Chinese search giant Baidu defends AI research capabilities after its ChatGPT-like Ernie Bot draws a turkey for Turkey
Chinese tech giant Baidu, which last week launched its ChatGPT alternative Ernie Bot, said on Thursday that the product was "totally self-developed" after some users expressed concerns that the company may have copied its chatbot's text-to-image capability from overseas.
Hilarious images generated by the Ernie Bot started to surface online earlier this week, as users found the chatbot drew a turkey, the bird, when asked for the country Turkey, and a crane - the bird - instead of the machine used for lifting heavy objects. These images were generated in spite of the Chinese-language prompts and use of completely different characters for each word.
Art influencer Liu Dake posted several of the artificial intelligence-generated images on Wednesday night, questioning if the Baidu product simply "translated the Chinese prompt into English and put it in foreign image generators like Stable Diffusion".
Do you have questions about the biggest topics and trends from around the world? Get the answers with SCMP Knowledge, our new platform of curated content with explainers, FAQs, analyses and infographics brought to you by our award-winning team.
He also said Ernie Bot would by default create a white person if nationality was not specified.
Baidu asserted that its chat bot has been "totally self-developed" and that the text-to-image capabilities were trained by its Ernie-ViLG model, according to a post on its official Weibo account on Thursday.
Baidu chairman and chief executive Robin Li Yanhong told Chinese media 36Kr on Wednesday that even GPT-4 is only able to understand an image, not generated one from scratch like Stable Diffusion.
Baidu also clarified that it uses "publicly available data worldwide to train the model, which is common industry practice". It promised that users will see "rapid fine-tuning and iterations" of the product in future.
Baidu founder Robin Li asks Ernie Bot, its answer to ChatGPT, to generate a poster at the company headquarters in Beijing on March 16, 2023. Photo: Handout alt=Baidu founder Robin Li asks Ernie Bot, its answer to ChatGPT, to generate a poster at the company headquarters in Beijing on March 16, 2023. Photo: Handout>
By Thursday noon, Ernie Bot could differentiate a crane bird from the machine, and did not mix up turkey and the country. It also drew an Asian face when asked for a "handsome guy". When the prompt was "to draw Turkey (the country) with wings," it generated a turkey (the bird) with colourful feathers as wings, according to a test by the Post.
The company has received mixed reviews for its product, which debuted less than three months after ChatGPT's November 2022 launch which generated headlines around the world.
Baidu initially came under fire when the Ernie Bot launch event only provided pre-recorded demonstrations. Its Hong Kong-listed shares slumped as much as 10 per cent that day. But subsequent reviews about the product's Chinese language skills and its ability to provide real-time information have seen the company's stock price climb 15 per cent since.
At the launch, CEO Li said Ernie Bot was "good at processing Chinese language" but he admitted its handling of English was not as good.
The Baidu product launch was overshadowed by OpenAI's upgraded GPT-4 model, unveiled just two days earlier. Compared with GPT-3.5 that powers ChatGPT, GPT-4 is able to process images, a feature still being worked out between OpenAI and its partner Be My Eyes.
This article originally appeared in the South China Morning Post (SCMP), the most authoritative voice reporting on China and Asia for more than a century. For more SCMP stories, please explore the SCMP app or visit the SCMP's Facebook and Twitter pages. Copyright © 2023 South China Morning Post Publishers Ltd. All rights reserved.
Copyright (c) 2023. South China Morning Post Publishers Ltd. All rights reserved.
