May General Meeting – Rachel Pierce’s Whirlwind Demo of ChatGPT
My earliest interactions with ChatGPT being somewhat less than fruitful, I concluded that the tasks I tried to use it for were still faster and better done when completed by a human. Rachel Pierce’s whirlwind introduction to ChatGPT at the May General Meeting of the Northern California Translators Association changed my point of view and I may just implement some of the tricks that she showed us.
The NCTA General Meeting was held virtually over Zoom, which allowed our members living outside of the Bay Area to participate. Jennifer Santiagos started off the General Meeting with a short overview on California Assembly Bill (AB) 432. This so-called Court Interpreters Workforce Pilot Program bill seeks to create an additional pathway for aspiring court interpreters to obtain full-time employment in California courts. In principle, this a very good initiative; however, the devil, as always, is in the details. The bill as it is now written would have various unintended, negative consequences for Californian court interpreters and language access in general. You can read more in this article by the American Translators Association. Jennifer urged us to contact our representatives to point out the flaws in the current draft of the bill.
ChatGPT – Caveat Emptor
The meeting then shifted gears to Rachel Pierce’s feature presentation. Rachel is an ATA-certified, French-to-English translator and copywriter specializing in high-visibility marketing and communications content. You can find out more at www.rachelpierce.com.
Rachel briefly introduced the concept of artificial intelligence (AI) and large language models (LLMs). LLMs are huge neural networks with billions to trillions of parameters (numerical values – “knobs,” if you will – that can be adjusted) that have been trained on enormous amounts of text and other data. ChatGPT is an example of such an LLM. LLMs can perform a wide variety of tasks involving language, including translation; however, purpose-trained neural machine translation engines are currently still better at translation than LLMs. ChatGPT’s appeal lies in the tasks it can perform that go beyond translation.
Before embarking on a quick demonstration of these other tasks, Rachel cautioned us about a few of ChatGPT’s drawbacks. For one, it should not be treated as a search engine because it is unable to pull real-time data from specific websites or knowledge bases like Wikipedia. In addition, AI models frequently “hallucinate;” that is, they present the user with completely made-up information so convincingly that unsuspecting users treat these creations as fact. Furthermore, the data used to train ChatGPT ends sometime in fall 2021 and does not contain any information prior to that date. Finally, it lacks privacy safeguards, so users should be very careful regarding confidentiality agreements and other sensitive data.
What can ChatGPT really do?
Based on the above, you might think that LLMs are completely useless. Not so, as Rachel demonstrated. ChatGPT can be viewed as an editorial assistant, which can be quite useful with proper supervision and given the right prompts. When prompting ChatGPT, you should give it as much information as possible and give it a scenario for context. You can also use the “Regenerate response” button to request an alternative, parallel reworded response or refine the prompts iteratively.
Rachel demonstrated in real time how ChatGPT can perform the following tasks:
- Terminology and glossary work
- Compare and contrast multiple ideas, theories, words, etc.
- Summarize an article, book or other text
- Draft an email
- Fix punctuation, spelling and word choice
- Help you organize your thoughts
- Creative prompting
If you want to try it out yourself, you can follow Rachel’s instructions in her article on ChatGPT for ATA’s Next Level Blog.
The message I took away from Rachel’s phenomenal presentation was that Large Language Models won’t replace translators or other experts anytime soon. However, when used correctly, they can make our lives a whole lot easier by completing all the mundane, ancillary tasks so that we can concentrate on what really matters: translation.