Wanted: Poets by AI companies
Recently, I came across an intriguing trend where several AI tech companies are actively recruiting creative writers, many of them poets (as reported by Andrew Deck). AI companies, such as Scale AI, have been recruiting “talented creative writers with Masters or PhD degree to help train generative artificial intelligence models to become better writers” (here).
In light of the looming concerns of job replacement by AI, this news of recruiting human writers comes as refreshingly surprising, at least initially.
The performance of LLM is constrained by a simple reality: garbage in, garbage out. The quality of the output is a direct reflection of the input. The primary source of data for these LLMs is the web, which, as we know, is full of redundant, boring, low-quality, and harmful content.
A group of researchers meticulously uncovered that billion-sized datasets used to train many large vision and language models contain a huge amount of “misogyny, pornography, and malignant stereotypes” (Birhane, Prabhu, and Kahembwe 2021).
Controlling the quality of AI outputs is no easy task. The painstaking job of improving data quality often falls on human shoulders. It was revealed that the “quest to make ChatGPT less toxic” had been built on the exploited labor of Kenyan workers (as reported by Billy Perrigo).
0 Comments