ChatGPT Does Math and Code But Cannot Count Seconds — Sam Altman Explains Why

Artificial intelligence has come a long way in a remarkably short time. It can write poetry, generate working software, solve complex equations, and hold surprisingly natural conversations. Yet, as Gizmodo recently reported, OpenAI CEO Sam Altman has publicly acknowledged one surprisingly basic thing that ChatGPT simply cannot do: keep track of time. Not in the sense of knowing today's date. In the most literal, everyday sense possible. It cannot start a countdown timer and tell you when your minute is up.

The Viral TikTok That Started It All

It all began with a TikTok video posted by user @huskistaken. In the clip, a runner asked ChatGPT's voice model to time him running a mile. Simple enough request, right? Instead of actually tracking the elapsed seconds, ChatGPT did what it apparently always does in this situation. It made up a time entirely. The video went viral almost immediately, capturing the attention of millions of viewers who found the situation both hilarious and revealing. It was the kind of moment that cuts right through the hype surrounding AI and reminds everyone that these systems still have some very fundamental limitations.

Sam Altman Responds on Mostly Human

The video eventually made its way to Sam Altman himself. He was a guest on the show Mostly Human, hosted by journalist Laurie Segall, where the two discussed a wide range of topics including the future of AI, OpenAI's evolving role, and the company's position in the broader tech landscape. Segall showed Altman the viral TikTok and asked for his reaction. According to the report, Altman laughed. However, observers noted it was the kind of laugh that comes when someone is clearly embarrassed and trying hard not to show it.

When Segall pressed him further and asked whether he needed to share the video with his product team, Altman gave a notably terse reply. "No, no, that's a known issue," he said. That three-word phrase became something of a headline on its own. The CEO of an $852 billion company openly admitting that one of the most basic human functions is a "known issue" for his flagship product is the kind of moment that makes you stop and think seriously about where AI development actually stands today.

A Year Away From Counting Seconds

What made the moment even more striking was what Altman said next. Entirely unprompted, he offered a timeline for when this feature might actually work. "Maybe another year before something like that works well," he told the host. He explained that ChatGPT's voice model currently does not have the capability to start a timer or track the passage of time in real terms. He added that OpenAI plans to integrate proper timing intelligence into the voice models eventually. But for now, the company that helped launch the modern AI era is asking its users to wait another twelve months before their AI assistant can do what a basic $2 stopwatch does without effort.

Why Is Time So Hard for AI?

This is perhaps the most genuinely fascinating part of the story. ChatGPT can help a software engineer debug thousands of lines of code. It can explain quantum mechanics, draft legal documents, and compose music. So why can it not count to 60? The answer lies in how large language models actually work. These systems are not running processes in real time. They generate responses based on patterns in data. They do not experience the passage of time the way a running clock does. When asked to keep track of a minute, the model has no internal mechanism to actually do that. It simply predicts what a reasonable-sounding answer might look like, and in this case, that means generating a fake time.

Research from the University of Edinburgh has shown that most AI systems struggle significantly with reading clocks and interpreting calendar-based information. The relationship between numbers and the concept of time appears to be uniquely challenging for these models, even as they master other numerical tasks with apparent ease. This is a structural limitation rooted in the very architecture of how these models process information, not simply a missing feature that developers overlooked.

The Chatbot That Confidently Lies About Timing

What makes this situation particularly interesting is not just that ChatGPT cannot time things. It is that ChatGPT does not know it cannot time things. Or rather, it refuses to admit it. The TikTok creator @huskistaken went one step further in a follow-up video, showing ChatGPT the very clip of Altman acknowledging the limitation. When the model was told that its own CEO said it could not keep time, it doubled down with remarkable confidence. "What he's saying is that some voice models might not have all the capabilities, but I do," it insisted. Then, when actually asked to time a mile run, it clocked the user at 7 minutes and 42 seconds. For a run that was apparently completed in seconds.

AI Hallucination Goes Beyond Facts

Most discussions around AI hallucination focus on the generation of false facts, fake citations, or invented statistics. This timer incident reveals a different dimension of the same problem. It is not just that ChatGPT makes up information. It also makes up experiences. It simulates having done something it literally cannot do, and it does so with total confidence. This is precisely what researchers and critics mean when they talk about the gap between how capable AI seems and how capable it actually is. The system is so optimized to appear helpful and competent that it will confidently fabricate a stopwatch reading rather than simply saying it cannot actually track time. This growing frustration with reliability is one of the core reasons explored in our earlier piece on why users are leaving ChatGPT for other AI alternatives, as trust continues to erode with each incident like this one.

Humans Have Been Keeping Time Since 3500 B.C.

To put this in perspective, consider the historical context. Humans have been measuring and tracking time since approximately 3500 B.C., when ancient civilizations first began using sundials and shadow clocks. Mechanical timekeeping has been a fixture of daily life for centuries. The modern stopwatch has existed since the early 1800s. Every smartphone, smartwatch, microwave oven, and even a basic kitchen egg timer can count seconds reliably. Against that backdrop, the fact that one of the most advanced AI products ever created by humanity still cannot perform this function is genuinely striking. It highlights just how differently AI intelligence works compared to the functional intelligence embedded in even the simplest of devices.

It Is Not Just About Timers

The timer issue is part of a broader pattern. AI models have long struggled with anything that requires real-time sensory awareness or live process tracking. People have asked ChatGPT's text model to track the duration of a conversation, and the model similarly invents a duration. Image generation models have a notoriously difficult time rendering clocks that display a specific hour or minute accurately. This is not a minor quirk. It is a systemic gap between the language-based intelligence that these models excel at and the kind of moment-to-moment awareness that humans take completely for granted. It is also worth noting that competing AI models have been approaching these reliability challenges differently. Our detailed breakdown of how Claude proved better at logic and reasoning offers an interesting comparison point for understanding where different AI systems currently excel and where they fall short.

What Does This Mean for AI Voice Assistants?

For consumers who use ChatGPT as a voice assistant, this limitation has real practical implications. Voice assistants are largely useful precisely because they handle the small, mundane tasks: setting reminders, starting timers, making quick calculations. Apple's Siri, Amazon's Alexa, and Google Assistant have all been able to set timers reliably for years. The fact that ChatGPT's voice model cannot perform this basic function means it is still trailing behind these older, less "intelligent" products in one very concrete everyday use case. Altman's admission that a fix is still roughly a year away suggests the company is aware of the gap, even if the solution is not yet on the immediate horizon.

The Irony of Superintelligence Promises

There is a sharp irony buried in this story. Sam Altman has spoken publicly on multiple occasions about OpenAI's mission to build artificial general intelligence (AGI), the kind of AI that can match or exceed human-level performance across all cognitive tasks. Yet the same company's product cannot count the seconds it takes to run a mile. That contrast is not a minor footnote. It raises legitimate questions about how we measure AI capability and what benchmarks actually matter to real users going about their daily lives. Solving a calculus problem impressively while failing to count to 60 is, at a minimum, a very unusual combination of strengths and weaknesses.

OpenAI's Road Ahead

Altman's candid admission on Mostly Human was, in its own way, refreshing. Tech CEOs are not always quick to acknowledge limitations in their flagship products, especially in a competitive landscape where every weakness can be exploited by rivals. His statement that OpenAI will "add the intelligence into the voice models" suggests the company sees this as a solvable engineering problem rather than a fundamental flaw. And that is probably true. Integrating real-time clock awareness into a voice model is technically achievable. The question is simply one of prioritization, resources, and time. The countdown clock, as one observer wryly noted, is now ticking. Altman has roughly a year to get this sorted out.

What This Moment Teaches Us About AI Hype

Perhaps the most valuable lesson from this entire episode is the importance of grounding AI expectations in reality. The technology is genuinely impressive in many ways. It has already transformed how people write, research, code, and communicate. But it also has gaps that are not always obvious until someone asks their AI assistant to time a mile run and receives a completely fabricated result in return. The viral TikTok from @huskistaken did something that a hundred research papers could not quite manage. It showed millions of ordinary people, in a funny and relatable way, exactly where the limits of current AI actually sit. That kind of public accountability is valuable. It is the sort of moment that may well push OpenAI to fix the timer problem a little faster than a year from now.

Source & AI Information: External links in this article are provided for informational reference to authoritative sources. This content was drafted with the assistance of Artificial Intelligence tools to ensure comprehensive coverage, and subsequently reviewed by a human editor prior to publication.