Users) and the mainstream users OpenAI needs to reach.

The Real Problem OpenAI Is Solving

OpenAI isn't optimizing for people who know the difference between Sonnet 3.7, Opus 4, Opus 4.1, GPT-4o, o1-mini, and GPT-5. Those users have already adopted AI. They're solving for the millions who haven't touched these tools yet.

Automatic model switching makes sense for someone who wants an answer.

They don't care if GPT-5 or GPT-4 (or any other poorly named model) responds as long as the answer works.

Early adopters care deeply because they've learned which models excel at specific tasks.

One group needs simplicity. The other demands control.

Sam Altman's weekend updates show OpenAI listening to both groups. They're keeping automatic switching while adding transparency about model selection and manual override options.

Altman’s phrase "good per-user customization will take longer" signals they understand that different users need different experiences, but they won’t have that soon.

The capacity challenge Altman mentions matters more than most realize. API traffic doubled in 24 hours, according to his post. They're making tradeoff decisions between serving everyone adequately versus giving power users precisely what they want.

This tension will define AI products for the next several years.

To those freaking out over GPT-5 not meeting your expectations: might I remind you about the famous Louis C.K. bit? It’s one of my favorites.

He talks about getting on a plane where they announce Wi-Fi for the first time. Everyone's amazed. No one even knew this was possible.

The signal goes to space, to satellites, then back to Earth.

Then the Wi-Fi stops working.

"The guy next to me goes 'This is bullshit.'

Like, how quickly the world owes him something he knew existed only 10 seconds ago."

- Louis CK (Youtube Clip)

Please calm down.

We got GPT-5. It processes billions of parameters, understands context across massive conversations, and generates responses faster than humans can read.

Traditional Interfaces will Change Drastically

Eric Schmidt recently said something provocative: user interfaces will go away.

His reasoning is worth diving into as we discuss this Nerds vs. Everyone else in AI.

Why use windows, icons, menus, and pointers invented at Xerox PARC 50 years ago when AI can generate exactly what you need?

Your UX is wimp.

Schmidt's vision sounds radical until you consider dialogue as humanity's original interface. We've spent decades forcing people to learn software interfaces. Click here, drag there, find the menu, remember the shortcut.

Natural language eliminates that cognitive overhead. And if AI can switch from model to model, it can optimize the UX and UI to whatever best fits the consumer’s experience and personal journey.

I wrote about this in my piece on Jakob's Law, where I argued the key reversal: instead of you remembering the interface, the AI remembers you.

Jakob's Law states that users spend most of their time on other sites, so they prefer your site to work the same way. Natural dialogue breaks this principle by returning to an even more fundamental pattern.

Everyone already knows how to talk. No training required.

Aaron Levie offers the counterargument, and it’s worth the read. Enterprises need deterministic workflows for high-stakes operations like payroll, ERP, and customer support. These platforms will become AI-first but maintain consistent interfaces.

Users want predictability when mistakes cost real money.

I think the answer sits between these two extremes. Core business platforms will keep stable interfaces augmented with AI. Everything else moves toward dynamic, conversational experiences.

But this is another reason why GPT-5 (and Grok) already provide automatic model-switching. For interfaces to generate on the fly, they need to have the autonomy to switch between different models automatically.

The Exponential Adoption Curve

Balaji Srinivasan shared a chart showing Waymo's weekly rides in California. Look at it closely.

AI Adoption requires Automation.

From 2022 to early 2024, the line barely moved. Then, around September 2024, it goes vertical. 20,000 rides become 150,000 rides in months.

This pattern reminds me of the automobile's evolution from manual to automatic transmission. Early drivers insisted on stick shifts for control and performance. They could feel the engine, optimize gear changes, and manage fuel efficiency.

The mainstream market just wanted to get from point A to point B without stalling at traffic lights.

Sound familiar?

Today's AI power users demanding manual model selection are the stick shift drivers of artificial intelligence.

They want precise control over which model handles their query. Meanwhile, millions of potential users want answers without thinking about GPT-4o versus GPT-5 versus Claude Opus 4.1.

The automobile industry's shift to automatic transmission didn't happen because manual transmissions failed.

It happened because reaching mass adoption required removing friction. Now, those same automatic systems control our driverless vehicles, and nobody complains about losing the clutch pedal in a Waymo.

Technical users will always demand control, while mainstream users want results. Successful platforms offer both through intelligent defaults with accessible overrides. OpenAI learned this lesson publicly last week.

The next two years will separate companies that understand this shift from those still building for yesterday's patterns.

You are welcome to reminisce about driving a stick shift.

But automatic transmissions for AI are inevitable.

Reply

or to participate.