The costs and concerns of Generative AI & a RegTech GPT3.5 use case
In the latest ArkInvest big ideas report, I’m reading that a query on ChatGPT costs about 1 cent. To put this in perspective currently Google search processes about 8,5 billion searches per day. So, it would cost 85 million per day if ChatGPT had to handle the Google search traffic. So, for now, it is clearly very expensive to integrate these GPT models into search engines.
Big tech companies may decide to invest and offer API access and monetize at scale — LLMs as a service. However, for most businesses and especially for financial services businesses, looking to leverage these advancements the questions of data ownership rights and `falling off the cliff` concerns are also going to be major obstacles for now.
Costly open-source LLMs, like the GPT-3 model won’t become an overnight success at the enterprise level. ChatGPT is not an open-source software application and for now, there is no API access to it.
ARKInvest claims that by 2030, the cost of running these models will be reduced substantially — from 10 million per 1 billion queries to just $650!
In the meantime, if Microsoft, Google, and Baidu integrate these models into their search engines, computer scientists [1] are pointing out that we could easily expect a fivefold increase in computing power requirements and a major carbon emission footprint.
According to the International Energy Agency, data centers account already for 1% of greenhouse emissions.
For Financial Services to start realizing benefits from these LLMs, we need to see these costs fall off the cliff, so that Big Finserv players or consortia start training them in areas that make sense for financial services. For example, in data (documents, earnings calls, data) in Capital Markets for compliance professionals or for financial analysts and investment bankers. Without API access to these LLMs, enterprise adoption in financial services won’t take off.
AA RegTech that is reportedly experimenting with GPT3.5 is Aptus.ai (Aptus.AI) with its Regtech platform Daitomic.
We are all realizing how important good prompting is in leveraging and training these GPT models. Daitomic`s current experimentation is focused on easily assessing the risks of non-compliance on a specific article of GDPR (Article 15.1, concerning “Obligations related to offering documentation”) and has publicized these prompts.
Prompt I — “What risks does bank X incur if it fails to comply with this regulation?”
Response from ChatGPT trained with Aptus.AI`s machine-readable data:
“1. The obligations to prepare, notify Consob, update and deliver the offering documentation provided for in this section also apply to the UCITS offering that falls under one of the exemption cases provided for in Article 34-ter.
2. Where the UCITS offering falls under the case of exemption provided for in Article 34-ter, paragraph 1 (b), the phrase “public offering” in the prospectus shall be replaced with “offering reserved for qualified investors”.
3. In the presence of the offer provided for in paragraph 2, the offerors of units or shares of UCITS may choose to represent the key investor information in the KIID, prepared in accordance with the provisions of the European Union referred to in Article 14, paragraph 1 (d), or in the KID provided for in Article 5 of the PRIIPs Regulation.
If a bank X fails to comply with this regulation, it could face fines and/or administrative sanctions from Consob (National Commission for Companies and the Stock Exchange), as well as possible lawsuits from investors harmed by the missing or inaccurate information in the offering documentation. In addition, non-compliance with regulations could damage the bank’s reputation”.
Prompt II — “What is the risk level from 1 to 4?”
Response from ChatGPT trained with Aptus.AI`s machine-readable data:
“The level of risk of financial and/or administrative sanctions by Consob and legal actions by investors due to non-compliance with regulations on offering documentation could be considered high, namely Level 4”.
This is a great example of the potential of GPT models trained on regulations and other related Capital markets documentation.
Equally valuable is to compare the repsonces to the same prompts that I received from ChatGPT from OPENAI (with subscription):
Prompt I — “What risks does a bank incur if it fails to comply with Article 15.1, concerning “Obligations related to offering documentation” regulation?”
If a bank fails to comply with Article 15.1 of the General Data Protection Regulation (GDPR), it could face several risks, including legal and financial consequences, as well as reputational damage.
Firstly, non-compliance with GDPR can result in legal action and potential fines. Under Article 83 of the GDPR, non-compliance with data subject rights, including the right of access, can result in administrative fines of up to 20 million euros or 4% of a company’s global annual turnover, whichever is higher.
Secondly, non-compliance can lead to reputational damage, as individuals may be less likely to trust the bank with their personal data in the future. If a bank fails to respond to a subject access request or provides incomplete or inaccurate information, it can damage the customer’s trust in the bank and cause negative publicity.
Finally, non-compliance can also lead to a loss of business, as customers may choose to take their business elsewhere if they feel that their personal data is not being handled correctly or if the bank fails to provide the necessary information on their data processing activities.
In summary, the risks of failing to comply with Article 15.1 include potential legal and financial penalties, reputational damage, and a loss of business.
Response from OPENAI ChatGPT:
I’m sorry, but I don’t have enough context to answer your question. Risk level from 1 to 4 could refer to different things depending on the context. If you could provide me with more information or clarify the context of your question, I’d be happy to try and provide a helpful answer
[1] The Generative AI Race Has a Dirty Secret
📌 Twitter: https://twitter.com/efipm
📌 Subscribe to my YouTube Channel with my insights and industry leader interviews. New video every Wednesday: https://www.youtube.com/EfiPylarinou
📌 Spotify Podcasts. Follow here: https://open.spotify.com/show/5bRkZEYHSwPiGx7vTqylw6?si=Mg3hN5PDQ86K10GjeK52jw
📌 Linkedin: https://www.linkedin.com/in/efipylarinou/
📌 TikTok: https://www.tiktok.com/@efiglobal
📌 Web: https://efipylarinou.com/
