Google’s Gemini AI launch marred by questions over capabilities

Google unveiled its much-anticipated synthetic intelligence system Gemini on Wednesday, touting benchmarks suggesting it may compete with OpenAI’s industry-leading GPT-4 mannequin in reasoning skills. However the launch has rapidly been overshadowed by accusations that the tech big overstated Gemini’s capabilities.

In a tightly choreographed video demonstration, Google confirmed Gemini interacting with visible information via a digital camera mounted above a desk, fielding questions and reasoning via issues as a human assistant manipulated objects. The slick presentation implied Gemini may function an clever digital assistant able to subtle dialog and help with every day duties. 

But tech consultants analyzing the underlying know-how behind the scenes say Gemini could fail to stay as much as Google’s lofty aspirations. The corporate is rolling out Gemini in three variations — Gemini Professional, Gemini Gentle and Gemini Extremely. However early opinions of the mid-range Professional model made public on Wednesday point out it nonetheless struggles with duties that ought to be routine for a state-of-the-art AI system.

“I’m extraordinarily disenchanted with Gemini Professional on Bard,” stated Victor de Lucca, an early tester of the Bard replace, in an X.com put up displaying that the AI system was not capable of accurately record the 2023 Oscar winners. “It nonetheless offers very, very unhealthy outcomes to questions that shouldn’t be onerous anymore with RAG.”

Others identified discrepancies between the capabilities Google claimed in its benchmark testing and what seems attainable with the publicly accessible Professional model. 

“Google Gemini Extremely [is] solely 4% higher…utilizing completely different prompts versus GPT-4-0613?” requested developer Nick Dobos in a broadly shared put up on X.com, suggesting the comparability was deceptive. 

The slick Gemini video additionally got here underneath hearth after a Google spokesperson confirmed to Bloomberg that the footage was pre-recorded and narrated after the very fact, quite than representing a stay conversational demo.

The controversy illustrates the challenges Google faces in advertising AI programs to shoppers. Whereas techies eagerly dissect benchmark numbers and educational papers, most people responds extra to inspirational movies promising a revolutionary future.

This disconnect has tripped up large tech corporations earlier than, maybe most infamously in 2016 when Microsoft’s Tay chatbot was yanked offline after studying hate speech from Twitter customers. That is additionally the second time Google Bard has been accused by the tech group of falling wanting the corporate’s promise. In September, VentureBeat reported that Google Bard was nonetheless failing to ship on its promise — even after main updates.

Google is, in fact, aiming to recuperate rapidly, promising to make Gemini extra broadly accessible to builders and researchers who can absolutely put it via its paces. However the rocky begin reveals the tech big nonetheless has work to do if it needs its AI assistant to measure as much as the hype.