Google's Gemini Demo Fails to Back Up Promises of AI Breakthrough

Google's Gemini demo fails to back up its AI breakthrough claims

Does a single announcement justify retiring humanity? Google unveiled Gemini, its multimodal artificial intelligence model, wrapping it in a video where the machine describes images, solves problems, and converses in real time. The repeated message: compared to this, humans are apes. Then came the fine print. Leaked prompts revealed the demo was a polished production, not a historical leap. The ensuing debate focused less on what Gemini does and more on what Google claims it does.

What Gemini promises versus what the demo actually showed

The official promise is multimodality: text, image, and video in a single pass, with object and trajectory tracking. Practical applications include camera glasses for the visually impaired, instant conference summaries, or analyzing body language in professional settings. None were verified in production. Leaked materials showed the video was fabricated, with no real-time analysis and reasoning below GPT-4-v. The popular conclusion: Google sold a movie, and the public mistook the trailer for the release.

Bard, GPT-4, and a race where Google arrives late

The competitive landscape doesn't help the narrative. The testable version of Bard is described as weak, with Gemini integration scheduled for sometime in 2024. Microsoft acted faster. Some see this as an excuse to hide the product's immaturity. Meanwhile, experimental Bard is close to GPT-4, with expectations that the Pro version will close the gap. None of this was verifiable then: the good version was unavailable.

Which jobs withstand the test: from programmer to plumber

If the tool works, what's worth studying? An uncomfortable recommendation emerges: don't train as a deep programmer and pursue hands-on trades that can't be teleworked or copied with a few clicks. Plumbers, electricians, bricklayers. The catch: these aren't studied online. Tech workers are warned that if your job involves predictable coding, you have an expiration date. Security guards appear as a refuge, with IT professionals switching sectors. One exception remains constant: caring for children and the elderly won't be done by machines.

The vet who already passes diagnosis to AI

Where replacement isn't theory is in medical imaging. A vet explains bluntly: with an image and a few parameters, the machine identifies lesions and proposes treatment. In a slaughterhouse line, it detects pathologies at speeds impossible for human inspectors, with near-total accuracy. It doesn't tire, doubt, or take holidays. For companies, this means direct labor savings. The same pattern appears in finance, where certain market processes already operate on automated criteria. Balance sheets, floor plans, delivery routes: tasks occupying entire departments are now handled with a single phrase.

Does AI replace humans or only mediocre ones?

Opinions diverge. One view holds that AI is just rehashed external material, creating nothing or thinking independently while creators censor it. The rebuttal is devastating: it's at the level of 90% of the population, which is a verdict for the average citizen. Others point out the obvious: humans tremble with a knife to the heart, can't work three shifts without bathroom breaks, and lose strength by mid-afternoon. The advantage isn't intelligence. It's endurance.

Antiestéticar, narratives, and those who already declare the world dead

Part of the conversation frames the announcement as part of a supposed narrative campaign from climate change to UFOs, backed only by distrust of big tech. There's no proof of this alleged artifice, and the argument doesn't hold. Oppositely, the view is that the known world model is dying, and denying it resembles video club owners when the internet arrived. The problem with this prophecy is that it brings no dates. Only certainty.

There's always solace in the idea that AI will plant potatoes and herd sheep on rainy days. For now, it only writes reports on how poorly the harvest is going.