OpenAI ne elaan kiya hai ki usne 88 ghante mein hi ek badi mushkil ganit ki samashya ko hal kar diya, jismein kareeb 10,000 AI agents ka jhund istemal kiya gaya. Yeh samashya, Navier-Stokes equations ka hissa hai jo fluid movement se sambandhit hai, aur iske key pehluon ke liye pichhle 90 saal se koi proof nahi mila tha. Lekin kaafi bots aur ek bhaari bijli ka bill ho, to jaise apparent hai, kuch bhi sambhav hai.

Mangalvaar ko, ChatGPT banane wali company ne bataya ki usne AI bots ka ek samooh Navier-Stokes existence aur smoothness samashya par kendrit kiya tha, jo itni ziddi hai ki yeh Clay Mathematics Institute ke Millennium Prize problems mein se ek hai, jiska pehla hal karne wale ko $1m ka inaam milta hai. Company ne apne is hal ko ek "milestone" aur is baat ka saboot bataya ki AI tools tezi se behtar ho rahe hain - haalanki, notably, ise abhi tak swatantra roop se verify nahi kiya gaya hai ya institute ne ise swikar nahi kiya hai.

Yahan timeline hai: August ke ant mein, OpenAI ne ek naya internal model train karna shuru kiya jisne ganit mein ruchi dikhayi. Pichhle hafte, 1 September ko, company ne afwahein sunin ki do Millennium Prize problems hal ho chuki hain, isliye usne hazaaron AI bots ko baaki bachi samashyaon par kaam karne ke liye set kar diya. 5 September tak - kareeb 88 ghante baad - uske paas Navier-Stokes samashya ka hal tha, jo turbulence ke dil mein hai, ek aisi ghatna jise insaan abhi tak acchi tarah nahi samajh paaye hain. Bots ne kareeb 3 million messages ka aadan-pradan kiya aur Navier-Stokes par akele 130 billion output tokens (AI model dwara utpann text aur code ki lineein) jala diye. OpenAI ki apni pricing par, is prayas ki lagat kareeb $10m (£7.3m) hoti.

Hal, report ke anusar, Millennium Prize dwara maange gaye proof ke char statements mein se do ko suljhaata hai. Lekin OpenAI inaam ka dava nahi kar raha hai. "Is result ko jaari karne mein hamara lakshya apne AI models ki kaafi pragati ki report karna hai," company ne kaha. "Ham is result ke liye Millennium Prize ka dava karne ka iraada nahi rakhte." Yeh suvidhajanak hai, kyunki inaam $1m ka hai aur unhone compute par pehle hi $10m kharch kar diye hain.

Swabhavik roop se, vivad chhid gaya hai. Tristan Buckmaster, New York University mein ganit ke professor, ka dava hai ki ve aur Levent Alpöge, OpenAI ke prati-spardhi Anthropic ke ganitagya, OpenAI ke Codex tool ka upyog karke isi samashya par kaam kar rahe the. Buckmaster ka kehna hai ki 3 September ko, unhein pata chala ki "hamari pragati ke baare mein jaankari OpenAI ko de di gayi thi." Unhone OpenAI ke saath emailon ka aadan-pradan saarvajanik kiya, jismein aarop lagaya ki company ne Navier-Stokes par apna kaam Buckmaster aur Alpöge ki pragati ke baare mein jaanne ke baad hi shuru kiya. "Vikalp yeh hai ki ghoshnaon ki ek sh्रृंखla ko kuch aisa kehne diya jaye jo main janta hoon ki jhooth hai," unhone kaha.

OpenAI ne jawab mein Buckmaster aur Alpöge ke "samkalin kaam" ko badhaai dete hue use "remarkable" bataya. Company ne zor diya ki usne unka kaam tab tak nahi dekha jab tak use saarvajanik roop se jaari nahi kiya gaya aur koi user data access nahi kiya gaya. Lekin ek spy novel ke yogya chhote aksharon mein, OpenAI ne add kiya: "Jabki sambhavna kam hai, hum isse inkar nahi kar sakte ki unke hamare products ke upyog se prapt de-identified data ne hamare models ko behtar banane mein madad ki." Doosre shabdon mein: aapka AI homework sirf grade nahi ho sakta - iska upyog duniya ki sabse kathin ganit samashyaon ko hal karne ke liye ho sakta hai bina aapko kabhi credit diye.