Yintoni i-OpenAI o1-preview? Intshayelelo ebanzi yohlalutyo lokuthelekisa kunye ne-GPT-4o

Imodeli yolwimi enkulu yamva nje ye -OpenAI o1: ukungenisa ixesha elitsha lokuqiqa.

Njengoko usenokuba uvile, emva kweenyanga zokuqikelela, i-OpenAI ikhuphe inguqulelo yayo yamva nje yomzekelo wayo wolwimi olukhulu ngoSeptemba 12, 2024 .

Okumangalisayo kukuba, le modeli intsha ayibizwa ngokuba yiGPT-5 . Endaweni yoko, ikhutshwe ngegama elilula nelithe ngqo: o1 . Kwenzeke ntoni ngasemva kweziganeko? Masifumanise!

Yintoni u-o1?

I-OpenAI o1 yimodeli yamva nje yolwimi olukhulu lwe-OpenAI, ekhutshwe nge-12 kaSeptemba 2024. Okumangalisayo kukuba, iyahluka kwindlela yokubiza amagama yangaphambili ye-"GPT", ikhetha ilebheli entsha ngokupheleleyo endaweni yoko. Kuthetha ukuthini oku? Ithetha ukuba ayisiyo nje ukuphuculwa kwi-GPT-4, kodwa sisiqalo sendlela entsha yokwenza izinto.

I-OpenAI ikhuphe iimodeli ezimbini ze-o1: i-o1-preview kunye ne -o1-mini . Babonakalise ubuchule obuqinileyo ekuqiqeni, kwizibalo, nakwisayensi , bedlula ngokuthe ngqo i -GPT-4o yangaphambili , ingakumbi kubuchule bokuqiqa, apho bafikelele kwinqanaba elikhulu phambili.

Ubuchule bokuqiqa yeyona nto iphambili kuthotho lwe-O1. I-OpenAI igxininisa ukuba ifana kakhulu nokucinga komntu, ekwaziyo ukuqiqa ngexesha langempela kunokuthembela kuphela kwidatha eqeqeshwe kwangaphambili. Oku kwahluke kakhulu kwiimodeli zangaphambili! Ngaba unomdla wokuba amandla okuqiqa e-O1 anamandla kangakanani? Qhubeka ufunda ukuze ufumane impendulo.

Ukukwazi kuka-o1 ukuqiqa: impumelelo engaphaya kwexesha elidlulileyo

Usenokuba ucinga, "Ngaba izakhono zokuqiqa zibaluleke ngokwenene?" Kungenxa yoko le nto i-o1 ikwazile ukufezekisa iziphumo ezimangalisayo kwiinkalo ezininzi.

1. Ukusebenza kwisayensi nakwimathematika : o1 Kwiimvavanyo ezininzi zesayensi, iqhube phantse ngokulinganayo nabafundi bezifundo zobugqirha. Kwiinkalo ezifana nefiziksi, ikhemistri, kunye nebhayoloji, ifikelele kwinqanaba eliphezulu lempumelelo kwezemfundo.

2. Iziphumo ze-US Mathematical Olympiad : U-O1 ufikelele ngokulula phakathi kwama-500 aphezulu kolu khuphiswano lunzima kakhulu. Khawuthelekelele ukhuphisana nabanye babantu abaselula abakrelekrele kwihlabathi, kwaye u-O1, oyimodeli nje, uphumelela ngokulula kubo.

3. Impumelelo yokubhala iikhowudi : i-o1 ibekwe kwindawo yama-89 kwiqonga lekhowudi elidumileyo kwihlabathi liphela iCodeforces , nto leyo imangalisayo! Ayinakusombulula iingxaki zezibalo kuphela, kodwa ikwabhala nekhowudi enzima.

Konke oku kungenxa yamandla e-O1 okuqonda izinto ngexesha langempela , avumela ukuba ikwazi ukusingatha imisebenzi eyahlukeneyo ngokulula kwaye ingasaxhomekeki kwidatha eqinileyo yangaphambi koqeqesho. Olu luguquko kwindlela yokuqikelela izinto.

Ngasemva kwegama elitsha: Kutheni i-GPT-5?

Okubangel’ umdla kukuba, i-OpenAI igqibe ekubeni ilishiye igama eliqhelekileyo elithi "GPT" ize isebenzise i-o1 endaweni yoko . Esi yayingesosigqibo esenziwe ngokungacwangciswanga. I-OpenAI ikholelwa ukuba i-o1 imele ukuqala kwe -inference paradigm entsha ngokupheleleyo , ngelixa uthotho lwangaphambili lwe-GPT lwalukwi-"pre-trained paradigm".

Kutheni kunjalo? Isizathu silula: I-O1 ayisaxhomekeki kwidatha eqeqeshwe kwangaphambili ; ichitha ixesha elininzi icinga, isondela ekuqiqeni kwabantu ngexesha langempela. Xa ithelekiswa ne-GPT-4, inokusombulula iingxaki ezinzima ngezakhono eziqiqileyo, ngakumbi kwicandelo lesayensi, lezibalo, kunye nelobunjineli.

Iiyantlukwano ezinkulu ezintlanu phakathi kwe-o1 kunye ne-GPT-4o

Sele sisazi ukuba i-o1 yokuqiqa ayinakuthelekiswa nanto, zeziphi ezinye iingenelo enazo?

1. Uphuculo olukhulu kubuchule bokuqiqa

Ukuba i-GPT-4 isazama ukusombulula ingxaki ye-inference, ngoko i-o1 sele iwele lo mqobo. I-o1 yenza ukuba kuthengiswe ngexesha lokwenyani kunokuxhomekeka kwidatha yoqeqesho lwangaphambili. Ngenxa yoku, iqhuba ngcono kwizibalo ezintsonkothileyo, kwisayensi, nakwimisebenzi yokukhowuda.

2. Kunzima ngakumbi kwi-jailbreak

Ngenxa yokwanda kwe-LLM, ukhuseleko luye lwaba yinto ebalulekileyo. Uthotho lwe-o1 lunzima kakhulu ukuluqhekeza kuneemodeli zangaphambili . Ngokweemvavanyo zokhuseleko, i-o1-preview ifumene amanqaku angama-84 kwi-100, ngelixa i-GPT-4o ifumene amanqaku angama-22 kuphela. Kubasebenzisi beshishini kunye nabaphuhlisi, oku kuthetha ukuba ngamava akhuselekileyo nanokuthenjwa.

3. Imigaqo emitsha yamagama

Isigqibo se-OpenAI sokusebenzisa u-"o1" ekubizeni kwayo sibonisa indawo yayo entsha . Oku akukokutshintsha nje igama, kodwa kukutshintsha kwengqiqo yokuqiqa ye-AI yonke. Ungacinga ngayo njengokutsiba ukusuka "kwidatha eqeqeshwe kwangaphambili engaguqukiyo" ukuya "ekucingeni ngexesha langempela eliguqukayo."

4. Eyona ndlela yokubonakalisa iteknoloji, imathematika nobunjineli

Ubuchule bokuqiqa obuphuculweyo buvumele i-O1 ukuba iphumelele kwizibalo nakwisayensi. Umzekelo, kwi-International Mathematical Olympiad, i-GPT-4 isombulule kuphela i-13% yeengxaki, ngelixa i-O1 isombulule i-83% . Olu lugqatso olusemgangathweni!

5. Ukwandiswa kwexesha lokulinda

Ngenxa yokuba i-o1 ifuna uqikelelo lwexesha langempela, ixesha layo lokuphendula lide kancinci kuneemodeli zangaphambili. Nangona oku kunokucaphukisa abanye abasebenzisi, olu tshintsho lwexesha lobuchule bovavanyo lufanelekile ukulinda. I-OpenAI ikwathe ukuba iya kuphucula isantya kwixesha elizayo.

Ngubani onokusebenzisa u-o1?

Mhlawumbi sele unomdla wokufumana amava e-o1, akunjalo? Iindaba ezimnandi zezokuba ukususela nge-12 kaSeptemba 2024, abasebenzisi beChatGPT Plus kunye neqela baya kuba nokufikelela kwi -o1-preview.

Kubasebenzisi abaqhelekileyo, nangona i-o1-mini ingavulwanga okwangoku, kulindeleke ukuba ivuleleke kuye wonke umntu kungekudala.

Okwangoku, imida yokusetyenziswa kweveki ye-o1-preview kunye ne-o1-mini zizinto ze-30 kunye ne-50 ngokulandelanayo, kodwa lo mda uya kwandiswa ngokukhawuleza. Sinokulindela ukuba ngokuphuculwa kwexesha elizayo, i-o1 iya kuthandwa ngakumbi.

Kufuneka ndiyisebenzisele ntoni i-o1?

Amandla okuqiqa e-o1 afaneleka kakhulu kwimisebenzi enzima efuna ukubalwa ngokuchanekileyo, ngakumbi kwizibalo, isayensi, kunye nokubhala iikhowudi . I-OpenAI igxininisa ukuba i-o1 ingasetyenziselwa ukusombulula iifomyula zezibalo ezintsonkothileyo kwifiziksi okanye kuhlalutyo lwedatha yeselula kuphando lwezempilo.

Ukuba ungumphuhlisi, u-o1 ngumncedisi wakho omkhulu. Inokukunceda ukuba ubhale ikhowudi entsonkothileyo kwaye ude uyilo lokuhamba kwamanyathelo amaninzi.

Unyino lwe-o1

Ngokuqinisekileyo, akukho modeli igqibeleleyo. Ngelixa i-o1 inomtsalane, iseyimboniso yangaphambili kwaye ayinazo zonke iimpawu ze-GPT-4o.

Ukuba ufuna ukukhangela iwebhu okanye ulayishe iifayile kunye nemifanekiso, i-o1 ayikayixhasi le misebenzi okwangoku.

Nangona kunjalo, njengoko i-OpenAI iqhubeka nokulungiswa, iinguqulelo zexesha elizayo ziyakwazi ukuzenzela ezi ntsilelo.

Iingcebiso kunye neengcebiso: Indlela yokusebenzisa i-o1 engcono?

Ngokungafaniyo neemodeli zangaphambili, i-o1 ihlakaniphile, ngoko akusekho mfuneko yokuba ufake ii-prompt enzima. Ngemiyalelo elula necacileyo kuphela, i-o1 inokuqonda iimfuno zakho kunye nesizathu. Kwelinye icala, ukhokelo oluninzi luya kucothisa isantya sokusebenza kwayo.

ukuququmbela

I-OpenAI o1 yinguqu enkulu kwi-AI inference. Ayisiyo nje inguqulelo ephuculweyo ye-GPT-4, kodwa yindlela entsha ngokupheleleyo yokuqonda . Ngezakhono zayo zokuqonda ngexesha langempela kunye nokusebenza okugqwesileyo kwezibalo kunye nesayensi, i-o1 sele ibonakalise amandla ayo angaqhelekanga kwiinkalo ezininzi.

Ukuba ufuna ukusombulula iingxaki zezibalo ezintsonkothileyo, bhala ikhowudi esebenzayo, okanye wenze imifuniselo yezenzululwazi, o1 lolona khetho lwakho lulungileyo. Kwaye, ngokusungulwa kweemodeli ezininzi kwixesha elizayo, i-o1 iya kuba namandla ngakumbi, ihlakaniphile kwaye kulula ukuyisebenzisa. Ngaba ukulungele ukwamkela ixesha elitsha le-o1?

Isishwankathelo: Kutheni i-o1 ibaluleke kangaka?

• Ukuphumelela ekuqiqeni: Ukuqiqa kwexesha langempela kufezekiswa okokuqala, kufutshane nokucinga komntu.
• Ukuphuculwa kokhuseleko: Kunzima ngakumbi ukuba i-jailbroken, olona khetho lungcono kumashishini kunye nabaphuhlisi.
• Indibano entsha yamagama: Vula i-paradigm entsha yokuqiqa kwaye uthi ndlela ntle kwindlela yoqeqesho lwangaphambili.
• Iinkokeli kwiMathematika neSayensi: Ukusebenza okungenakulinganiswa, ngakumbi kukhuphiswano lweOlimpiki.

Musa ukulinda, zama i-OpenAI o1 ngoku ! Lilinyathelo le-AI ongafuni ukuliphosa.

Ukuba ubhalisa i-OpenAI kwilizwe laseTshayina, ngokukhawuleza "OpenAI's services are not available in your country."▼

Imisebenzi ephucukileyo ifuna ukuba abasebenzisi baphucule ukuya kwi-ChatGPT Plus ngaphambi kokuba isetyenziswe, nangona kunjalo, kumazwe angayixhasi i-OpenAI, kunzima ukwenza i-ChatGPT Plus isebenze, kwaye kufuneka ujongane nemiba enzima efana namakhadi etyala angaphandle.

Apha sikwazisa ngewebhusayithi efikelelekayo kakhulu ebonelela ngeChatGPT Plus ngeeakhawunti ezabelwana ngazo zokurenta.

Nceda ucofe ikhonkco elingezantsi ukuze ubhalise iakhawunti yePlanet ▼

Cofa ikhonkco elingezantsi ukuze ufumane isikhokelo esineenkcukacha zokubhalisa iakhawunti yakho kwiPlanet▼

🔗Ikhonkco lokuthenga i-ChatGPT Plus nge-carpooling kwi-akhawunti yakho yeplanethi, fumana isaphulelo se-50% kwiintlawulo zenyanga!

Iingcebiso:

• Iidilesi ze-IP eRashiya, eChina, eHong Kong, naseMacau azinakubhalisela i-akhawunti ye-OpenAI. Kuyacetyiswa ukuba ubhalise kwenye idilesi ye-IP.