Kitajsko tehnološko dohitevanje oziroma celo prehitevanje se ne dogaja samo v zelenih tehnologijah in čipih, pač pa se je zgodilo tudi na področju umetne inteligence. Ta teden je kot bomba odjeknila novica, da je “neki” Liang Wenfeng na Kitajskem oblikoval nov bot umetne inteligence (velik jezikovni model) DeepSeek, ki se je na neodvisnih testih izkazal z boljšimi performansami kot bvis ostali ameriški boti, boljši tudi od zadnje verzije OpenAI GPT. “Strašljivo” dejstvo pa je, da Liangu uspelo svoj bot natrenirati zgolj za delček (3 % !!!) vsote, ki ga je Open AI porabil za treniranje svojega bota. Zadnja verzija DeepSeek-V3 je bila razvita v zgolj 2 mesecih in za zgolj 5.6 mio dolarjev.

Nekateri pravijo, da je DeepSeek za umetno inteligenco to, kar je BYD za električne avtomobile.

A Chinese computer nerd is stunning tech experts with a bargain price AI tool that beats the world’s top players on several benchmarks.

Liang Wenfeng’s system, called DeepSeek-V3, has powers of “reasoning” that match or beat those of the equivalent devices by the world’s top tech companies, at a better cost, tests show.

“DeepSeek-V3 has emerged as the strongest open-source model currently available and achieves performance comparable to leading closed-source models,” said Supreeth Koundinya, a tech analyst in India.

‘Brilliant young minds’

While Liang and his colleagues spent more than a billion yuan over recent years to research and develop AI systems, the ground-breaking new model, DeepSeek V3 LLM, was developed in two months with a budget of just US$5,580,000, the cost of a single nice apartment in Monaco or Hong Kong.

In contrast, Meta’s minimum AI budget for the financial year is a staggering US$38,000,000,000.

“Sometimes having less means innovating more,” said San Francisco angel investor Henry Shi, writing on X yesterday. “DeepSeek proves you don’t need billions in funding, hundreds of PhDs, a famous pedigree; just brilliant young minds, the courage to think differently and the grit to never give up.”

Cheap and cheerful

How did it happen? Liang, born in Guangdong, enrolled in Zhejiang University in Hangzhou where he worked on teaching computers how to see like humans do. Spotting an opportunity, he moved into the computer side of stock market trading, where he made solid money.

In his 30s, he saw a different opportunity and used the cash he had made to start an AI firm, called DeepSeek.

The Hangzhou-based start-up was invisible – until a recent series of tests showed that his cheap and cheerful recent model performed better at a lower cost than the best efforts of heavily financed top AI models developed by Meta Platforms and OpenAI, creator of ChatGPT.

Meeting China’s leaders

Liang, now 40, moved further into the spotlight on Monday this week, when he appeared at a symposium hosted by China’s Premier Li Qiang in Beijing–and became seen as the country’s face of AI.

The US government has been trying to halt China’s development by forcing companies and countries around the world, including in China-owned Taiwan island, to stop selling all but the most basic chips and chip-making equipment to the developing Asian country.

The western media justified the move by saying that China was using the chips for the military, but the west’s own experts, at MIT, said 99.9 per cent were used for civilian purposes, such as to enable the functioning of medical scanners and engineering equipment.

Vir: Nury Vittachi via X

Ta teden so socialna omrežja ponorela v tekmovanju, kdo bo bolj nazorno predstavil razlike med DeepSeek in OpenAI GPT. Spodaj več o tem. Meni se je zdela zanimiva razlika v tem, kako DeepSeek in OpenAI GPT odgovorita na vprašanje, naj vizualno predstavita Pitagorov izrek. Tudi jaz sem šel poskusit oba in v eni varianti zahteval vizxualno predstavitev, v drugi pa PPT predstavitev Pitagorovega izreka ter nekaterih drugih zanimivih vprašanj, ki so se mi utrnila. DeepSeek se je res izkazal kot boljši.

Ti Kitajci so nori!