Is ChatGPT ready for customer relations?
Feb. 1, 2023
The model ChatGPTmodel, put online by OpenAIhas made a name for itself with its ability to conduct incredibly fluid dialogs. But is it ready to meet real customer needs? What can we learn from it?
Arriving with great fanfare in November 2022, ChatGPT is the first online service to reach one million users in a single week. As explained in our article, the Large Language Model (LLM ) integrates textual analogies learned from very large corpora from the web and digital publication archives, covering science, industry, humanities, code debugging... ChatGPT has also been trained by reinforcement to conduct ongoing dialogues, and has also learned to be wary of user requests and to reject at first sight subjects deemed sensitive (violence, racism, phishing...), but it readily lends itself to games of prompting or prompt engineering to shape its conversations and its textual or code productions. As a result, essays and analyses are pouring in...
ChatGPT's strengths and weaknesses
On the plus side, there's no denying that the mega-model generates extremely fluent responses, whatever the topic, without any technical adaptation other than prompting. Responses are syntactically and semantically correct (at least in English, although some rare syntactical errors may occur in French) and remain fairly consistent throughout the conversation. The answers generated are also very dense in information, sometimes even when this information is not requested, as the model likes to insert a few elements of definition in the introduction to an answer, or to add a few additional elements of justification at the end.
This strength leads directly to ChatGPT's main weakness: the uncontrolled nature of the information provided. The information may be perfectly correct, but just as often it is false (inventions by authors and scientific articles, hyper-credible false computer diagnoses) or misleading (a mixture of correct, up-to-date information with that which is ten years out of date), for example when it comes to product range guidance. Or the answer may simply be too wordy or inappropriate, as in these examples of bank callbot simulation. In the examples below, the model's ability to generate sentences is impressive, but the first extract unfolds partly contradictory questions and proposals, without letting the interlocutor express himself, while the second reveals an apprehension of the bank card based on the American model. In fact, it's very difficult to check the professional conformity of the output generated on the fly.
Can ChatGPT be used in production?
The first eloquent response came from Stackoverflow, which banned ChatGPT from its forum for IT developers to help each other, due to the excessive noise induced by the mega-model's inventions, which were as false as they were credible.
Another response comes directly from the firm behind ChatGPT, OpenAI, which has been backed by Microsoft since 2020. OpenAI is projecting sales ofone billion dollars by 2024, but ChatGPT is not for sale. The LLM with millions of users that is making the blogs is clearly there to draw attention to OpenAI's other ready-to-use models, which served as the basis for its learning curve. The LLMs in the GPT3 family come with explicit pricing, which is multiplied by six if you want to use a fine-tuned version, i.e. one tailored to a specific task and domain, via supervised learning.
The answer, both externally and internally, is clear: No, ChatGPT is not for production. However, other fine-tuned LLMs may be...
What's the outlook for customer relations?
Once we've digested the fact that ChatGPT is a device designed to interact with beta-testers and influencers, and not a system intended for production, we're left with the phenomenon of LLMs and, more generally, deep learning for conversational interactions. Do these models have a role to play in the future of customer relations, or not?
At ViaDialog, we're convinced that we can: while our stack already integrates the latest generation of neural building blocks, in a controlled, production-proven environment, we're continuing to prepare for what's to come, to offer even more fluid, efficient and competent dialogue systems. Stay tuned for future communications on this subject!
Article written by Ariane Nabeth-Halber, AI Director @ViaDialog
To find out more...
- A good introduction to word2vec, the basic building block of Large Language Models (LLM), also known as embedding or "data plunging", this is a vector coding of words, which already captures certain semantic analogies.
- Another fundamental LLM mechanism is attention:
A. Vaswani, N. Shazeer, N. Parmar, J. Uszko- reit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, pages 5998-6008, 2017.
- BERT, The first LLM based on attention and self-supervised learning:
J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018.
- The first achievements of GPT-3 :
T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, et al. Language models are few-shot learners. In Advances in Neural Information Processing Systems, volume 33, pages 1877- 1901, 2020.
- Google Deepmind's equivalent of GPT-3:
J. W. Rae, S. Borgeaud, T. Cai, K. Millican, J. Hoffmann, et al. Scaling language models: Methods, analysis & insights from training Gopher. arXiv preprint arXiv:2112.11446, 2021.
- Google's equivalent of chatGPT:
R. Thoppilan, D. De Freitas, J. Hall, N. Shazeer, A. Kulshreshtha, et al. LaMDA: Language models for dialog applications. arXiv preprint arXiv:2201.08239, 2022.
Successes and setbacks for Facebook-Meta's Galactica :
