Google's Bard flubs its very first demo

When you rush out a product just to counter a rival, you almost always trip over your own shoelaces.

Google just officially unveiled Bard, its conversational AI designed to take on OpenAI's ChatGPT. But before regular users could even get their hands on it, the bot managed to publicly faceplant in Google's own promotional materials.

Astronomers quickly spotted a glaring factual error in the very first demo GIF shared by the company on Twitter, which racked up over 880,000 views. Alphabet's stock took an immediate hit as a result.

The space telescope flub

In the Twitter ad, Google pitched Bard as a launchpad for curiosity that could simplify complex subjects for anyone.

It showed someone asking: "What new discoveries from the James Webb Space Telescope can I tell my 9 year old about?"

Bard confidently spat out three bullet points. One of them claimed that the telescope "took the very first pictures of a planet outside of our own solar system."

That is completely untrue, and real astronomers wasted no time calling it out.

Astrophysicist Grant Tremblay pointed out on Twitter that JWST was definitely not the first to image an exoplanet. Bruce Macintosh, director of UC Observatories, chimed in by noting he personally imaged an exoplanet 14 years before JWST even launched.

According to NASA's own archives, the very first direct image of an exoplanet was actually captured back in 2004 by the European Southern Observatory's Very Large Telescope.

Google spokesperson Jane Park later scrambled to do damage control, stating that the incident highlights the need for rigorous testing, which the company is kicking off with its Trusted Tester program.

Why it matters

Large language models make stuff up all the time. But making a basic factual blunder in your marquee marketing ad is a catastrophic unforced error, especially for a company whose entire empire is built on search accuracy.

In our view, Google clearly panicked. Microsoft had just unveiled its AI-powered Bing the day before, complete with heavy disclaimers telling users that the AI might make mistakes and they should verify facts themselves. Google, desperate to show it was not falling behind, rushed this demo out without even running basic fact-checks on the prompt.

For regular users, this is a glaring reminder that chatbots are essentially autocomplete engines operating on probability, not determinism. They sound authoritative even when they are completely hallucinating. If you are using these tools for homework, research, or real work, the burden of truth still falls entirely on you.

For the tech industry, it exposes the messy reality of the AI search race: spitting out coherent paragraphs is easy, but getting an LLM to reliably stick to the truth remains an unsolved problem.