For a while, Microsoft looked like it would eat Google’s lunch. Its sluggish search engine Bing was being revolutionized with a new hot OpenAI chatbot. Those hopes have waned because no one—not even AI scientists—truly grasped the breadth of artificial intelligence’s capabilities for once. Early users of Bing reported unruly, emotional, even threatening responses to some questions from the AI system, which called one user a “bad researcher” and told a writer he would “happily Wasn’t getting married.” Bing, whose bot goes by the name Sydney, has overshadowed Google’s Bard error. However, these flaws are just the tip of the iceberg.
The technology behind chatbots like Bard and OpenAI’s ChatGPT comes from large language models (LLMs), computer programs trained on the billions of words on the public Internet that can generate human text. If ChatGPT is a car, this model is its engine, and OpenAI has been selling access to it since 2020. But amid today’s arms race for search bots, even those engines are being shared freely and their flaws are being patched.
OpenAI doesn’t disclose how many developers have access to its LLM, GPT-3, but it’s likely in the hundreds of thousands. While there are dozens of free, open-source LLMs, OpenAI is seen as the gold standard. Given Google’s resources, its model LaMDA may soon prove to be just as popular. Google kept its model a secret for years, explaining that its reputation could be damaged if launched prematurely. Yet, earlier this month, as Microsoft announced it would soon power Bing with GPT, Google appeared to reverse that position. Not only did it launch Bard the next day, but it also said it would open access to LaMDA. This strategy could irk Google, Microsoft and OpenAI, just as it did Facebook in 2018, when it was forced to shut down access to user data following the Cambridge Analytica scandal. All it took was one rogue user.
One of the big risks is bias. Twitch shut down an animated spoof of Seinfeld that featured AI-generated animation and dialogue after the show’s characters made transphobic and homophobic remarks. That dialogue was created by a “less sophisticated version” of GPT-3.
GPT-3 was trained on billions of words from a range of sources, including 7,000 unpublished books, Wikipedia entries and news articles, making it vulnerable to picking up even biased or hateful content. OpenAI has used human mediators to strip a lot out of its models, but the task has not been easy. Bias is also nearly impossible to detect when it is buried deep in LLM, a complex layered network of billions of parameters that acts like a black box even for its own creators. Misinformation affects these models as well. Tech news site CNET produced 77 articles on financial advice last November using the LL.M. It is to issue corrections on 41 of them. OpenAI doesn’t disclose the “hallucination rate” of its language model, but a January 2022 report on tech news site Protocol said it was between 21% and 41%, citing researchers. My own experience using ChatGPT has been misinformed. between 5% and 10%. Even if the rate is so low, companies looking to use LLM in programs need to take everything with a big grain of salt.
Abuse is probably the biggest unknown. OpenAI prohibits GPT-3 clients from using it to promote violence or spam. Criminals get ‘Content Policy Violation’ emails, but bad actors can ignore all that. Stephen Bayle, associate professor of security and political violence at the University of Exeter, used GPT-3 to generate fake ISIS propaganda as part of a study last year. He recalls receiving requests for clarification from OpenAI, and replies to explain. “We said, ‘This is academic research’,” he recalls. “We haven’t heard back.”
OpenAI says it has stopped “hundreds” of actors trying to misuse GPT-3 for a wide range of purposes, such as misinformation, and is constantly tweaking its models to filter out harmful content. Used to be. But there are other LLMs for bad actors to use.
In early 2019, OpenAI released a 70-page report on the social impact of language models, and said it would not release its latest LLM because it could be misused. The scene has changed a lot since then. Sure, its language models have become more accurate and less biased, its security filters more effective. But commercial pressure and the growing influence of Microsoft, which invested $1 billion in 2019 and another $10 billion this year in OpenAI, appears to have moved it to make a riskier bet on commercializing its technology. Google is doing the same now with its plan to sell access to LaMDA.
With Google stumbling and Microsoft Bing making bizarre comments, both companies need to slow down their AI arms race. Their revolutionary chatbots aren’t ready to go wide and neither are the language engines powering them.
Permi Olsson is a Bloomberg Opinion columnist covering technology.
©Bloomberg
catch all business News, market news, today’s fresh news events and Breaking News Update on Live Mint. download mint news app To get daily market updates.