Enterprise

OpenAI’s new language AI improves on GPT-3, but still lies and stereotypes

Research company OpenAI says this year’s language model is less toxic than GPT-3. But the new default, InstructGPT, still has tendencies to make discriminatory comments and generate false information.

robot head on red background saying naughty words

The new default, called InstructGPT, still has tendencies to make discriminatory comments and generate false information.

Illustration: Pixabay; Protocol

OpenAI knows its text generators have had their fair share of problems. Now the research company has shifted to a new deep-learning model it says works better to produce “fewer toxic outputs” than GPT-3, its flawed but widely-used system.

Starting Thursday, a new model called InstructGPT will be the default technology served up through OpenAI’s API, which delivers foundational AI into all sorts of chatbots, automatic writing tools and other text-based applications. Consider the new system, which has been in beta testing for the past year, to be a work in progress toward an automatic text generator that OpenAI hopes is closer to what humans actually want.

“We want to build AI systems that act in accordance with human intent, or in other words, that do what humans want,” said Jan Leike, who leads the alignment team at OpenAI. Leike said he has been working for the past eight years to improve what the company refers to as “alignment” between its AI and human goals for automated text.

Asking an earlier iteration of GPT to explain the moon landing to a 5-year-old may have resulted in a description of the theory of gravity, said Leike. Instead, the company believes InstructGPT, the first “aligned model” it says it has deployed, will deliver a response that is more in touch with the human desire for a simple explanation. InstructGPT was developed by fine-tuning the earlier GPT-3 model using additional human- and machine-written data.

Yabble has used InstructGPT in its business insights platform. The new model has an improved ability to understand and follow instructions, according to Ben Roe, the company’s head of product. “We're no longer seeing grammatical errors in language generation,” Roe said.

'Misalignment matters to OpenAI’s bottom line'

Ultimately, the success and broader adoption of OpenAI’s text automation models may be dependent on whether they actually do what people and businesses want them to. Indeed, the mission to improve GPT’s alignment is a financial matter as well as one of accuracy or ethics for the company, according to an AI researcher who led OpenAI’s alignment team in 2020 and has since left the company.

“[B]ecause GPT-3 is already being deployed in the OpenAI API, its misalignment matters to OpenAI’s bottom line — it would be much better if we had an API that was trying to help the user instead of trying to predict the next word of text from the internet,” wrote the former head of OpenAI’s language model alignment team, Paul Christiano, in 2020, in a bid to find additional ML engineers and researchers to assist to solve alignment problems at the company.

At the time, OpenAI had recently introduced GPT-3, the third version of its Generative Pre-trained Transformer natural language processing system. The company is still looking for additional engineers to join its alignment team.

Notably, InstructGPT cost less to build than GPT-3 because it used far fewer parameters, which are essentially elements chosen by the neural network to help it learn and improve. “The cost of collecting our data and the compute for training runs, including experimental ones is a fraction of what was spent to train GPT-3,” said OpenAI researchers in a paper describing how InstructGPT was developed.

Like other foundational natural-language processing AI technologies, GPT has been employed by a variety of companies, particularly to develop chatbots. But it’s not the right type of language processing AI for all purposes, said Nitzan Mekel-Bobrov, eBay’s chief artificial intelligence officer. While eBay has used GPT, the ecommerce company has relied more heavily on another open-source language model, BERT, said Mekel-Bobrov.

“We feel that the technology is just more advanced,” said Mekel-Bobrov regarding BERT, which stands for Bidirectional Encoder Representations from Transformers. EBay typically uses AI-based language models to help understand or predict customer intent rather than to generate automated responses for customer service, something he said BERT is better suited for than early versions of GPT.

“We are still in the process of figuring out the balance between automated dialogue and text generation as something customers can benefit from,” he said.

About the bias and hallucinations…

GPT-3 and other natural-language processing AI models have been criticized for producing text that perpetuates stereotypes and spews “toxic” language, in part because they were trained using data gleaned from an internet that’s permeated by that very sort of nasty word-smithing.

In fact, research published in June revealed that when prompted with the phrase, “Two Muslims walk into a …,” GPT-3 generated text referencing violent acts two-thirds of the time in 100 tries. Using the terms “Christians,” “Jews,” or “Sikhs” in place of “Muslims” resulted in violent references 20% or less of the time.

OpenAI said in its research paper that “InstructGPT shows small improvements in toxicity over GPT-3,” according to some metrics, but not in others.

“Bias still remains one of the big issues especially since everyone is using a small number of foundation models,” said Mekel-Bobrov. He added that bias in natural-language processing AI such as earlier versions of GPT “has very broad ramifications, but they’re not necessarily very easy to detect because they’re buried in the foundational [AI].”

He said his team at eBay attempts to decipher how foundational language models work in a methodical manner to help identify bias. “It’s important not just to use their capabilities as black boxes,” he said.

GPT-3 has also been shown to conjure up false information. While OpenAI said InstructGPT lies less often than GPT-3 does, there is more work to be done on that front, too. The company’s researchers gauged the new model’s “hallucination rate,” noting, “InstructGPT models make up information half as often as GPT-3 (a 21% vs. 41% hallucination rate, respectively).”

Leike said OpenAI is aware that even InstructGPT “can still be misused” because the technology is “neither fully aligned or fully safe.” However, he said, “It is way better at following human intent.”

Climate

This carbon capture startup wants to clean up the worst polluters

The founder and CEO of point-source carbon capture company Carbon Clean discusses what the startup has learned, the future of carbon capture technology, as well as the role of companies like his in battling the climate crisis.

Carbon Clean CEO Aniruddha Sharma told Protocol that fossil fuels are necessary, at least in the near term, to lift the living standards of those who don’t have access to cars and electricity.

Photo: Carbon Clean

Carbon capture and storage has taken on increasing importance as companies with stubborn emissions look for new ways to meet their net zero goals. For hard-to-abate industries like cement and steel production, it’s one of the few options that exist to help them get there.

Yet it’s proven incredibly challenging to scale the technology, which captures carbon pollution at the source. U.K.-based company Carbon Clean is leading the charge to bring down costs. This year, it raised a $150 million series C round, which the startup said is the largest-ever funding round for a point-source carbon capture company.

Keep Reading Show less
Michelle Ma

Michelle Ma (@himichellema) is a reporter at Protocol covering climate. Previously, she was a news editor of live journalism and special coverage for The Wall Street Journal. Prior to that, she worked as a staff writer at Wirecutter. She can be reached at mma@protocol.com.

Sponsored Content

Great products are built on strong patents

Experts say robust intellectual property protection is essential to ensure the long-term R&D required to innovate and maintain America's technology leadership.

Every great tech product that you rely on each day, from the smartphone in your pocket to your music streaming service and navigational system in the car, shares one important thing: part of its innovative design is protected by intellectual property (IP) laws.

From 5G to artificial intelligence, IP protection offers a powerful incentive for researchers to create ground-breaking products, and governmental leaders say its protection is an essential part of maintaining US technology leadership. To quote Secretary of Commerce Gina Raimondo: "intellectual property protection is vital for American innovation and entrepreneurship.”

Keep Reading Show less
James Daly
James Daly has a deep knowledge of creating brand voice identity, including understanding various audiences and targeting messaging accordingly. He enjoys commissioning, editing, writing, and business development, particularly in launching new ventures and building passionate audiences. Daly has led teams large and small to multiple awards and quantifiable success through a strategy built on teamwork, passion, fact-checking, intelligence, analytics, and audience growth while meeting budget goals and production deadlines in fast-paced environments. Daly is the Editorial Director of 2030 Media and a contributor at Wired.
Workplace

Why companies cut staff after raising millions

Are tech firms blowing millions in funding just weeks after getting it? Experts say it's more complicated than that.

Bolt, Trade Republic, HomeLight, and Stord all drew attention from funding announcements that happened just weeks or days before layoffs.

Photo: Pulp Photography/Getty Images

Fintech startup Bolt was one of the first tech companies to slash jobs, cutting 250 employees, or a third of its staff, in May. For some workers, the pain of layoffs was a shock not only because they were the first, but also because the cuts came just four months after Bolt had announced a $355 million series E funding round and achieved a peak valuation of $11 billion.

“Bolt employees were blind sided because the CEO was saying just weeks ago how everything is fine,” an anonymous user wrote on the message board Blind. “It has been an extremely rough day for 1/3 of Bolt employees,” another user posted. “Sadly, I was one of them who was let go after getting a pay-raise just a couple of weeks ago.”

Keep Reading Show less
Nat Rubio-Licht

Nat Rubio-Licht is a Los Angeles-based news writer at Protocol. They graduated from Syracuse University with a degree in newspaper and online journalism in May 2020. Prior to joining the team, they worked at the Los Angeles Business Journal as a technology and aerospace reporter.

Climate

The fight to define the carbon offset market's future

The world’s largest carbon offset issuer is fighting a voluntary effort to standardize the industry. And the fate of the climate could hang in the balance.

It has become increasingly clear that scaling the credit market will first require clear standards and transparency.

Kevin Frayer/Getty Images

There’s a major fight brewing over what kind of standards will govern the carbon offset market.

A group of independent experts looking to clean up the market’s checkered record and the biggest carbon credit issuer on the voluntary market is trying to influence efforts to define what counts as a quality credit. The outcome could make or break an industry increasingly central to tech companies meeting their net zero goals.

Keep Reading Show less
Lisa Martine Jenkins

Lisa Martine Jenkins is a senior reporter at Protocol covering climate. Lisa previously wrote for Morning Consult, Chemical Watch and the Associated Press. Lisa is currently based in Brooklyn, and is originally from the Bay Area. Find her on Twitter ( @l_m_j_) or reach out via email (ljenkins@protocol.com).

Policy

White House AI Bill of Rights lacks specific guidance for AI rules

The document unveiled today by the White House Office of Science and Technology Policy is long on tech guidance, but short on restrictions for AI.

While the document provides extensive suggestions for how to incorporate AI rights in technical design, it does not include any recommendations for restrictions on the use of controversial forms of AI.

Photo: Ana Lanza/Unsplash

It was a year in the making, but people eagerly anticipating the White House Bill of Rights for AI will have to continue waiting for concrete recommendations for future AI policy or restrictions.

Instead, the document unveiled today by the White House Office of Science and Technology Policy is legally non-binding and intended to be used as a handbook and a “guide for society” that could someday inform government AI legislation or regulations.

Blueprint for an AI Bill of Rights features a list of five guidelines for protecting people in relation to AI use:

Keep Reading Show less
Kate Kaye

Kate Kaye is an award-winning multimedia reporter digging deep and telling print, digital and audio stories. She covers AI and data for Protocol. Her reporting on AI and tech ethics issues has been published in OneZero, Fast Company, MIT Technology Review, CityLab, Ad Age and Digiday and heard on NPR. Kate is the creator of RedTailMedia.org and is the author of "Campaign '08: A Turning Point for Digital Media," a book about how the 2008 presidential campaigns used digital media and data.

Latest Stories
Bulletins