Kuyini ukubuka kuqala kwe-OpenAI o1? Isingeniso esiphelele sokuhlaziya okuqhathanisayo ne-GPT-4o

Imodeli yolimi olukhulu yakamuva ye -OpenAI o1: ukuletha inkathi entsha yokucabanga.

Njengoba kungenzeka ukuthi uke wezwa, ngemva kwezinyanga zokuqagela, i-OpenAI ikhiphe inguqulo yayo yakamuva yemodeli yayo yolimi olukhulu ngoSepthemba 12, 2024 .

Ngokumangalisayo, le modeli entsha ayibizwa ngokuthi i-GPT-5 . Kunalokho, ikhishwe ngaphansi kwegama elilula neliqondile: o1 . Kwenzekeni ngemuva kwezigcawu? Ake sithole!

Yini i-o1?

I-OpenAI o1 iyimodeli yakamuva yolimi olukhulu ye-OpenAI, eyakhishwa ngoSepthemba 12, 2024. Ngokumangazayo, ihlukile esimisweni sangaphambilini sokuqanjwa kwamagama "se-GPT", ikhetha esikhundleni salokho ilebula elisha ngokuphelele. Kusho ukuthini lokhu? Kusho ukuthi akuyona nje intuthuko ku-GPT-4, kodwa futhi kuyisiqalo sendlela entsha.

I-OpenAI ikhiphe amamodeli amabili e-o1: ukubuka kuqala kwe-o1 kanye ne-o1-mini . Babonise amakhono aqinile ekucabangeni, ezibalweni, kanye nasesayensini , bedlula ngqo i -GPT-4o yangaphambilini , ikakhulukazi ekhonweni lokucabanga, lapho befinyelele khona intuthuko enkulu phambili.

Ikhono lokucabanga liyisici esiyinhloko sochungechunge lwe-O1. I-OpenAI igcizelela ukuthi ifana kakhulu nokucabanga komuntu, okukwazi ukucabanga ngesikhathi sangempela kunokuthembela kuphela kudatha eqeqeshwe kusengaphambili. Lokhu kuhluke kakhulu kumamodeli adlule! Unesifiso sokuthi ikhono lokucabanga le-O1 linamandla kangakanani? Qhubeka ufunda ukuze uthole impendulo.

Ikhono lokucabanga lika-o1: impumelelo edlula okwedlule

Ungase ucabange, "Ingabe amakhono okucabanga abaluleke kangako?" Yingakho i-o1 ikwazile ukuzuza imiphumela emangalisayo emikhakheni eminingi.

1. Ukusebenza kwesayensi nezibalo : o1 Ezivivinyweni eziningi zesayensi, yenze cishe ngokulinganayo nabafundi bezobudokotela. Ezifundweni ezifana nefiziksi, ikhemistri, kanye nebhayoloji, ifinyelele esicongweni sempumelelo yezemfundo.

2. Imiphumela ye-US Mathematical Olympiad : U-O1 wangena kalula phakathi kwabangu-500 abaphezulu kulo mncintiswano onzima kakhulu. Cabanga ukuthi uncintisana nabanye babantu abasha abahlakaniphe kakhulu emhlabeni, kanti u-O1, oyimodeli nje, ubadlula kalula.

3. Impumelelo yokubhala amakhodi : i-o1 ibekwe endaweni yama-89 epulatifomu yokubhala amakhodi edumile emhlabeni wonke i-Codeforces , okuyinto emangalisayo! Ayikwazi ukuxazulula izinkinga zezibalo kuphela, kodwa futhi ibhale nekhodi eyinkimbinkimbi.

Konke lokhu kungenxa yamakhono okuphetha e-O1 ngesikhathi sangempela , avumela ukuthi ikwazi ukusingatha imisebenzi ehlukene ngendlela eguquguqukayo futhi ingasathembeli kudatha eqinile yangaphambi kokuqeqeshwa. Lokhu kuyinguquko ekuphetheni.

Ngemuva kwegama elisha: Kungani kungenjalo i-GPT-5?

Ngokuthakazelisayo, i-OpenAI yanquma ukushiya igama elijwayelekile elithi "GPT" yasebenzisa i-o1 esikhundleni salokho . Lesi kwakungesona isinqumo esingahleliwe. I-OpenAI ikholelwa ukuthi i-o1 imelela ukuqala kwe -inference paradigm entsha ngokuphelele , kuyilapho uchungechunge lwangaphambilini lwe-GPT lwalungolwe-"paradigm eqeqeshwe kusengaphambili".

Kungani kunjalo? Isizathu silula: I-O1 ayisathembeli kudatha eqeqeshwe kusengaphambili ; ichitha isikhathi esiningi icabanga, isondela ekucabangeni kwabantu ngesikhathi sangempela. Uma iqhathaniswa ne-GPT-4, ingaxazulula izinkinga eziyinkimbinkimbi ngamakhono aqinile okunengqondo, ikakhulukazi emikhakheni yesayensi, yezibalo, kanye neyobunjiniyela.

Umehluko omkhulu emihlanu phakathi kwe-o1 ne-GPT-4o

Sesiyazi kakade ukuthi ikhono lokucabanga lika-o1 alinakuqhathaniswa, ngakho yiziphi ezinye izinzuzo enazo?

1. Intuthuko enkulu ekhonweni lokucabanga

Uma i-GPT-4 isazama ukuxazulula inkinga ye-inference, khona-ke u-o1 usevele weqa lesi sithiyo. i-o1 inika amandla ukubikezela kwesikhathi sangempela kunokuthembela kudatha yokuqeqeshwa yangaphambilini. Ngenxa yalokhu, yenza kangcono emisebenzini eyinkimbinkimbi yezibalo, isayensi, kanye nekhodi.

2. Okunzima kakhulu ku-jailbreak

Njengoba ukwanda kwe-LLM kukhula, ukuphepha sekuyinto ebaluleke kakhulu. Uchungechunge lwe-o1 kunzima kakhulu ukuluqeda kunamamodeli angaphambilini . Ngokusho kokuhlolwa kokuphepha, ukubuka kuqala kwe-o1 kuthole amaphuzu angu-84 kwangu-100, kuyilapho i-GPT-4o ithole amaphuzu angu-22 kuphela. Kubasebenzisi bebhizinisi nabathuthukisi, lokhu kuholela kokuhlangenwe nakho okuphephile nokuthembekile.

3. Imithetho emisha yokuqamba amagama

Isinqumo se-OpenAI sokusebenzisa u-"o1" ekuqambeni kwayo sibonisa indawo yayo entsha . Lokhu akusikho nje ukushintsha igama, kodwa ukuguqulwa kwendlela yokucabanga ye-AI yonke. Ungakucabanga njengokugxuma kusuka "kudatha eqeqeshwayo engaguquki" kuya "ekucabangeni kwesikhathi sangempela okuguquguqukayo."

4. Ukuvezwa kokugcina kobuchwepheshe, izibalo kanye nobunjiniyela

Ikhono lokucabanga elithuthukisiwe lenze i-O1 yakwazi ukuphumelela ezibalweni nasesayensini. Isibonelo, kwi-International Mathematical Olympiad, i-GPT-4 yaxazulula izinkinga ezingu-13% kuphela, kanti i-O1 yaxazulula ezingu-83% . Lokhu kuyigxathu elibalulekile!

5. Ukwelulwa kwesikhathi sokulinda

Ngenxa yokuthi i-o1 idinga ukuqagela ngesikhathi sangempela, isikhathi sayo sokuphendula side kancane kunamamodeli angaphambilini. Nakuba lokhu kungase kubakhungathekise abanye abasebenzisi, lokhu kushintshisana kwesikhathi sokukwazi ukuqagela kufanelekile impela ukulinda. I-OpenAI iphinde yathi izothuthukisa isivinini esikhathini esizayo.

Ubani ongasebenzisa i-o1?

Cishe usuvele ulangazelela ukuzwa i-o1, akunjalo? Izindaba ezinhle ukuthi kusukela ngoSepthemba 12, 2024, abasebenzisi be-ChatGPT Plus kanye nethimba bazokwazi ukufinyelela ku- o1-preview.

Kubasebenzisi abajwayelekile, nakuba i-o1-mini ingavuliwe okwamanje, kulindeleke ukuthi ivuleleke kuwo wonke umuntu esikhathini esizayo esiseduze.

Njengamanje, imikhawulo yokusetshenziswa yamasonto onke yokubuka kuqala kwe-o1 kanye ne-o1-mini yizinto ezingu-30 nezingu-50 ngokulandelana kwazo, kodwa lo mkhawulo uzokwenyuswa maduze. Singalindela ukuthi ngokuthuthukiswa okuzayo, i-o1 izothandwa kakhulu.

Yini okufanele ngiyisebenzise u-o1?

Amakhono okucabanga e-o1 afaneleka kakhulu emisebenzini eyinkimbinkimbi edinga izibalo ezinembile, ikakhulukazi kwizibalo, isayensi, kanye nokubhala ikhodi . I-OpenAI igcizelela ukuthi i-o1 ingasetshenziswa ukuxazulula amafomula ezibalo ayinkimbinkimbi kwifiziksi noma ekuhlaziyweni kwedatha yeselula ocwaningweni lwezempilo.

Uma ungunjiniyela, u-o1 ungumsizi wakho omkhulu. Ingakusiza ukuthi ubhale ikhodi eyinkimbinkimbi futhi uklame ukuhamba komsebenzi okunezinyathelo eziningi.

Imikhawulo ye-o1

Yiqiniso, ayikho imodeli ephelele. Nakuba i-o1 ihlaba umxhwele, isewukubuka kuqala futhi ayinazo zonke izici ze-GPT-4o.

Uma udinga ukuphequlula iwebhu noma ulayishe amafayela nezithombe, i-o1 ayikayisekeli le misebenzi okwamanje.

Kodwa-ke, njengoba i-OpenAI iqhubeka nokuthuthukiswa, izinguqulo zesikhathi esizayo zingakwazi ukuvala lezi ziphutha.

Amathiphu neziphakamiso: Ungayisebenzisa kanjani i-o1 kangcono?

Ngokungafani namamodeli wangaphambilini, i-o1 ihlakaniphile, ngakho-ke awusadingi ukufaka imiyalo eyinkimbinkimbi. Ngemiyalo elula necacile kuphela, i-o1 ingaqonda izidingo zakho nesizathu. Ngakolunye uhlangothi, isiqondiso esiningi kakhulu sizonciphisa isivinini sokucubungula.

Ekuphetheni

I-OpenAI o1 iyinguquko ekuqondeni kwe-AI. Akuyona nje inguqulo ethuthukisiwe ye-GPT-4, kodwa iyindlela entsha ngokuphelele yokucabanga . Ngamakhono ayo okucabanga ngesikhathi sangempela kanye nokusebenza okuvelele kwezibalo nesayensi, i-o1 isivele ibonise amandla ayo angavamile emikhakheni eminingi.

Uma udinga ukuxazulula izinkinga zezibalo eziyinkimbinkimbi, bhala ikhodi ephumelelayo, noma wenze izivivinyo zesayensi, i-o1 iyisinqumo sakho esingcono kakhulu. Futhi, ngokwethulwa kwamamodeli amaningi esikhathini esizayo, i-o1 izoba namandla, ihlakaniphe futhi kube lula ukuyisebenzisa. Ingabe usukulungele ukwamukela inkathi entsha ye-o1?

Isifingqo: Kungani i-o1 ibaluleke kangaka?

• Ukuphumelela ekhonweni lokucabanga: Ukucabanga kwesikhathi sangempela kufinyelelwa ngokokuqala ngqa, eduze nokucabanga komuntu.
• Ukuthuthukiswa kokuphepha: Kunzima ukuboshwa, ukukhetha okungcono kakhulu kwamabhizinisi nabathuthukisi.
• Umhlangano omusha wokuqamba amagama: Vula i-paradigm entsha yokucabanga bese uthi sala kahle kumodi yokuqeqeshwa kwangaphambili.
• Abaholi beMathematika neSayensi: Ukusebenza okungenakuqhathaniswa, ikakhulukazi emiqhudelwaneni yama-Olympic.

Ungalindi, zama i-OpenAI o1 manje ! Kuyingqophamlando ye-AI ongafuni ukuyiphuthelwa.

Uma ubhalisa i-OpenAI ezweni laseChina, ngokushesha "OpenAI's services are not available in your country."▼

Ngenxa yokuthi imisebenzi ethuthukisiwe idinga abasebenzisi ukuthi bathuthukele ku-ChatGPT Plus ngaphambi kokuthi isetshenziswe, kunzima ukwenza i-ChatGPT Plus isebenze emazweni angayisekeli i-OpenAI, futhi badinga ukubhekana nezinkinga ezinzima njengamakhadi esikweletu abonakalayo angaphandle...

Lapha sethula kuwe iwebhusayithi ethengeka kakhulu ehlinzeka ngama-akhawunti okuqasha okwabelwana ngawo e-ChatGPT Plus.

Sicela uchofoze isixhumanisi esingezansi ukuze ubhalise i-akhawunti yePlanet ▼

Chofoza isixhumanisi esingezansi ukuze uthole umhlahlandlela onemininingwane wokubhalisa i-akhawunti yakho ku-Planet▼

🔗Xhumanisa ne-ChatGPT Plus ukuthenga nge-carpool yabantu abaningi kwi-akhawunti yeplanethi, thola isaphulelo esingu-50% emalini yenyanga!

Amathiphu:

• Amakheli e-IP e-Russia, e-China, e-Hong Kong, nase-Macau awakwazi ukubhalisela i-akhawunti ye-OpenAI. Kunconywa ukuthi ubhalise elinye ikheli le-IP.