Tattalin Arzikin Manyan Harsunan Na'ura: Rarraba Token, Daidaitawa, da Farashin Mafi Kyau

Teburin Abubuwan Ciki

• 1 Gabatarwa
• 2 Tsarin Ka'idar
  • 2.1 Saitin Samfurin
  • 2.2 Bambancin Mai Amfani
• 3 Hanyoyin Farashi Mafi Kyau
  • 3.1 Kudaden Kashi Biyu
  • 3.2 Rarraba Token Mai Yarjejeniya da Wanda Ba Yarjejeniya Ba
• 4 Aiwarwarin Fasaha
  • 4.1 Tsarin Lissafi
  • 4.2 Aiwarwarin Lamba
• 5 Sakamakon Gwaji
• 6 Aikace-aikacen Gaba
• 7 Bincike Na Asali
• 8 Nassoshi

1 Gabatarwa

AI mai ƙirƙira da Manyan Harsunan Na'ura (LLMs) suna kawo sauyi a fagage daga binciken kimiyya zuwa masana'antu masu ƙirƙira, amma sanya farashin samun damar yin amfani da waɗannan kayan aikin yana gabatar da rikitattun ƙalubalen tattalin arziki. Wannan takarda ta ƙirƙira wani tsarin ka'ida don nazarin mafi kyawun farashi da ƙirar samfurin LLMs, tana ɗauke da mahimman siffofi ciki har da canjin farashin aiki, keɓancewar samfurin ta hanyar daidaitawa, da babban bambancin mai amfani mai girma.

2 Tsarin Ka'idar

2.1 Saitin Samfurin

Muna ƙirƙira mai siyarwa mai cin gashin kansa wanda ke ba da nau'ikan LLM da yawa ta cikin jerin samfura. Tsarin ya haɗa da canjin farashin sarrafa token shigarwa da fitarwa, keɓantawa ta hanyar daidaitawa, da buƙatun masu amfani daban-daban a cikin ayyuka daban-daban.

2.2 Bambancin Mai Amfani

Masu amfani suna nuna babban bambancin girma a cikin buƙatun aiki da kuma hankalin kuskure. Ƙimar daidaito bayanan sirri ne, yana nuna aikace-aikace daban-daban daga ƙirƙirar abun ciki mai ƙirƙira zuwa ayyukan bincike masu sarƙaƙiya.

3 Hanyoyin Farashi Mafi Kyau

3.1 Kudaden Kashi Biyu

Za a iya aiwatar da mafi kyawun tsarin ta hanyar jerin kudaden kashi biyu, tare da ƙara farashi ga masu amfani masu yawan amfani. Wannan yana ba da hujjar ayyukan masana'antu da aka lura na farashi mai matakai dangane da keɓancewar samfurin da matakan amfani.

3.2 Rarraba Token Mai Yarjejeniya da Wanda Ba Yarjejeniya Ba

Muna bincika wuraren yarjejeniya guda biyu: ɗaya inda mai bayarwa ke sarrafa rarraba token a cikin ayyuka, da kuma wani inda masu amfani ke rarraba token cikin 'yanci. Mafi kyawun tsarin farashi ya dogara ne akan ko rarraba token yana yin yarjejeniya da kuma ko masu amfani suna fuskantar ƙuntatawa.

4 Aiwarwarin Fasaha

4.1 Tsarin Lissafi

An ayyana aikin amfanin mai amfani kamar haka: $U(\theta, q, t) = \theta \cdot v(q) - t$, inda $\theta$ ke wakiltar nau'in mai amfani, $q$ shine inganci (cin token da matakin daidaitawa), kuma $t$ biyan kuɗi ne. Matsalar mai siyarwa ita ce haɓaka kudaden shiga bisa la'akari da ƙuntatawa na dacewar ƙwaƙƙwaro da na hankali.

4.2 Aiwarwarin Lamba

class LLMPricingModel:
    def __init__(self, cost_per_token, fine_tuning_cost):
        self.cost_per_token = cost_per_token
        self.fine_tuning_cost = fine_tuning_cost
    
    def optimal_two_part_tariff(self, user_types):
        # Aiwarwarin mafi kyawun tsarin farashi
        fixed_fees = []
        per_token_prices = []
        for theta in user_types:
            # Lissafa mafi kyawun (F, p) ga kowane nau'in mai amfani
            F = self.calculate_fixed_fee(theta)
            p = self.calculate_per_token_price(theta)
            fixed_fees.append(F)
            per_token_prices.append(p)
        return fixed_fees, per_token_prices

5 Sakamakon Gwaji

Tsarin ya nuna cewa masu amfani masu kamanceceniya na halayen ƙima-girma suna zaɓar matakan daidaitawa da cin token makamancin haka. Simintin lambobi ya nuna cewa farashi mai matakai tare da kudaden kashi biyu yana ƙara kudaden shiga na mai siyarwa da kashi 15-30% idan aka kwatanta da farashi ɗaya, yayin da yake kiyaye shigar masu amfani a sassa daban-daban.

6 Aikace-aikacen Gaba

Za a iya faɗaɗa tsarin tattalin arziki don nazarin sabbin aikace-aikacen LLM ciki har da ƙirar da aka ƙarfafa da dawo da bayanai, tunani na sarkar, da samfurori masu yawa. Hanyoyin bincike na gaba sun haɗa da kasuwanni masu fa'ida, farashi mai ƙarfi, da tasirin jin daɗin jama'a na tsarin farashi daban-daban.

7 Bincike Na Asali

Wannan takarda ta ba da gudummawa mai mahimmanci ga tattalin arzikin na'urorin masarrafi ta hanyar tsara matsalar farashi na Manyan Harsunan Na'ura. Tsarin mawallafan yana haɗa ka'idar tattalin arziki ta ƙanana tare da ƙirar sabis na AI na zahiri, yana magance wani muhimmin gibi a cikin wallafe-wallafen. Idan aka kwatanta da tsarin farashin software na gargajiya, LLMs suna gabatar da ƙalubale na musamman saboda canjin farashin aikinsu da kuma yanayin bambancin mai amfani mai girma. Ƙarfafawa da takarda ta yi akan kudaden kashi biyu ya yi daidai da ayyukan masana'antu da aka lura daga masu bayarwa kamar OpenAI da Anthropic, waɗanda ke amfani da farashi mai matakai dangane da matakan amfani da iyawar samfurin.

Hanyar ka'idar ta ginu ne akan wallafe-wallafen ƙirar tsari, musamman aikin Myerson (1981) akan mafi kyawun ƙirar gwanjo, amma ya faɗaɗa shi zuwa mahallin ayyukan AI tare da siffofi masu ci gaba. Bambanci tsakanin rarraba token mai yarjejeniya da wanda ba yarjejeniya ba yana ba da muhimman bayanai ga yanke shawara na ƙirar dandamali. Wannan bincike ya dace da binciken fasaha akan ingancin LLM, kamar aikin akan gine-ginen ƙwararru na gauraye waɗanda ke ba da damar rarraba albarkatu mafi ƙanƙanta (Fedus et al., 2022).

Daga mahangar aiki, tsarin yana taimakawa wajen bayyana dalilin da yasa muke ganin irin wadannan dabarun farashi daban-daban a cikin kasuwar sabis na AI. Binciken da ya nuna cewa masu amfani masu yawan amfani suna fuskantar farashi mafi girma yana nuna dabarun farashi na tushen ƙima da ake gani a cikin software na kasuwanci, amma tare da ƙarin rikitarwa na ƙuntatawa na albarkatu na tushen token. Kamar yadda aka lura a cikin Rahoton Fihirisar AI na Stanford 2024, farashin lissafi na gudanar da manyan samfura ya kasance mai yawa, yana sa mafi kyawun farashi yana da mahimmanci don samar da sabis mai dorewa.

Ƙuntatawa na takarda sun haɗa da mayar da hankali kan saitunan cin gashin kansa, yana barin ƙwararrun gasa don aikin gaba. Bugu da ƙari, samfurin yana ɗaukar cikakken bayani game da tsarin farashi, wanda ƙila ba zai kasance a aikace ba. Duk da haka, wannan binciken yana ba da tushe mai ƙarfi don fahimtar ƙa'idodin tattalin arziki da ke tattare da ƙirar sabis na LLM kuma zai yi tasiri ga duka binciken ilimi da aikin masana'antu yayin da ayyukan AI ke ci gaba da haɓaka.

8 Nassoshi

1. Bergemann, D., Bonatti, A., & Smolin, A. (2025). Tattalin Arzikin Manyan Harsunan Na'ura: Rarraba Token, Daidaitawa, da Farashin Mafi Kyau.
2. Myerson, R. B. (1981). Mafi kyawun ƙirar gwanjo. Lissafi na Binciken Aiki.
3. Fedus, W., Zoph, B., & Shazeer, N. (2022). Maɓalli na Canza Siffofi: Sikelin zuwa Samfurori Mafi Girman Siffofi Tiriliyan. Jaridar Binciken Koyon Na'ura.
4. Stanford HAI (2024). Rahoton Fihirisar Na'ura mai wadata Hankali na 2024. Jami'ar Stanford.
5. OpenAI (2023). Rahoton Fasaha na GPT-4. OpenAI.