📈 Data to start your week
Exponential View Azeem Azhar
New research says AI models are getting faster at learning on the fly, roughly doubling every few months. Meanwhile execs are quietly backing off doom predictions about AI killing jobs, and ChatGPT just lost its majority market share.
Based on reporting by Exponential View, Azeem Azhar — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
There's a quiet data dump making the rounds this week, and it's more interesting than the usual AI headline churn. ByteDance researchers looked at how quickly successive generations of AI models pick up new skills on the job, not in training, but during actual deployment. Their finding: models released just three months apart are learning at roughly double the speed of their predecessors. That's a scaling curve nobody was explicitly chasing, and it suggests the improvements aren't just about bigger parameter counts or more tokens crammed into pretraining. Something about how these systems adapt in real time is compounding fast.
At the same time, the people who run companies are singing a different tune than they were eighteen months ago. In January 2025, 46% of CEOs said they expected AI to force significant headcount reductions at their firms. By May 2026, that number had dropped to 20%. That's not a small shift. It's more than half the executive class walking back predictions that, at the time, sounded almost inevitable. Whether that's because the automation hasn't materialized as fast as feared, or because leaders got spooked by backlash and started softening their language in public, is genuinely unclear. Probably some of both.
And then there's the ChatGPT number, which is the one that'll get the most attention despite being the least dramatic in isolation. OpenAI's flagship product dipped below 50% share of the AI chatbot audience for the first time in March. That's still a commanding position, but it marks the end of an era where ChatGPT was basically synonymous with the category. Gemini, Claude, and a growing list of regional and specialized alternatives have been chipping away steadily rather than dramatically, and the crossing point just happened to land in March.
Taken together, these three signals don't point in one tidy direction, and that's sort of the point. Models are getting sharper faster than expected. Executives are less certain about the jobs story than they were telling shareholders a year ago. And the market that seemed like a foregone conquest for one company is fragmenting in real time. None of this settles the bigger arguments about AI's trajectory, but it does suggest the confident narratives from early 2025 were built on thinner ice than anyone admitted.
My take — AI-written commentary, not fact-checked reporting
The CEO reversal is the tell here, not the scaling law. Executives don't walk back headcount predictions because they suddenly feel generous, they do it because the automation curve turned out messier and slower than the slide decks promised. I'd bet the ChatGPT share drop is the more durable trend of the three, because moats built on being first rarely survive once the tech becomes commoditized, and that's exactly what's happening across the open and closed model race right now.
Read more about this at: Exponential View