In the early years, AI models like ChatGPT or its rival Foghere required huge teams of low-cost workers to deliver human-like responses that helped the models distinguish basic facts such as whether an image is a car or a carrot. Was.
But more sophisticated updates to AI models in the highly competitive field are now demanding a rapidly growing network of human trainers who have specialized knowledge – from historians to scientists, some with doctoral degrees.
“A year ago, we could generally hire graduate students to teach us how to improve AI,” said Ivan Zhang, co-founder of Foghere, talking about its internal human trainers.
“We now have licensed therapists who are teaching models how to behave in a medical environment, or financial analysts or accountants.”
For more training, Cohere, which was last valued at more than $5 billion, works with a startup called Invisible Tech. Cohere is one of the main rivals of OpenAI and specializes in AI for businesses.
Startup Invisible Tech employs thousands of trainers working remotely, and has key partners from AI companies ranging from AI21 to Microsoft to train its AI models to reduce errors known in the AI world as hallucinations. Has become one of.
“We have 5,000 people in over 100 countries around the world who are PhDs, master’s degree holders and knowledge work experts,” said Francis Pedraza, founder of Invisible.
Depending on the employee’s location and complexity of the work, Invisible pays up to $40 per hour. Some companies like Outlier pay up to $50 per hour, while another company called Labelbox said it pays up to $200 per hour for “high expertise” topics like quantum physics, but $15 for basic topics. Starts with dollars.
Invisible was founded in 2015 as a workflow automation company that provided services to companies like food delivery company DoorDash to digitize their delivery menus. But things changed when a relatively unknown research firm called OpenAI contacted them in the spring of 2022, ahead of ChatGPT’s public launch.
“OpenAI came to us with a problem, which was that when you were asking a question in an early version of ChatGPT, it was hallucinatory. You couldn’t trust the answer,” Pedraza told Reuters.
“They needed an advanced AI training partner to facilitate reinforcement learning with human feedback.”
OpenAI did not respond to a request for comment.
Generative AI generates new content based on previous data used to train it. However, sometimes it cannot differentiate between true and false information and produces false results which are called hallucinations. In one notable example, a Google chatbot in 2023 shared incorrect information in a promotional video about which satellite first took photos of a planet outside Earth’s solar system.
AI companies are aware that hallucinations could derail GenAI’s attractiveness to businesses and are trying various methods to reduce it, including using human trainers to teach the concept of fact and fiction. also includes.
Since joining forces with OpenAI, Invisible says it has become the AI training partner for most GenAI companies, including Cohere, AI21, and Microsoft. Cohere and AI21 confirmed they are customers. Microsoft did not confirm that it is a customer of Invisible.
“These are all companies that had training challenges, where their number one cost was calculation power, and then the number two cost was quality training,” Pedraza said.
How does this work?
OpenAI, which started the craze around GenAI, has a team of researchers aptly named the “Human Data Team” who work with AI trainers to gather specialized data for training their models, like ChatGPT. Works with.
OpenAI researchers come up with various experiments like reducing hallucinations or improving writing style and work with AI trainers from Invisible and other vendors, said a source familiar with the company’s processes.
At any given point, dozens of experiments are being run, some with tools developed by OpenAI and others with tools from other vendors, the person said.
Depending on what AI companies want – from getting better at Swedish history or doing financial modeling – Invisible hires workers with degrees relevant to those projects, reducing the burden of AI companies managing hundreds of trainers.
“OpenAI has some of the most incredible computer scientists in the world, but they’re not necessarily experts on Swedish history or chemistry questions or biology questions or whatever you might ask,” Pedraza said. alone.
Coheer’s Zhang said he has personally used Invisible’s trainers to find a way to teach his GenAI models to extract relevant information from large data sets.
Competition
Among competitors in this space is Scale AI, a private start-up valued at $14 billion that provides sets of training data to AI companies. It has also ventured into the field of providing AI trainers, and counts OpenAI as a customer. Scale AI did not respond to interview requests for this story.
Invisible, which has been profitable since 2021, has raised only $8 million in seed capital,
“We are 70% owned by the team, and only 30% is owned by the investors,” Pedraza said. “We facilitate secondary rounds, and the most recent trade was valued at a valuation of half a billion dollars.” Reuters could not confirm that assessment.
Human trainers first received AI training through data-labeling work, which required less qualifications and was also less paid, sometimes less than $2, mostly by people from African and Asian countries. Was done.
As AI companies launch more advanced models, the demand for specialized trainers and in dozens of languages is increasing, creating a well-paid niche where employees from a variety of disciplines can become AI trainers without even knowing how to code.
The demand for AI companies is leading to the creation of more companies that are providing similar services.
“My inbox is basically full of new companies popping up here and there,” Zhang said. “I see this as a new place where companies hire humans to generate data for AI labs like ours “
(Except for the headline, this story has not been edited by NDTV staff and is published from a syndicated feed.)