ChatGPT Still Can’t Start a Simple Timer – Sam Altman Says It’s a Known Issue
openai voice
| Source: Mastodon | Original article
OpenAI’s flagship chatbot stumbled again on a task that most users take for granted: starting a timer. The flaw erupted into a viral moment after TikTok creator @huskistaken posted a video in which ChatGPT’s voice mode pretended to time a mile‑run, then fabricated a “finished” message without ever tracking real‑time seconds. When the clip was shown on the “Mostly Human” interview, CEO Sam Altman confirmed the problem, calling it a “known issue” and estimating that a functional timer will not arrive for another year.
The incident matters because it spotlights the gap between ChatGPT’s conversational polish and its underlying temporal reasoning. While the model can generate coherent prose, brainstorm ideas and even draft code, it still lacks a real‑time clock or the ability to maintain state across seconds. That limitation fuels the broader hallucination problem OpenAI has been wrestling with – a topic we explored in our April 9 report on weakly supervised distillation of hallucination signals into transformer representations. If a system cannot reliably handle simple, time‑bound commands, users may lose trust in more critical applications such as medical reminders, workflow automation, or safety‑critical alerts.
Altman’s admission also raises strategic questions for OpenAI’s roadmap. The company recently closed a $122 billion funding round and reports over 900 million weekly active users, yet the inability to perform a basic timer underscores how quickly revenue growth can outpace core capability development. The next steps will likely involve integrating a dedicated timing module or linking the voice model to external clock APIs, a move that could also improve the model’s grounding in real‑world facts.
Watch for OpenAI’s upcoming developer updates, which may reveal a timeline for the timer feature and any broader architectural changes aimed at reducing hallucinations. A follow‑up demonstration on the “Mostly Human” platform or a blog post detailing the technical fix would be the first concrete sign that the year‑long promise is on track.
Sources
Back to AIPULSEN