It’s high praise indeed when Bill Gates says that his demonstration of ChatGPT in 2022 was only the second time in his life that he had seen a revolutionary technology.
Before you write-off Gates’ predictions as self-promotion, bear in mind that most of his outlandish predictions in his 1995 book, The Road Ahead, came true more quickly than he foresaw – translation product managers take note.
Slator’s March 24, 2023 poll revealed that 69.4% of respondents said that their customers were not asking about ChatGPT, this shows that people are still processing the magnitude of the coming revolution that Gates predicts. We believe forward thinking LSPs will connect the dots for them.
In our view, one of ChatGPT’s key differentiating factors is that it is the first technology that can edit poorly written English and translate from one language to another with a high degree of accuracy in both tasks. This has important productivity ramifications for multinational companies whose staff are non-native English speakers.
By using one integrated application, a non-native English executive can now prepare a PowerPoint in English, have it proof-read by ChatGPT and translated on the fly to share with colleagues who don’t read English well. The reality, however, is that most executives will not be able to figure out how best to perform that task on language models that are suited to their specific domain, diluting accuracy, and compromising effectiveness.
What Customers Think
We thought it would be useful to find out how our English editing customers will use ChatGPT to proofread their work, so we surveyed 600 users of our human English editing service (Brand GUARDIAN). The findings (from 67 respondents) paint an informative picture.
- 82% wanted ChatGPT integrated into our human editing service (so that they have a choice)
- 60% would use ChatGPT for correcting non-critical work
- 7% would use ChatGPT for correcting external communication
- 42% would use ChatGPT for correcting internal communication
- 72% were concerned about confidentiality risks of using ChatGPT
We expect these numbers to change rapidly when adoption accelerates and people become more comfortable with the technology. Specifically, we would expect the percentage using ChatGPT to edit non-critical work to rise as ChatGPT’s quality improves, and as LSPs carve out their own language models to suit their customers’ specific industry domains.
ChatGPT Drawbacks
Italy became the first liberal democracy to ban ChatGPT on March 31 for security breaches that they claim breach General Data Protection Regulation (GDPR). If this proves to be correct, then the whole of the EU will need to ban ChatGPT to comply with its own laws. With China off-limits to ChatGPT, losing the EU would be a big short-term blow.
Our survey shows that at the company level confidentiality is also a major concern. ChatGPT users need to bear in mind that whatever they plug into ChatGPT is stored by OpenAI forever and used to improve its own algorithms. This fact alone would breach many companies internal protocols on data protection of sensitive information and customer data.
The other drawback is that for highly technical and nuanced writing ChatGPT performed poorly in our test cases. Financial research, legalese, and medical research all scored badly, as did very specific edits on things like promoting a new brand or new property, where detailed knowledge is essential.
For example, when we asked ChatGPT to write a 500 word press release for our client’s April 2023 hotel opening in Ho Chi Minh, Vietnam, it said the hotel was in the heart of the city, had 355 rooms and invented a quote from a member of staff who left the company in 2021. The reality is that the hotel has 160 rooms, is close to the airport, and the new GM was hired in January 2023.
Drawbacks aside, we believe that the winners in this race will be the LSPs that integrate ChatGPT into their customer work-flows, train specific domain language models, and fuse the technology with professional tech-savvy linguists, providing customers with real-time optionality based on their needs.