Thekiniki Leyintshwa Yo Hlohlotela Yi Vula Leswaku Yi Pfuna AI Ku Ehleketa Ku Fana Ni Vanhu

Xana u tshame u tivutisa leswaku u nga endla njhani leswaku AI a ehleketa ku fana ni munhu? Hambileswi Timodelo ta Ririmi letikulu (LLM) ti humelelaka eku vhumbheni ka rito leri landzelaka hi ku landzelelana, ti tala ku khunguvanyeka loko ti langutane na swiphiqo leswi lavaka ku ehleketa hi ndlela ya maendlelo. Nghena eka ku hlohlotela ka Chain-of-Thought – thekiniki yo cinca ntlangu leyi hundzulaka ndlela leyi hi tirhisanaka ha yona na AI.

Ku Pfuxeta Nxaxamelo Wa Miehleketo

Anakanya u dyondzisa n’wana ku tlhantlha xiphiqo lexi rharhanganeke. Ematshan’weni yo va komba xifaniso xo hetelela, u va kongomisa eka goza rin’wana ni rin’wana. Sweswo hi xiviri leswi ku hlohloteriwa ka Chain-of-Thought (CoT) swi swi endlaka eka ti-LLM. Hi ku nyika swikombiso leswi kombisaka ku anakanya hi magoza hi magoza, hi pfuna timodeli leti ku fikelela swintshuxo leswi kongomeke hi ku tirhisa nhluvuko lowu twisisekaka.

Fomula Ya Masalamusi: Ha Yini Yi Tirha

Maendlelo ya ndhavuko yo hlohlotela hakanyingi ya yisa eka mbuyelo wa ku ba kumbe ku hundza. CoT prompting, hambiswiritano, yi avanyisa swiphiqo leswi rharhanganeke hi swiphemu leswi gayekaka, leswi pfumelelaka modele ku langutana na xiphemu xin’wana na xin’wana hi ndlela leyi hlelekeke. Endlelo leri ri tikombise ri humelela swinene lerova timodeli ta xiyimo xa le henhla to fana na tihakelo ta sweswinyana ta OpenAI ti ri nghenise eka ntirho wa tona wa nkoka.

Ku Tirhisa ka Misava Ya Xiviri: Ku tlhantlha Tiphazili ta Tinhlayo

A hi ngheneni eka xikombiso lexi tirhaka. Anakanya u pfuna AI ku tlhantlha xiringaniso lexi

5x - 4 = 16.

Hi leyi ndlela leyi CoT prompting yi kongomisaka modele ha yona:

1. Goza ro sungula: Engeta 4 eka matlhelo hamambirhi
   • 5x - 4 + 4 = 16 + 4. Xitsonga
   • 5x = 20
2. Goza ra vumbirhi: Avanyisa matlhelo hamambirhi hi 5

   • 5x/5 = 20/5

   • x = 4

     
     

Endlelo leri leri hlelekeke ri tiyisekisa ku pakanisa naswona ri kombisa endlelo ro anakanya—nchumu wa nkoka swinene loko ku tirhana ni matirhelo lama rharhanganeke ya tinhlayo.

Chain-of-Thought vs. Few-Shot Prompting: Ku twisisa ku hambana

Maendlelo lawa hamambirhi ma nga ha vonaka ma fana loko u ma languta ro sungula kambe ma tirhela xikongomelo xo hambana:

• Few-Shot Prompting : Yi nyika swikombiso leswi kombisaka ku nghenisa na vuhumelerisi, ku fana na xilotlelo xa nhlamulo xa swihlawulekisi swo tala.

• Ku hlohlotela ka Nxaxamelo wa Miehleketo : Ku kombisa endlelo ro anakanya leri heleleke, ku fana na ku kombisa ntirho hinkwawo eka xiphiqo xa tinhlayo.

  
  

Few-Shot Prompting yi nga languteka hi ndlela leyi:

 Example 1: Problem: If a store sells books for $5 each and markers for $3 each, how much would 2 books and 4 markers cost? Answer: $22 Example 2: Problem: If a store sells notebooks for $4 each and pens for $2 each, how much would 3 notebooks and 5 pens cost? Answer: $22

Ku hlohlotela ka Nxaxamelo wa Miehleketo ku ta languteka hi ndlela leyi:

 Problem: If a store sells books for $5 each and markers for $3 each, how much would 2 books and 4 markers cost? Thinking: 1. Calculate cost of books: 2 books × $5 = $10 2. Calculate cost of markers: 4 markers × $3 = $12 3. Total cost = Cost of books + Cost of markers = $10 + $12 = $22 Answer: $22 Problem: If a store sells notebooks for $4 each and pens for $2 each, how much would 3 notebooks and 5 pens cost? Thinking: 1. Calculate cost of notebooks: 3 notebooks × $4 = $12 2. Calculate cost of pens: 5 pens × $2 = $10 3. Total cost = Cost of notebooks + Cost of pens = $12 + $10 = $22 Answer: $22

Zero-Shot Nxaxamelo wa Miehleketo

Hi ku tirhisa xiga xo olova kambe xa matimba xa " A hi ehleketeni hi goza, " Zero-shot CoT prompting yi tirhisiwa eka swiyimo leswi u nga riki na swikombiso swa milandzu evokweni. Xikombiso, "A hi ehleketeni hi goza: Hlamusela fizikisi ya quantum."

Xana u fanele ku tirhisa rini xitsundzuxo xa CoT?

Ku hlohlotela ka CoT ku voninga loko ku tirhana na:

• Tinhlayo leti rharhanganeke
• Ku anakanya loku twisisekaka ka magoza yo tala
• Ku twisisa loku tolovelekeke
• Ku Endliwa ka Vukanganyisi hi ndlela yo fanekisela

Vumundzuku: Nxaxamelo wa Miehleketo ya Multimodal

Nhluvuko wa sweswinyana eka ku anakanya ka AI wu hlanganisa swiphemu swa swivono na swa matsalwa eka maendlelo ya vutivi. Anakanya hi AI yi kambela ku lunghisa xikanyakanya, yi twisisa buku ya swiletelo, na ku nyika nkongomiso lowu voniwaka na lowu twalaka eka xiteji xin’wana na xin’wana xa endlelo. Endlelo ra multimodal ri kombisa nhluvukiso lowu taka eka nseketelo wa AI.

Ku kurisa Ku Humelela ka Switsundzuxo swa Wena swa CoT

Hambileswi ku hlohloteriwa ka nxaxamelo wa mianakanyo ku nga ni matimba, ku tirha hi swipimelo swo karhi. Ndlela ya ku anakanya leyi modele yi yi endlaka a yi tiyisisiwi ku va yi nga ri na swihoxo naswona yi hambanile nyana eka mimbuyelo.

Hi leswi switsundzuxo swo va ni vuyelo byo antswa:

• Ringeta hi vukheta hi maendlelo yo hambana.
• Anakanya hi ku hlanganisa ku hlohlotela ka CoT na tindlela tin’wana to hlohlotela.
• Tirhisa timodeli letikulu leti nga na ti parameter to tlula 100 wa tibiliyoni, loko swi koteka.

Mahetelelo

Loko hi languta ndlela leyi AI yi endleke nhluvuko ha yona, swi le rivaleni leswaku ku hlohlotela hi Chain-of-Thought (CoT) i endlelo leri tirhaka ro tirhisa vuswikoti byo anakanya bya xiyimo xa le henhla swinene. I swa nkoka ku xiya leswaku ku tirhisa CoT ku hlohlotela eka nhluvukiso wa switirhisiwa kumbe ku fikelela mbuyelo lowunene swinene eka ku nghenelela ka AI swi tlakusa swinene minkarhi yo fikelela ku humelela.

Xikongomelo a hi ku fikelela nhlamulo leyi kongomeke ntsena, kambe na ku twisisa endlelo leri endliweke ku endla tano, naswona sweswo i ku saseka ka ku hlohlotela ka Chain-of-Thought.

Tinhlamuselo ta le hansi

1. Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E., Le, Q., & Zhou, D. (2022). Ku hlohlotela ka Nxaxamelo wa Miehleketo Ku Humesa Ku Anakanyisisa eka Timodelo Letikulu ta Ririmi.
2. Kojima, T., Gu, S. S., Reid, M., Matsuo, Y., na Iwasawa, Y. (2022). Timodelo Letikulu ta Ririmi i Zero-Shot Reasoners.
3. Zhang, Zhuosheng, na van’wana. "Ku anakanyisisa ka nxaxamelo wa miehleketo ya multimodal eka timodeli ta ririmi." arXiv ku kandziyisiwa ka le mahlweni arXiv:2302.00923 (2023).