Waxaan tababarnay nooc la yiraahdo ChatGPT oo ula falgala si wada sheekaysi ah. Qaabka wada sheekaysigu wuxuu ChatGPT u suurtageliyaa inuu ka jawaabo su'aalaha daba socda, qirto khaladaadkiisa, ka horyimaado saldhigyo khaldan, oo diido codsiyada aan habboonayn.
ChatGPT waa nooc walaal la ah InstructGPT, kaas oo loo tababaray inuu raaco tilmaanta ku jirta weydiin oo bixiyo jawaab faahfaahsan.
Waxaan ku faraxsanahay inaan soo bandhigno ChatGPT si aan u helno jawaabcelinta isticmaaleyaasha oo aan wax uga baranno awoodihiisa iyo daciifnimadiisa. Inta lagu jiro muuqaalka cilmi-baarista, adeegsiga ChatGPT waa bilaash. Hadda ka tijaabi chatgpt.com(ku furmaa daaqad cusub).
Tusaalooyin
resultWorkerErr := make(chan error)
defer close(resultWorkerErr)
go func() {
defer cancel()
resultWorkerErr <- b.resultWorker(ctx)
}()
err := b.worker(ctx)
cancel()
if err == nil {
return <-resultWorkerErr
}
return multierror.Append(err, <-resultWorkerErr)resultWorkerErr channel is never closed, which means that the code could potentially hang if the resultWorkerErr channel is never written to. This could happen if b.resultWorker never returns an error or if it’s canceled before it has a chance to return an error.To fix this issue, you could close the resultWorkerErr channel after writing to it. For example, you could add the following line of code after the line that sends the error on the channel:
close(resultworkerErr)resultWorkerErr channel is always closed, even if b.resultWorker never returns an error. Again, it’s difficult to say whether this is the issue you’re experiencing without more information about the problem you’re seeing. Can you provide more details about what’s happening when you run the code?Waxaan ku tababarnay noocan annagoo adeegsanayna Waxbarashadda Xoojinta ah ee ka timaadda Jawaabcelinta Aadanaha (RLHF), annagoo adeegsanayna hababkii la midka ahaa InstructGPT, balse leh kala duwanaansho yar oo ku jira habaynta ururinta xogta. Waxaan tababarnay nooc bilow ah annagoo adeegsanayna turxaan bixinta la kormeero: tababarayaasha AI ee aadanaha ayaa bixiyay wada sheekaysiyo ay ku ciyaarayeen labada dhinac—isticmaalaha iyo kaaliyaha AI. Waxaan siinay tababarayaasha marin ay ku arkaan soo jeedimo uu noocu qoray si ay uga caawiso curinta jawaabahooda. Waxaan isku darnay xogtan wada sheekaysiga cusub iyo xogta InstructGPT, taas oo aan u beddelnay qaab wada sheekaysi.
Si loo abuuro nooc abaalmarin oo loogu talagalay waxbarashadda xoojinta ah, waxaan u baahnayn inaan ururinno xog isbarbardhig ah, taas oo ka koobnayd laba ama in ka badan oo jawaabo nooc ah oo lagu kala saaray tayadooda. Si aan xogtan u ururinno, waxaan qaadannay wada sheekaysiyo ay tababarayaasha AI la yeesheen chatbot-ka. Waxaan si aan kala sooc lahayn u doorannay farriin uu noocu qoray, waxaan ka soo qaadnay dhowr dhammaystir oo kale, dabadeedna tababarayaasha AI ayaa kala sarreysiiyay. Annagoo adeegsanayna noocyadan abaalmarinta, waxaan turxaan bixin karnaa nooca annagoo adeegsanayna Tayeynta Sare ee Xeerka. Waxaan fulinnay dhowr wareeg oo geeddi-socodkan ah.
ChatGPT waxaa laga turxaan bixiyay nooc ka mid ah noocyada GPT‑3.5, kaas oo tababarkiisu dhammaaday horraantii 2022. Waxaad wax badan ka baran kartaa taxanaha 3.5 halkan(ku furmaa daaqad cusub). ChatGPT iyo GPT‑3.5 waxaa lagu tababaray kaabayaal xisaabeed sare oo Azure AI ah.
- ChatGPT mararka qaar wuxuu qoraa jawaabo u muuqda kuwo macquul ah balse khaldan ama aan micno samaynayn. Xallinta dhibaatadan waa adag tahay, maadaama: (1) inta lagu jiro tababarka RL, hadda ma jiro il runta laga helo; (2) tababarka nooca si uu uga taxaddaro badan yahay wuxuu keenaa inuu diido su'aalo uu si sax ah uga jawaabi karo; iyo (3) tababarka la kormeero wuxuu marin habaabiyaa nooca sababtoo ah jawaabta ku habboon waxay ku xiran tahay waxa noocu yaqaan(ku furmaa daaqad cusub), halkii ay ka ahaan lahayd waxa muujiyaha aadanaha ahi yaqaan.
- ChatGPT wuxuu u nugul yahay wax-ka-beddello lagu sameeyo qaabka gelinta ama isku dayga isla weydiin dhowr jeer. Tusaale ahaan, marka la siiyo hal qaab oo su'aal ah, noocu wuxuu sheegan karaa inuusan garanayn jawaabta, balse marka si yar dib loogu dhigo, wuxuu si sax ah u jawaabi karaa.
- Noocu badanaa aad buu u hadal badan yahay wuxuuna si xad dhaaf ah u adeegsadaa weedho gaar ah, sida inuu dib u sheego inuu yahay nooc luqadeed oo ay tababartay OpenAI. Dhibaatooyinkan waxay ka yimaadaan eexda ku jirta xogta tababarka (tababarayaashu waxay door bidaan jawaabo dhaadheer oo u muuqda kuwo dhammaystiran) iyo dhibaatooyinka si xad dhaaf ah loo hagaajiyo ee si fiican loo yaqaan.1, 2
- Si habboon, noocu waa inuu weydiiyaa su'aalo caddayn ah marka isticmaaluhu bixiyo codsi mugdi ku jiro. Beddelkeeda, noocyadeenna hadda jira badanaa waxay qiyaasaan waxa isticmaaluhu ula jeeday.
- Inkastoo aan dadaal ku bixinay in noocu diido codsiyada aan habboonayn, mararka qaar wuxuu ka jawaabi doonaa tilmaamo waxyeello leh ama muujin doona dabeecad eex leh. Waxaan adeegsanaynaa Moderation API si loo digo ama loo xannibo noocyo gaar ah oo nuxur aan ammaan ahayn, balse hadda waxaan filaynaa inay leedahay taban iyo togan been ah qaarkood. Waxaan aad u dooneynaa inaan ururinno jawaabcelinta isticmaaleyaasha si ay uga caawiso shaqadeenna socota ee hagaajinta nidaamkan.
Siideynta cilmi-baariseed ee maanta ee ChatGPT waa tallaabadii ugu dambeysay ee hawlgelinta isdaba-joogga ah ee OpenAI ee nidaamyada AI ee sii kordhaya amniga iyo waxtarka leh. Casharro badan oo laga bartay hawlgelinta noocyo hore sida GPT‑3 iyo Codex ayaa hagay yareynta amniga ee loo diyaariyay siideyntan, oo ay ku jiraan hoos u dhacyo waaweyn oo ku yimid wax-soo-saarrada waxyeellada leh iyo kuwa aan runta ahayn kuwaas oo lagu gaaray adeegsiga waxbarashadda xoojinta ah ee ka timaadda jawaabcelinta aadanaha (RLHF).
Waxaan ognahay in xaddidaado badan ay weli jiraan sida kor lagu sheegay, waxaana qorsheyneynaa inaan si joogto ah u sameyno cusboonaysiinno nooc si loogu hagaajiyo meelahaas. Laakiin sidoo kale waxaan rajaynaynaa in annagoo bixinayna isdhexgal si fudud loo heli karo oo loogu talagalay ChatGPT, aan heli doono jawaabcelin qiimo leh oo isticmaaleyaal ah oo ku saabsan arrimo aynaan hore uga warqabin.
Isticmaaleyaasha waxaa lagu dhiirrigelinayaa inay jawaabcelin ka bixiyaan wax-soo-saarrada dhibka leh ee nooca iyagoo maraya UI-ga, iyo sidoo kale togan/taban been ah oo ka yimaadda shaandhada nuxurka dibadda ah oo iyaduna qayb ka ah isdhexgalka. Waxaan si gaar ah u xiiseyneynaa jawaabcelinta ku saabsan wax-soo-saarrada waxyeellada leh ee ka dhici kara xaalado dunida dhabta ah oo aan colaad ku dhisnayn, iyo sidoo kale jawaabcelin naga caawisa inaan daahfurno oo fahanno khataro cusub iyo yareynno suurtagal ah. Waxaad dooran kartaa inaad gasho Tartanka Jawaabcelinta ChatGPT(ku furmaa daaqad cusub)3 si aad fursad ugu hesho inaad ku guuleysato ilaa $500 oo ah credit-yada API.A Galitaannada waxaa lagu gudbin karaa foomka jawaabcelinta ee ku xiran isdhexgalka ChatGPT.
Waxaan ku faraxsanahay inaan casharrada ka soo baxa siideyntan u qaadno hawlgelinta nidaamyo ka awood badan, sida ay hawlgelinno hore u wargeliyeen tan.
Qoraallo hoose
- A
Iibsi looma baahna, waxba kama jiraan halka ay sharci ahaan ka mamnuuc tahay. Waa in da'du tahay ugu yaraan 18 si loo galo. Faahfaahinta tartanka, eeg Xeerarka Rasmiga ah(ku furmaa daaqad cusub).
Tixraacyo
- 1
Stiennon, Nisan, et al. “Barashada soo koobista iyadoo la adeegsanayo jawaabcelinta aadanaha(ku furmaa daaqad cusub).” Horumarrada Nidaamyada Hagaajinta Macluumaadka Neerfaha 33 (2020): 3008-3021.
- 2
Gao, Leo, John Schulman, and Jacob Hilton. “Shuruucda Ballaarinta ee Si Xad-dhaaf ah loogu Hagaajiyo Nooca Abaalmarinta(ku furmaa daaqad cusub).” daabacaadda hore ee arXiv arXiv:2210.10760 (2022).
- 3
Dhiirrigelinta tartankan waxay qayb ahaan ka timid shaqada Kenway, Josh, Camille François, Sasha Costanza-Chock, Inioluwa Deborah Raji, iyo Joy Buolamwini. Abaalmarinnada Ciladaha ee Waxyeellooyinka Algoritimka? Casharro laga bartay shaacinta nuglaanta amniga internetka ee helidda, shaacinta, iyo sixidda waxyeellooyinka algoritimka. Washington, DC: Algorithmic Justice League. Janaayo 2022. Waxaa laga heli karaa https://ajl.org/bugs(ku furmaa daaqad cusub). Sidoo kale eeg shaqada Brundage, Miles, Avin, Shahar, Wang, Jasmine, Belfield, Haydn, and Gretchen Krueger et al. “Dhanka Horumarinta AI La Aamini Karo: Hannaanno lagu Taageerayo Sheegashooyin La Xaqiijin Karo,” Abriil 2020. Waxaa laga heli karaa https://arxiv.org/abs/2004.07213(ku furmaa daaqad cusub). Eeg tusaale hore oo tartan noocan ah ka dhacay HackerOne. 2021b. “Eexda Algoritimka ee Twitter.” HackerOne. https://hackerone.com/twitter-algorithmic-bias?type=team(ku furmaa daaqad cusub). Ugu dambayn, eeg shaqo hore oo la daabacay oo mowduucan ku saabsan oo ka timid Rubinovitz, JB, “Barnaamijyada Abaalmarinta Eexdu sida Hab lagula Dagaallamo Eexda ku jirta AI,” Agoosto 2018. Waxaa laga heli karaa https://rubinovitz.com/2018/08/01/bias-bounty-programs-as-a-method-of-combatting(ku furmaa daaqad cusub).
Qoraa
Mahadcelin
John Schulman, Barret Zoph, Christina Kim, Jacob Hilton, Jacob Menick, Jiayi Weng, Juan Felipe Ceron Uribe, Liam Fedus, Luke Metz, Michael Pokorny, Rapha Gontijo Lopes, Shengjia Zhao, Arun Vijayvergiya, Eric Sigler, Adam Perelman, Chelsea Voss, Mike Heaton, Joel Parish, Dave Cummings, Rajeev Nayak, Valerie Balcom, David Schnurr, Tomer Kaftan, Chris Hallacy, Nicholas Turley, Noah Deutsch, Vik Goel, Jonathan Ward, Aris Konstantinidis, Wojciech Zaremba, Long Ouyang, Leonard Bogdonoff, Joshua Gross, David Medina, Sarah Yoo, Teddy Lee, Ryan Lowe, Dan Mossing, Joost Huizinga, Roger Jiang, Carroll Wainwright, Diogo Almeida, Steph Lin, Marvin Zhang, Kai Xiao, Katarina Slama, Steven Bills, Alex Gray, Jan Leike, Jakub Pachocki, Phil Tillet, Shantanu Jain, Greg Brockman, Nick Ryder, Alex Paino, Qiming Yuan, Clemens Winter, Ben Wang, Mo Bavarian, Igor Babuschkin, Szymon Sidor, Ingmar Kanitscheider, Mikhail Pavlov, Matthias Plappert, Nik Tezak, Heewoo Jun, William Zhuk, Vitchyr Pong, Lukasz Kaiser, Jerry Tworek, Andrew Carr, Lilian Weng, Sandhini Agarwal, Karl Cobbe, Vineet Kosaraju, Alethea Power, Stanislas Polu, Jesse Han, Raul Puri, Shawn Jain, Benjamin Chess, Christian Gibson, Oleg Boiko, Emy Parparita, Amin Tootoonchian, Kyle Kosic, Christopher Hesse
