In November 2022, OpenAI released ChatGPT, gaining 100 million users within two months and setting a record for the fastest-growing user base. Arguably, never before have we seen such an event get so much hype in the history of technology. Language technology suddenly got propelled into the center of the wider public’s attention in a way unheard of since Google Translate entered the scene. ChatGPT has become part of the public discourse, talked about at business meetings and at the dinner table, discussed both by mainstream news outlets and on social media. Some relish the debate, others are sick of it by now, and Nimdzi is listening to both sides — and all sides in between — weighing perspectives and offering the analyst’s take on just how this technology is transforming both our professional and personal lives.
The first wave of LLM adoption, and what’s coming next
Nimdzi has been tracking the transformation of large language models (LLMs) into real-life use cases and applications within the language industry, and nearly half a year into the age of ChatGPT, we’re bringing you the key findings from our ongoing research. These observations are based on fieldwork conducted by Nimdzi over the past two months and include over two dozen interviews, discussions, and demo sessions with a wide range of industry players ranging from language service and technology providers to enterprise language service buyers.
While this is an open-ended research project, some clear themes have already emerged that we wanted to share before the next tech disruption knocks on the door.
The following are some of the research questions we’ve been looking to find answers to:
- How is the “ChatGPT disruption” different from previous technological disruptions the language industry has quite well adapted to, especially the advent of neural machine translation (NMT)?
- Does ChatGPT lead to fewer work opportunities for language professionals? Or, inversely, does it expand the horizons of work for language talent and service providers, perhaps even creating vast new opportunities not normally within the purview of language services?
- Is the tool driving cost efficiencies, increasing revenue, or opening new revenue streams?
- How does it compare with the tried-and-tested language tech stack in terms of implementation and customization?
- In what ways do LLMs change the industry’s customer-vendor relationships?
- What are the key benefits, shortcomings, constraints, and best practices for the use of ChatGPT-like models right now, and what is the outlook?
Five months, five key insights (plus one extra)
- Our investigation tells us that LLMs are close to the top of the hype cycle (see where Gartner put generative AI in 2022). Language services – both on the client side and the provider side – are natural early adopters of the technology. While excitement is high, the first constraints, challenges, and limitations to operationalizing ChatGPT have already begun surfacing. It remains to be seen just how deep and long the trough of disillusionment will be, however. Gartner places Generative AI’s real impact reaching maturity within the next three to six years.
- ChatGPT and GPT-like models have all the benefits of a superfood: They can create new revenue streams, attract new clients and buyers, create new customer experiences, speed up engineering, scripting, and coding, and are looked at by practically all departments at both enterprises and LSPs alike. The opportunities seem endless at first glance, and all these use cases are being explored across the spectrum of language industry players. It indeed seems to be a general-purpose, jack-of-all-trades tool compared to the current, narrow-purpose application language technologies.
- Implementation of LLMs is (currently) being hindered by complexity manifested on multiple levels. This is very similar to what we experienced during the rise of NMT: a plethora of model options outside the obvious ones (from OpenAI), the problem of trust (due to the models’ tendency to introduce both hallucination and bias), lack of quality assessment standards and frameworks, and privacy and security concerns, as well as questions around ROI.
- Confirming the resilience of the language industry, no one is seriously concerned about LLMs reducing the relevance or adversely affecting the growth of the language industry. In fact, LLMs undoubtedly increase complexity when deployed in the context of global, multilingual communications, and the general expectation is that the industry’s knowledge and expertise in language technology implementation, as well as the extensive language talent pool’s ability to post-edit or validate the output will be needed now more than ever. If anything, this piece of technology opens new growth opportunities.
- For the most obvious application of using the technology as a machine translation engine, LLMs won’t come close to being the next generation MT engines anytime in the short- and mid-term (hallucinations and latency being just some of their as-of-yet unsolved problems). This has not prevented some small and medium businesses from thinking about replacing their current LSPs and NMT workflows, though the quality of outcomes and savings results are questionable at best and disillusionment is looming in the air. However, when it comes to augmenting MT workflows, LLMs show great new potential in tasks such as improving source text quality, improving TM suggestions, and contextualizing and correcting MT outputs (AutoPE). As a technology director from an LSP commented in their interview with Nimdzi: “[LLMs] are way better at evaluation than at actual translation.”
Are you as excited about LLMs as we are? Contact us at firstname.lastname@example.org to participate in our research or fill out the survey below!
Editor’s note: A more detailed version of this piece is available here for Nimdzi subscribers.