Columnists

AI getting smarter, thanks to smarter human trainers

In the early years, getting AI models like ChatGPT or its rival Cohere to spit out human-like responses required vast teams of low-cost workers helping models distinguish basic facts, such as if an image was of a car or a carrot.

But more sophisticated updates to AI models in the competitive arena are now demanding a rapidly expanding network of human trainers who have specialised knowledge, from historians to scientists, some with doctorate degrees.

"A year ago, we could get away with hiring undergraduates, to just generally teach AI on how to improve," said Cohere co-founder Ivan Zhang, talking about its internal human trainers.

"Now we have licensed physicians teaching the models how to behave in medical environments, or financial analysts or accountants."

For more training, Cohere, which was last valued at over US$5 billion, works with a startup called Invisible Tech.

Cohere is one of the main rivals of OpenAI and specialises in AI for businesses.

The startup Invisible Tech employs thousands of trainers, working remotely, and has become one of the main partners of AI companies ranging from AI21 to Microsoft to train their AI models to reduce errors, known in the AI world as hallucinations.

"We have 5,000 people in over 100 countries who are PhDs, master's degree holders and knowledge work specialists," said Invisible founder Francis Pedraza.

Invisible pays as much as US$40 per hour, depending on the location of the worker and the complexity of work.

Some companies such as Outlier pay up to US$50 per hour, while another company called Labelbox said it pays up to US$200 per hour for "high expertise" subjects like quantum physics, but starts with US$15 for basic topics.

Invisible was founded in 2015 as a workflow automation company catering to the likes of food delivery company DoorDash to digitise its delivery menu.

But things changed when a relatively unknown research firm called OpenAI contacted it in the spring of 2022, ahead of the public launch of ChatGPT.

Pedraza said: "OpenAI came to us with a problem, which is that when you were asking an early version of ChatGPT a question, it was going to hallucinate. You couldn't trust the answer.

"They needed an advanced AI training partner to provide reinforcement learning with human feedback."

OpenAI did not respond to request for comment.

Generative AI produces new content based on past data used to train it.

However, sometimes it can't distinguish between true and false information and generates false outputs known as hallucinations.

In one example, in 2023 a Google chatbot shared inaccurate information about which satellite first took pictures of a planet outside the Earth's solar system in a promotional video.

AI companies are aware that hallucinations can derail GenAI's attractiveness to businesses and are trying to reduce them, including using human trainers to teach the concept of fact and fiction.

Since getting onboard with OpenAI, Invisible says it has become AI training partners to most of the GenAI companies, including Cohere, AI21 and Microsoft. Cohere and AI21 confirmed they are clients. Microsoft did not confirm it is a client of Invisible.

Pedraza said: "These are companies that had training challenges, where their number one cost was compute power, and then the number two cost is quality training."

OpenAI, which started off the frenzy around GenAI, has a team of researchers aptly named Human Data Team that works with AI trainers to gather specialised data for training its models like ChatGPT.

OpenAI researchers come up with experiments like reducing hallucinations or to improve writing style and work with AI trainers from Invisible and other vendors, a source familiar with the company's processes said.

At any point, dozens of experiments are being run, some with tools developed by OpenAI and others by tools of vendors, the person said.

Based on what the AI companies want, from getting better at Swedish history or doing financial modelling, Invisible hires workers with relevant degrees for those projects, reducing the burden of managing hundreds of trainers by the AI companies.

Pedraza said 1,000 contract workers catered to OpenAI alone.


* The writers are from Reuters

Most Popular
Related Article
Says Stories