DeepMind is readying a more 'grown-up' ChatGPT rival

While ChatGPT continues to break the internet, Google's top AI lab is finally stepping into the ring with a different philosophy.

DeepMind CEO Demis Hassabis revealed in an interview with Time that the company is considering releasing its own chatbot, called Sparrow, for a private beta sometime this year.

DeepMind originally introduced Sparrow last year in a research paper, pitching it as a dialogue agent designed to drastically reduce the risk of unsafe and fabricated answers.

The bot that shows its receipts

ChatGPT is wildly fun, but its tendency to hallucinate facts and occasionally generate toxic responses has set off alarm bells across the industry.

DeepMind's answer to that problem is straightforward: make the bot cite its sources. Hassabis explained that the deliberate rollout timeline comes down to ensuring Sparrow includes key features ChatGPT currently lacks, most notably citing verifiable references. In his words, it is right to be cautious on that front.

In early testing, Sparrow answered factual questions with plausible explanations backed by evidence 78% of the time.

It is also built with explicit behavioral constraints. If a query touches on territory that should be left to human judgment, Sparrow is trained to politely decline rather than guess.

Even OpenAI CEO Sam Altman recently acknowledged on Twitter that opening up these models comes with real dangers, conceding that OpenAI cannot anticipate every issue as deployment scales.

Why it matters

We think Sparrow marks the moment the chatbot race shifts from novelty parlor tricks to enterprise-grade reliability.

For everyday users and knowledge workers, a chatbot that admits what it does not know and provides real citations is infinitely more useful for actual work than a creative bullshitter. If Sparrow can consistently deliver verifiable answers, it solves generative AI's biggest trust deficit.

For the tech landscape, this puts real pressure on OpenAI and Microsoft. OpenAI is racing toward monetization with a paid ChatGPT tier, but DeepMind is betting that safety and accuracy will win in the long run.

The real risk for DeepMind is over-sanitization. If guardrails are tuned too aggressively, the bot risks becoming robotic, frustrating, and unhelpful compared to ChatGPT's fluid freedom. The upcoming private beta will show whether safety and charisma can actually coexist.