Former Tesla AI director and OpenAI founding member Andrej Karpathy said on X that more than 99% of people paying attention to AI today only came into the field within the past year.
In the post, he wrote that this leaves “AI dinosaurs” confused — people who were already in AI before 2026, and even more so those who entered before 2012. According to ABMedia, the post drew more than 2,300 likes and nearly 130,000 views.
A reply to a viral “we already did this” comment
Karpathy made the remark while responding to Elvis Saravia, founder of the AI research community DAIR.AI, whose X handle is @omarsar0.
Earlier, Karpathy had shared ways to better understand AI outputs, including controlled English, diagrams, web pages, and explainer videos. ABMedia said that Oct. 2 post received more than 50,000 likes.
Saravia wrote, “50K likes? Most of what Karpathy shared, we were doing over a year ago.” He added that the exchange opened up an interesting question: whether there is a general interface or language that could let humans communicate more smoothly with AI agents.
Karpathy’s response offered an explanation for why such posts still spread widely. For many readers, these ideas do not feel old because they only started using or following AI recently.
Experience gaps are shaping the discussion
ABMedia said Karpathy’s view points to a problem that often goes unnoticed: confusion in AI discourse is driven in large part by differences in experience, not only by differences in opinion.
For long-time AI researchers, prompt engineering, structured model outputs, and visual methods for understanding model behavior are already familiar practices. For a large wave of newer users, those same methods still feel new. What veterans see as obvious can look novel to people who just arrived, leaving both sides with very different baselines for what counts as new.
A rapid influx of new users
The article also cited user growth figures. OpenAI said at DevDay that ChatGPT had reached 1.2 billion total users. Meta’s AI agent Muse, meanwhile, recorded 1.8 million downloads in its first 12 days after launch.
That pace of adoption has expanded public discussion around AI very quickly and pushed it beyond specialist circles just as fast.
“99%” was presented as intuition, not data
ABMedia stressed that Karpathy’s “99%” was not backed by data. It described the figure as his personal mental model, and said he used that wording himself.
The article also noted that a larger share of newcomers does not make newcomer discussions less valuable. Many practical AI applications and workflows have been explored by users who are just getting started and are not constrained by established frameworks.
That framing can also be read as a form of veteran arrogance. Saravia’s complaint carried a clear “we already knew this” tone, while Karpathy’s lighter reply still served as a reminder to experienced people in AI: when the audience changes, the way ideas are communicated has to change too.
What this means for media, companies, and product design
If Karpathy’s observation is broadly right, ABMedia said media coverage of AI, internal AI rollouts at companies, and AI interface design should all start from the assumption that the audience is made up of beginners rather than experts with deep technical context.
What looks self-evident to veterans is often exactly the information new users need most. That also helps explain why content that does not look new inside the AI community can still go viral on social platforms.
ABMedia said that while most users are still in the learning stage, the people and products most likely to win the next wave of adoption will be those that can explain complex technology clearly.
