The great AI hallucination

WAYNE KUBLALSINGH

ARE YOU familiar with people who suffer from hallucinations? Like paranoid people, or those on LSD and other narcotics? Who hear imaginary voices, or see imaginary people? Or those news anchors on global mainstream media, CNN, the BBC, Fox, who gaze into the camera (you) with the deepest of earnestness, directness, the utmost pathos in their voices, but continually generate, repeat and reinforce fake news?

What about AI (artificial intelligence) hallucinations? AI programs which generate and repeat information, with the greatest of aplomb, confidence and poise, from the most “established,” “secure," “reliable” and “trusted” data banks, but information which is biased, ambiguous, or plainly inaccurate and wrong?

For years we have been subject to a grand hallucination from Big Tech AI gurus. That to develop credible, resourceful AI programs you need astounding volumes of energy. Such as would be produced by large power plants, nuclear power. That you need astounding sums of money, hundreds of billions (US).

That you need astounding levels of compute, computer processing power to train models, large GPUs working in tandem. That you need cutting-edge, high-end chips solely produced by elite companies. That the cost of generating AI was so high, licensing fees and subscriptions had to be paid for its use.

Big Tech AI savoured and swanked on these myths. On Wednesday last week, Donald Trump called a press conference to promote Stargate, a US$500 billion AI joint-venture infrastructure project, backed by ChatGPT backers OpenAI and Oracle, as well as Japanese Softbank.

The very same CEOs of Big Tech, Google, Facebook/Meta, Yahoo, the former Twitter, who had flagellated and vilified Trump, were, at his inauguration ceremony, lined up on the front row, like those turncoat barons and dukes of the Ancien Régime. These narratives of Big Tech AI naturally lead one to suppose that AI is deserving of astronomical investments from banks and substantial returns from the market.

We have been also fed the narrative that the US is the "natural" home of AI and worthy of global AI "dominance," while China is a thief, a spy, a threat to the privacy and freedoms of the citizens and US state.

In 2019, following the placing of Huawei, the Chinese telecommunications creator, on an export blacklist, cutting China off from key technologies, the US instigated the house arrest (in Canada) of Meng Wanzhou, the daughter of the company’s founder Ren Zhengfei.

It also imposed strict controls on exporting high-end chips, chip-making equipment, and advanced semi-conductors to China, even from Taiwan. It banned TikTok in the US, which has since been temporarily reversed by Trump.

The narrative was that China was a national security risk; that its technologies have "back doors" to steal sensitive information of US citizens and the US government, and that China steals US technologies for its own innovation and advancement.

More predictable than ironic, Edward Snowden is in exile in Russia today, called a "traitor" for revealing that the US is a global surveillance (spy) state, which uses back doors to hack and store data from everyday US mobile and computer systems used by US citizens, by foreign citizens, by foreign governments and by alleged "enemies."

Now, lo and behold, up comes January 27! On this day a relatively small Chinese lab based in Hangzhou launched its AI model Deep Seek R1. Immediately, the model was downloaded by millions of users globally. For free! What is called open source. No subscriptions, no licence fees. And built, it reports, for US$6 million. From hedge fund income.

This is a tiny fraction of the cost of production, hundreds of billions of dollars, incurred by other AI models like ChatGPT and Anthropic. It surpassed ChatGPT downloads from the Apple App Store. And when compared (benchmarked) with the existing AI language models, outperformed them in speed, accuracy and resourceful.

It used large existing AI models to train its model. It did not use high-end chips. No need for astounding volumes of energy, or brute force processing power. It applied sophisticated algorithms, innovative machine-training strategies, using a small group of highly-talented AI developers.

It had also been transparently publishing its models, advances, benchmarks, since its founding in May, 2023.

America panicked. On Monday, the stocks on the tech-heavy Nasdaq exchange plummeted. Nvidia, the elite chip-manufacturer, whose high-tech chip was banned from export to China, found its stocks tumbling just under 17 per cent. Broadcom Inc, another chip maker, fell to 17.4 per cent. Oracle, the manufacturer of software, hardware and Cloud applications, dropped by 13.4 per cent. Big Tech executives immediately sought their back rooms, deep diving into this new disruptive technology. How had this happened?

US commentators called it the "Sputnik moment," harkening back to the Soviet Union’s artificial satellite launch into Earth’s orbit. Others called it an "AI arms race." On Tuesday morning, the first question posed to Karoline Leavitt, Trump’s Press Secretary, at her first press conference, was whether Trump felt there was something “fishy” about Deep Seek.

On Wednesday morning, Trump’s Commerce Secretary, Howard Lutnick, at his confirmation hearing, laid bare his fangs: "They stole things, they broke in. It's got to end, and I'm going to be rigorous in our pursuit of restrictions and enforcing those restrictions to keep us in the lead, because we must stay in the lead…We need to stop it, we need to end it.”

Other state and Big Tech actors have likewise attacked Deep Seek, vowing to win back America’s global "AI dominance."

But ordinary AI users and AI professionals are happy with Deep Seek, an AI which seeks to democratise the current AI hegemony. To unplug us from the unipolar US database. To decapitalise AI. To smash the monopolistic AI myth.

In the US system, if the commanding heights of transformative technologies are not agglomerated, monopolised, controlled by government and large corporations, all hell breaks loose. The good news, and opportunity, for aspirants to AI development, individuals or small nation-states, is that AI’s fake narrative, its hallucination, has been smashed, exposed, laid bare.