AI-powered story of delve overuse in science

AI-powered story of delve overuse in science

The unexpected relations between generative artificial intelligence, medical publications, and regional variants of business English were unearthed in spring 2024 when, out of the sudden, the frequency of the word delve in the scientific articles included in PubMed (a research publication base) rose hundred times comparing to pre-ChatGPT era.

Why did biomed folk fell in love with the word delve?

Jeremy Nguyen from the Swinburne University of Technology in Melbourne made the search across PubMed to find that since the beginning of 2024 the word delve has been used hundreds time more frequently than at the beginning of the century. PubMed itself is a database collecting biological science publications on a global scale; and delve used to be (I didn’t check whether it still is) one of the favorite words ChatGPT used. No surprisingly, the tweet with the finding concluded with the question “Are medical studies being written with ChatGPT?”. The subject stirred AI community and attracted mainstream media for GenAI had been over a year in the centre of hype. Soon The Guardian published the piece which heavily supported the positive answer to Mr. Nguyen’s question. Moreover, the likely explanation was given for the phenomenon. The article How cheap, outsourced labour in Africa is shaping AI English bore the title which spoke for itself, but anyway, let’s delve into the subject.


The original chart from J. Nguyen’s tweet shows incredible growth of ‘delve’ usage in 2023-2024.

RLHF done in Africa was a reason

RLHF – the reinforced learning with human feedback was the most likely reason behind the recent spectacular career of delve. Scientific term RLHF means that large language model is adopted to become suitable for chatting with humans by the refinement of its output. The refinement is based on the work thousands human workers who asses LLM’s answers and give the feedback. The gargantuan effort, which reportedly was outsourced by OpenAI to English-speaking countries of Africa, made ChatGPT the great tool it is today. AI had learned from humans how to conduct meaningful conversations. At te same time it picked up some? favorite words used by its human trainers. While ChatGPT uses dvelve far more often than it is statistically expected for the global Internet, there is a place where the word is used similarly often: the African web. It looks like ChatGPT is talking a bit African-ish dialect of English. Scientists from every part of the globe, using the tool to support or improve their professional publications, enabled that dialect to spread and the process was very quick due to the extremely fast adoption of the solution.

Not only delve and not only medics

The case of delve inspired further studies of the phrases commonly associated with AI-powered writing. The black list of such words was prepared with the suggestion to block their usage to make AI-texts look more human-like. The list includes top ten words to avoid: explore, captivate, tapestry, leverage, embrace, resonate, dynamic, testament, elevate, and, of course delve. Moreover, AI-favorite phrases were also identified. In the article on AI Phrase Finder fifteen such phrases were enlisted and delve into was one of them. Some others include: It is important to note that, in the dynamic world of, in this digital world, and embark on a journey. Does it sound familiar? I think for everybody in corporate business world those phrases should ring a bell. They sound like directly taken from the presentations and reports of the global consulting leaders. To be sure it is a correct connotation I made a quick search for the word delve in the recent reports of some leading consultancies. Indeed, the word is commonly iused and all reports Google has shown on the top are those from ChatGPT era. Apparently, not only bio-scientists use ChatGPT to improve their writing, the consultants do it as well.


Searching for ‘delve’ within BCG, McKinsey, and EY reports (Google search, September 2024); note all entries are from GenAI era.

Lessons and thoughts

The whole story is meaningful and inspirational on many levels. Many of aspects of the phenomenon were discussed in the aforementioned The Guardian article. Here are some of my takes:

·?????? AI is able to set linguistic trends, like motion pictures and TV before, but will it be flexible enough to make the language constantly evolving

·?????? AI has a potential to amplify the influence of underrepresented regions of the world and popularize the language traits which are used outside the old Anglophonic world

·?????? On the other hand it could be a trap of speaking English which sounds AI-like just because somebody live in the geography with extensive AI-training activities

·?????? There is still relatively easy to identify AI-driven writing, at least it was still possible in April 2024

·?????? The story points out to the necessity of monitoring AI-favorite vocabulary to make your prompts more efficient in delivering human-like results

At the end I’d like to show what is ChatGPT’s view on the word delve. It’s optimistic and maintains that the use of the word make conversation “more expressive and engaging’.


ChatGPT opinion on the use of ‘delve’ in conversation, answer got in September 2024.

?

Artur Miernik

Partner at EY

2 个月

Wojciech R. Bolanowski, MD PhD Wojtek, I will definitely read it with interest!

回复

要查看或添加评论,请登录

社区洞察

其他会员也浏览了