From Octopus Paul to AI prompts: How Fifa World Cup predictions evolved

When Paul picked Germany to defeat Argentina and that panned out, an Argentine chef threatened to pan-sear him. The recipe had started with detailed braising of the sea animal turned football pundit. His German minders pledged to allow no such thing, but then Paul was declared traitor when he offended German fans by predicting the home team would lose. They too announced they would eat him.

But then the Spanish prime minister offered to “send state protection” to Germany after he correctly identified them as winners and was under attack, and a diplomatic crisis might’ve been averted. In his 33 months of lifetime, Paul attracted a transfer fee proposal of 30,000 pounds, and warded off competition from German porcupines, pygmy hippopotamus and tamarins who all failed.

A Singapore parakeet fetched better efficiency, but Paul had by then retired from dipping into jars with mussels, oysters and two competing national flags, his first choice of snack considered the winner. He died in late 2010, but not on a plate. The smart-armed creature posthumously received attention from iPhone apps, Google doodles and even Chinese movie thrillers.

Sixteen years hence, though, not many are staring at aquariums. They are squinting into smartphones and at whirring supercomputer screens to identify the FIFA World Cup winners. Getting AI to guess the winner on July 19 however, is irresistible.

USAToday used Copilot, though they offered an insightful disclaimer up-front. “Large Language Models aren’t keeping up with sports results and matches.” This led to refining of Prompts. Nevertheless, Copilot had France ejecting Argentina in the round of 16, Morocco making the semis, Canada beating South Korea and France defeating Brazil in the final.

AI engineering developers had resoundingly proven one basic English question fed into a comp is smarter than 8-armed octopuses. Frank Andrade on Artificial Corner deployed Claude to analyse 49,000 matches, with old ones diminishing in importance, and used an Elo score (fewer points for defeating lower ranked) to find his way to Spain at 27% odds, Argentina 21%, with France and England in semis. AI re-ran the testing 50,000 times, to be sure. And a retrospective AI test on 2018 and 2022 WCs led to 52% correct results.

ActionNetwork.com picked France, using 1200 data points across 25 variables, with Norway predicted to throw up a surprise. A spokesperson said “the 2026 World Cup outcome was predicted based on key data, including national team form, World Cup history, market value of squads and coach profiles.” France’s market value? €1.48 billion, as per the website.

Depending on precise Prompts, Opta AI can summon France and Spain as favourites, while slotting Norway into the Top 10, while ChatGPT lines up Spain, France, Argentina, England, Portugal, Brazil. It is not too different from what a casual fan, not enamoured by Messi, might pick.

Between Paul and Prompts, mathematicians have weighed in too, picking Holland as a left-field choice, Portugal closely behind. Grok patriotically and predictably gave a leg-up to USA’s chances, calling them ‘disruptors’. China meanwhile is turning the entire predicting-business into a nation-wide game of its own. IChonqing reported that their own tech tranche Lenovo with Migu have planned a “human-versus-AI” World Cup prediction event.

“Lenovo’s Tianxi AI agent will bring together 12 Chinese AI models, including DeepSeek, Kimi, ERNIE Bot, Qwen, and China Mobile’s Jiutian, to compete with fans in match predictions. The program is designed to turn AI prediction into a participatory viewing experience, rather than a background technology tool,” IChonqing quoted.

RIP Paul. Machines 1 def Mollusk 0.