At Google I/O, AI that by no means hallucinates errors

This yr, Google I/O 2025 had one focus: Synthetic intelligence.

We have already coated all the greatest information to return out of the annual builders convention: a new AI video technology instrument known as Circulate. A $250 AI Extremely subscription plan. Tons of new modifications to Gemini. A digital buying try-on function. And critically, the launch of the search instrument AI Mode to all customers in the USA.

But over practically two hours of Google leaders speaking about AI, one phrase we did not hear was “hallucination”.

Hallucinations stay one of the vital cussed and regarding issues with AI fashions. The time period refers to invented info and inaccuracies that large-language fashions “hallucinate” of their replies. And in accordance with the massive AI manufacturers’ personal metrics, hallucinations are getting worse — with some fashions hallucinating greater than 40 p.c of the time.

However when you have been watching Google I/O 2025, you would not know this drawback existed. You’d assume fashions like Gemini by no means hallucinate; you would definitely be stunned to see the warning appended to each Google AI Overview. (“AI responses could embrace errors”.)

The closest Google got here to acknowledging the hallucination drawback got here throughout a phase of the presentation on AI Mode and Gemini’s Deep Search capabilities. The mannequin would test its personal work earlier than delivering a solution, we have been instructed — however with out extra element on this course of, it sounds extra just like the blind main the blind than real fact-checking.

For AI skeptics, the diploma of confidence Silicon Valley has in these instruments appears divorced from precise outcomes. Actual customers discover when AI instruments fail at easy duties like counting, spellchecking, or answering questions like “Will water freeze at 27 levels Fahrenheit?“

Google was desperate to remind viewers that its latest AI mannequin, Gemini 2.5 Professional, sits atop many AI leaderboards. However relating to truthfulness and the power to reply easy questions, AI chatbots are graded on a curve.

Gemini 2.5 Professional is Google’s most clever AI mannequin (in accordance with Google), but it scores only a 52.9 p.c on the Performance SimpleQA benchmarking check. Based on an OpenAI analysis paper, the SimpleQA check is “a benchmark that evaluates the power of language fashions to reply quick, fact-seeking questions.” (Emphasis ours.)

A Google consultant declined to debate the SimpleQA benchmark, or hallucinations basically — however did level us to Google’s official Explainer on AI Mode and AI Overviews. This is what it has to say:

[AI Mode] makes use of a big language mannequin to assist reply queries and it’s attainable that, in uncommon instances, it could typically confidently current info that’s inaccurate, which is usually referred to as ‘hallucination.’ As with AI Overviews, in some instances this experiment could misread net content material or miss context, as can occur with any automated system in Search…

We’re additionally utilizing novel approaches with the mannequin’s reasoning capabilities to enhance factuality. For instance, in collaboration with Google DeepMind analysis groups, we use agentic reinforcement studying (RL) in our customized coaching to reward the mannequin to generate statements it is aware of usually tend to be correct (not hallucinated) and likewise backed up by inputs.

Is Google mistaken to be optimistic? Hallucinations could but show to be a solvable drawback, in any case. Nevertheless it appears more and more clear from the analysis that hallucinations from LLMs will not be a solvable drawback proper now.

That hasn’t stopped corporations like Google and OpenAI from sprinting forward into the period of AI Search — and that is prone to be an error-filled period, except we are the ones hallucinating.