Sign up to see the future, today<\/p>\n
Can\u2019t-miss innovations from the bleeding edge of science and tech<\/p>\n
For tech companies racing to be the king of the AI hill, there are few things more precious than raw, original data. <\/p>\n
To keep the large language models underlying our favorite AI chatbots up to date, tech companies have to feed them reams of fresh inputs. As one study found<\/a>, the amount of data being used to train AI has doubled every nine months since 2010 \u2014 exponential growth which may soon hit a wall as stores of clean data run critically low<\/a>.<\/p>\n When there\u2019s no more original content to pilfer, companies have started paying workers to generate fresh training data, offering them low-quality contracts<\/a> to train AI in hyper-specific tasks like running weekly payroll for Broadway musicians<\/a>. Others have been hired for to film themselves doing degrading or menial chores like folding laundry<\/a> or distinctly adult activities<\/a>.<\/p>\n Predictably, this growing workforce behind the AI boom has started cutting corners en masse, turning to other AI chatbots to supply the data meant to feed AI chatbots. Talking to New Scientist<\/a>, numerous insiders said this practice of AI cannibalism \u2014 a method experts have long warned can destabilize LLMs<\/a> \u2014 is shockingly commonplace.<\/p>\n \u201cIt\u2019s very widespread,\u201d a worker identified as Alice told NewSci. \u201cEvery company I\u2019ve worked for has had explicit guidelines around it and they clearly do try to catch people out, so I think they do care. But I don\u2019t think they can stop it.\u201d<\/p>\n In other words, AI companies are learning an ironic lesson: after purloining everybody else\u2019s<\/a> content<\/a> without permission to create a product that threatens employment across the economy, the new precariat they\u2019ve created are using the same tech to do the few human tasks they still need in as lazy a fashion as possible.<\/p>\n Though workers have to be careful not to be too obvious, Alice says it isn\u2019t hard to pass AI-generated data off as her own, provided she scrubs the obnoxious linguistic tics<\/a> of chatbots like ChatGPT before she submits it. \u201cIt\u2019s only the sloppiest of users that get caught,\u201d the AI contractor told NewSci. \u201cAnyone with a modicum of awareness around AI hallmarks can tell their output not to use them, and at that point what are you going to do?\u201d<\/p>\n \u201cIf these companies want quality data, then they should\u00a0offer quality contracts,\u201d Alice continued. \u201cInstead they\u2019re low-balling struggling people, employing them for the barest possible amount of time and tossing them aside as projects are finished with no warning.\u201d<\/p>\n Other contractors told NewSci they use LLMs in order to avoid making mistakes and losing their gig entirely.<\/p>\n \u201cI was terrified of not having an income source, and then after that, it just became easier to run everything through LLMs,\u201d one explained. \u201cFor a\u00a0lot of the projects that I do now, it\u2019s creating scenarios, so I will use one LLM to help me create the scenario and then I\u2019ll use a different LLM to help me create the files that go along with the scenario. I do feel guilty but like I said, in the beginning it was more about trying to make sure I wasn\u2019t making any errors.\u201d<\/p>\n Whatever the reason, it\u2019s clear workers aren\u2019t above feeding AI companies a taste of their own slop \u2014\u00a0a situation which could have drastic consequences<\/a> for the AI race as a whole.<\/p>\n