Adopting Generative AI
By Marjolein Groot Nibbelink
With more than 100 million sign-ups in its initial two months, ChatGPT is the most quickly adopted consumer application in the history of the world. Compare this to the adoption of the telephone — arguably an even more influential communication device — which took roughly 75 years to gain 100 million users. Yes, it took infrastructure, hardware, and a learning curve to get a telephone in your home whilst new forms of generative AI are at your fingertips on a device you already own, on a platform you already know. As this friction to adopt new technology is reduced with time, the question turns from, “What would this cost me in time and money, and what value does it add to my life?” to, “What keeps me from trying, and what’s the risk of not adapting at speed?” The FOMO is real. Our learning curves are shorter, tech changes faster, and our time to think is reduced to a fear-driven nudge to jump on board or risk missing the boat forever. Generative AI promises us an increase in productivity, efficiency, and the public projection of our intellect in a user-friendly, intuitive, groundbreaking AI application. For free. How do your peers who focus on tech development in localization use it in their daily lives? We asked them a few questions about it.
VP of marketing at Smartcat
Our team uses Smartcat’s own software — which includes generative AI capabilities — to create various types of multilingual content such as email headlines, blog post ideas, and job descriptions. It also enables research and the gathering of intelligence on various topics. Many of our clients also use us for corporate learning content development. We’ve learned that prompts require finesse to generate valuable results and there’s a learning curve for that.
One of the most valuable use cases of generative AI has emerged from our usage of Smartcat’s technology to develop new content by first training it on our own existing creative assets, which already exist in various languages. It gets to know our brand voice, tone, and specific terminology to create new content in each language from scratch using this intelligence. We believe every global enterprise needs a multilingual AI solution to allow it to communicate consistently and effectively both internally and externally. Maintaining brand voice and terminology in one language is hard enough, but in multiple languages, it becomes hard to maintain without a collaborative platform that serves a central system of record for all employees that touch the content lifecycle. Those who don’t embrace it risk being left behind. As we hear time and again from our clients, reviewing and editing content using generative AI requires no additional time than reviewing original assets, but of course the turnaround time is 100 times that of a human.
Senior technical writer at AlayaCare
In my capacity as a linguist, I occasionally use ChatGPT to develop a pitch paragraph whenever I wish to reach out to new clients. In my day job, where I serve as a technical writer in the home healthcare industry, I primarily use generative AI as a source of inspiration for my daily writing tasks. Specifically, whenever I am given new product information, and I am looking to get started, I’d occasionally use generative AI to create a draft. I’d then read the draft to understand whether a user learning this feature for the first time would understand what I am trying to convey. I otherwise play with ChatGPT for my personal Instagram. Earlier in 2023, I started an Instagram account where I talk about my current hometown Scarborough, one of the most multicultural cities in Canada. The account is run entirely in traditional Chinese, so I use generative AI to create ideas and drafts for posts.
Let me share with you a secret — this current profile photo I have has been enhanced by generative AI. Traditionally, you can’t increase the resolution of an image other than retaking it, but the photographer who did this photo series had already removed my images from his server, so I had to do it myself. I used Upscale.media to touch up on the image but retained its fidelity largely. Imagine if you are a small business and you need to work on your website on a very tight budget. You have to create profile bios for your team, and you run into this situation where everyone submits blocks of text without a single tone of voice or images which are too low-res for formal publishing. These open-sourced tools can help you resolve challenges like these.
My advice to other industry users is we must remember not to let the hype take over while not treating it like some sort of mythical creature that brings nothing but doom. I am always of the belief that generative AI is here to assist, not here to take over, but you can only achieve this if you actively push yourself to learn how to use these tools to the best of your ability. The translation industry had its scare a few years already — I remember in 2017-2019 there was all the talk about how machine translation (MT) was going to “take over.” Clearly, it hasn’t, but it will impact you negatively if you still vehemently refuse to adopt the tool in some way. MT or AI won’t take over, but the people who learn how to maximize the benefits of these tools will be the ones to take over and replace you. As for hype, I think it’s important to tread carefully in this world of generative AI tools. We may be pressured by our workplace or colleagues to have to use generative AI tools, but for example, in my daytime job, while I encourage colleagues to use DeepL, I always ask them to submit the output for review before using it as is. The same thing goes for writing-focused generative AI tools. Put restraints and boundaries on how you’d use it, such as limiting its use to creating drafts for you to get started.
Co-founder and CEO of Translated
I use generative AI a lot and would like to remind everyone that MT was the first form of generative AI. Regarding the new generative AIs, content adaptation to a specific target and use case looks to me the most important application so far. It will allow people to learn more by consuming content that is aligned with the time, attention, and education they have. To my competitors, I would suggest: Don’t use it. It is like machine translation. It is evil and will put all LSPs and translators out of business soon.
Community manager at Creative Words
Of course, I’m amused by generative AI. Befriending technology is the only way to embrace progress and bend it to your needs rather than surrendering to it. This is the mindset we have at Creative Words and that I fully support. So far, I have mostly been using it for work, but I am pretty sure I will bring it to my personal world as soon as I see an opportunity where it can really be advantageous.
On a company level, we have tested it on various projects, both for internal use and for offering a broader range of services to a broader range of customers. More personally, as I have a thing for stats and data, I have found it very useful to create formulas for my spreadsheets or to rearrange data where Excel or Google Sheets wouldn’t do the trick (or I wasn’t good enough to figure out the solution by myself).
One of our customers recently said, “You can’t stop water with your hands.” Technology is here to stay, and we in the industry have become well aware of that already when the first CAT tools came into our lives. Trying to resist it is just an illusion. Early adoption, on the contrary, may help you gauge its pros and cons yourself and understand where it can really support your own workflows and make you faster, better, and stronger. Don’t let the others decide that for you!
Founder and CEO of Bureau Works
think there is a misconception generated by using generative AI too loosely. Its prompt-based nature lends itself to such a wide range of applications and approaches that it’s easy to mistakenly compare apples to oranges. I enjoy open-ended conversations with GPT-4 that don’t necessarily have an ultimate output. I love how the continuous dialogue refines and improves the quality of the suggestions. At a content level, I use it less and less frequently. Initially, I was amazed at how many coherent sentences it could pull together. But giving these sentences a closer glance, I could not feel like the content was mine. It became more troublesome to continuously re-prompt for improvements, weed out misinformation and falsehoods and try to introduce personality to the text than just writing things from scratch. In many regards, using GPT-4 for writing content made me become more appreciative of my own quirks and idiosyncrasies when it comes to crafting text. I am currently exploring the potential and limitations of generative AI, particularly regarding translation quality and productivity. I focus on embracing the computational costs and the API rate limits and squeezing the most benefits possible out of prompt engineering and machine learning principles.
So far, the most valuable use we have found is the ability to make seemingly small decisions that take a lot of work with incredible accuracy. We don’t rely on generative AI as a panacea — rather, we are exploring the dialogue between humans and machines without having to think about prompts or copying and pasting content but as a seamless experience. I’ve also found immense value in the ability to pick up on potential semantic deviations, as well as other potential issues that minimize the translator burden. It’s a simpler and more productive translation experience if employed within the right framework.
I would recommend other language industry pioneers not to adopt it but to explore it with both optimism and skepticism at the same time. This understated expectation allows me to explore opportunities without pushing too hard on a given direction. Some of the things I betted on initially did not work out due to exorbitant computational costs, ineffective results, or API rate limits, but some of our initiatives were incredibly scalable, predictable, and effective. I think as long as one approaches it with an open mind, it will typically yield more good things than bad things. I think the problems begin when I am unwilling to budge on a particular vision, even though results tell me otherwise. If anything, all this advent with gen AI has made me more aware of how in the end, continuous learning is perhaps what most defines us as a species.
President of tbo
If you look at machine translation in its current form (MT based on artificial neural networks), then tbo, just like everyone else in the translation field, has been using generative AI successfully for many years. Machine translation is a generative kind of AI in its essence, though many people might overlook this with all fanfare that we’ve seen with ChatGPT and the like. Regarding the newer developments, we’ve been exploring ChatGPT and using it to a limited extent on some live projects. In our linguistic department, we’ve been doing some tests with prompt-based MT to see how it behaves with style and terminology instructions. We foresee that prompt functionality will be part of the CAT tool environment within the next generation of technology, so it’s definitely something that we are getting ready for. Regarding live projects, we have also used ChatGPT to help generate strings for clients involved in developing new language models for different consumer technologies. We’ve also used it somewhat curiously to develop candidate texts for internal communications, like trying out different wordings for holiday wishes and whatnot. Understanding that machine translation per se is not the stated goal of ChatGPT, or other similar large language models, we know that they are not ready for large-scale production as MT engines in their current state, but clearly, there are some features that we can see merging with the traditional LSP workflow.
Like the printing press and the internet before it, we feel that generative AI is in itself a massive multiplier. There are obviously pros and cons that come with any of these technologies, but the net result for human cultures and economies is that they multiply the possibilities to create, organize, and transmit information. For anyone in the field of localization or translation, that means an expanding demand for language services of every sort imaginable. The translation world has already been living through this emergence for many years now — everyone is familiar with the conversations around translation vs. post-editing, for example, and the different effects that has had. Undoubtedly, we will see new instances of human specialists engaging with AI-created texts, whether to correct them or modify them for specific situations; we will also see huge new streams of content that enter into circulation thanks to AI users that, in turn, promote all sorts of new activity — some cultural, some economic, some “who knows.” All of that benefits our chosen field, and it’s all due to generative AI’s function as a multiplier.
Anyone who is a professional in the world of language services should at least be familiar with the technology, even if that is not going to be your daily specialty as a professional. It’s a good idea to know what’s going on and what’s possible. One might not be personally interested in sitting there at the keyboard having long, thoughtful conversations with ChatGPT, but your kids are going to be doing it, and your clients are going to be doing it. The technology will transition from being a fad to a more ubiquitous kind of functionality that will simply become part of our technosphere and everyday life. Those of us in the world of translation and language services won’t simply come into contact with the text that these kinds of apps create — as a profession, we will need to be power users, for lack of a better term. This means not only seeing out the transition of conventional communications as they pass into the realm of generative AI but also being proficient in the new kinds of tools that seem to have emerged magically from this “black box.”
CEO and principal co-founder of Global Saké
I’ve been playing around with generative AI to test the extent of its latest technologies, especially in experimenting with the current capabilities of multilingual generative AI performance. However, I prefer to not use generative AI in my everyday compositions. That said, I’m still curious about this powerful tool and will continue staying updated. Whether we like it or not, human biology needs to catch up to technology.
Generative AI is a useful writing tool for repetitive copy tasks. For example, LinkedIn has recently launched a gen AI feature for hiring managers, auto composing their replies in LinkedIn messenger. To preserve my voice in my professional and personal uses, though, I’d rather compose my own script than use an auto-generated copy. Where I do see great value today in gen-AI impact is in the ability of trained large language models (LLMs) to effectively enhance neural machine translation (NMT) performance in a rapidly changing world, where global-first real-time multilingual deploys may become the singular scalable solution. We already see ChatGPT and other generative AI successful adoption in mainstream MTs for error detection, accuracy post-edit, and as a look-up context tool for multilingual content creation. A world where everyone can generate multilingual local content origination from the get-go is a decentralized world that questions the translation paradigm of a single source into multiple target languages.
On one hand, generative AI may cannibalize our human creativity and inspiration and make us “comfortably numb.” On the other hand, it may just be the transformative revolution trigger we’ve been waiting for in enabling global-first product launches for the diverse populations of the world, with its potential of generating multilingual, geo-fit, in-context content origination in real-time instant deploys.
Are there big-rock challenges and risks along the way? Absolutely! But there’s also a sky-high opportunity for rendering more relevant, usable, and valuable product and content experiences for people across the globe.
CEO of Smartling
I’ve been using generative AI for the past year or so, both professionally and personally. As quality improves, so does its usefulness. Gen AI is a massive productivity boost for me. I use it to create and improve ideas, write, edit, and perform analyses and summaries of large amounts of data.
Generative AI is already being used in the localization industry for source and target copy creation, as a standalone translation engine, or as a quality estimation tool. However, some of these implementations do not have a scalable and repeatable process behind them, and limitations like hallucinations, false fluency, and factual inaccuracy exist, especially outside English. Smartling has patents pending and is focused on enabling our customers to harness generative AI now and in the future.
Founder of custom.mt
I use generative AI for work, personal development, and other situations. The most valuable use is for my company. We integrated it as an MT engine now available in Trados, memoQ, and Smartling. OpenAI now makes up almost 50% of our volume of machine translations. For myself, I find use in it for information extraction from unstructured texts. Also, code. I say treat it as a daily productivity tool, just like Microsoft Office and Excel.
CTO at Yamagata Europe
I am currently utilizing generative AI for professional purposes. Rather than exploring ways of selling AI-driven products or services, I prefer to think about how I can leverage it to make my own life easier from a business perspective. Personal experiences are often an excellent starting point for a variety of commercial initiatives.
At this point, every idea is a potential use case, but I see immense potential in generative AI for augmented machine translation. Gender-inclusive MT is a perfect example. Additionally, I like to use generative AI solutions for coding and development assistance. Finally, generative AI is an excellent resource to help with all kinds of content solutions, which includes content creation and review, text summarization, and even providing context in the translation process.
Generative AI comes with two assets that are very innovative in the language technology space, especially when combined. On the one hand, generative AI solutions can take context and reference data into account, while on the other, they can follow instructions from natural language descriptions. This combination is extremely powerful and opens doors towards many interesting opportunities to enrich some of the most traditional processes in the language industry, from authoring to translation, review, and quality assurance.
Director of localization at Coupa Software
I use generative AI on a daily basis in multiple contexts. At work, as soon as I open my computer, I use an implementation that automatically summarizes e-mail threads and JIRA tickets to speed up my reading. I use GPT-4 to replace web searches for general information and ideation. I have not struggled with starting off a blank page since ChatGPT came out — I start every document with a few bullet points from AI and work from there. I also use extensively some AI services that can summarize videos and web pages. I also leveraged some tools that recorded meetings and produced highlights and meeting notes for a while, but I stopped doing so due to security concerns at work. Outside of work, my favorite use by far is image generation. I am one of those who embarrass themselves when trying to draw anything, and generative AI has given me a way to visualize things that before only existed in my mind. It feels like getting superpowers, and it’s deeply rewarding.
Quick ideation is by far the most valuable use of generative AI. Having easy access to an on-demand brainstorming mechanism is priceless.
Everybody should be incorporating generative AI into their toolbox, as it just makes so many processes faster and easier. The fact that is a language-driven tool means that we, language professionals, are particularly well-equipped to get the most out of it. I think that, in general, we need to be better at adopting state-of-the-art technology as quickly as other industries do, but, in this case, it is a must to know what is and is not possible and set the right expectations for ourselves and our internal and external customers.
Subscribe to stay updated between magazine issues.