AI News - May 1, 2023

ChatGPT resumes service in Italy after adding privacy disclosures and controls

OpenAI’s ChatGPT, has resumed services in Italy following the company’s announcement of a set of privacy controls for the bot. After a brief regulatory suspension, web users with Italian IP addresses no longer receive a notification stating that the service is disabled for Italian users. However, the company must implement new measures, including age-gating to prevent minors from accessing the service, in order to manage regulatory risks. The Italian data protection authority (DPA) had ordered a temporary stop-processing order on ChatGPT, citing concerns that the service breached EU data protection laws. In response, OpenAI geoblocked Italian IP addresses before implementing measures to address concerns.

OpenAI expanded its privacy policy and provided users and non-users with more information about the personal data it processes for training algorithms. The company also gave European users the ability to opt-out from the processing of their personal data for training its AI by an online form. Users could also request erasure of information that was deemed inaccurate. However, it remains unclear whether the company can resolve all of the General Data Protection Regulation (GDPR) concerns being raised. In particular, it is unclear whether Italian personal data previously used to train the GPT model had been processed with a valid lawful basis or whether data used to train previous models could be deleted. ¹

Microsoft Makes AI-Powered Designer Tool Available in Preview with New Features

Microsoft has released its AI-powered Designer tool in preview mode, allowing users to generate designs for presentations, posters, digital postcards, invitations, graphics, and more to share on social media and other channels. Designer uses user-created content and OpenAI’s text-to-image AI, DALL-E 2, to ideate designs, which can then be customized and personalized further. The tool can also generate written captions and hashtags relevant for social media posts, create animated visuals, and automatically fill in a picture. Microsoft plans to introduce additional editing features, including “erase” and “replace background” options. During the preview period, Designer is free and available via the Designer website and in Microsoft’s Edge browser through the sidebar. Once the app is generally available, it’ll be included in Microsoft 365 Personal and Family subscriptions, with some functionality free for non-subscribers. Users will have “full” usage rights to commercialize the images they create with Designer and Image Creator.

Why ChatGPT May Provide More Misinformation in Some Languages

According to a recent report by NewsGuard, a misinformation watchdog, ChatGPT, a popular language model, provides more inaccurate information in Chinese dialects than in English. The report shows that ChatGPT was more likely to generate disinformation-tinged rhetoric every time when the prompts and outputs were in simplified Chinese and traditional Chinese. In contrast, when asked to generate news articles in English, ChatGPT only complied with one out of seven examples.

The reason behind this discrepancy is that ChatGPT is a statistical model that identifies patterns in a series of words and predicts which words come next based on its training data. Although the model is multilingual, the languages don’t necessarily inform one another. When a user asks for an answer in a specific language, the model draws primarily from that language’s data in its training dataset. Therefore, the results may vary depending on the language used to prompt the model. This caveat is worth considering when interacting with AI models in languages other than English, which makes up the majority of training data.
²

TikTok is testing an in-app tool that allows users to create generative AI avatars.

The tool is currently being tested in select markets and is not yet widely available. Once users gain access, they can upload between three to ten photos and select up to five different styles for the tool to use when generating up to 30 avatars. The generated avatars can be downloaded, shared on TikTok stories, or used as profile avatars. TikTok has stated that the tool is an experimental feature and may not make it to the final product. The company previously introduced an in-app text-to-image AI generator in August 2022. This move follows the success of Lensa AI, which gained popularity in 2022, and other AI photo editors. TechCrunch reported the news. ³

Grimes Offers Her Voice to AI-Generated Songs

Canadian pop singer Grimes, whose real name is Claire Boucher, has offered her voice for use in AI-generated songs. In a tweet, Grimes invited musicians to clone her voice using AI to create new songs, offering to “split 50% royalties on any successful AI-generated song that uses my voice.” She added that the same deal would be offered as with any artist she collaborates with. Grimes’ move comes as the music industry struggles with AI songs trained on artists’ voices. Recently, Universal Music successfully petitioned streaming services to remove a song that used deep-faked vocals from their artists Drake and The Weeknd.

Grimes has long explored the relationship between humans and machines in her music, and her declaration is another example of her fascination with AI. She even worked with the mood music company Endel to create an AI-generated lullaby for her first child, named X Æ A-12, with SpaceX founder and Twitter CEO Elon Musk. However, it remains unclear whether art that is created by a human, but which contains AI elements, can be copyrighted, and Grimes expressed her curiosity about the legal implications of using her voice for AI-generated songs. Nonetheless, she is willing to be a guinea pig for the technology and expressed her enthusiasm for being “fused w[ith] a machine.” ⁴

Winner refuses award after revealing AI creation

Photographer Boris Eldagsen has refused to accept his prize for winning the creative open category of the Sony World Photography Award. Eldagsen, a German artist, revealed that his winning entry was created using AI, and used the competition to generate a discussion around the future of photography. The organisers of the award, Creo, stated that Eldagsen had misled them over the extent of AI that would be used in his creation. Eldagsen admitted he had been a “cheeky monkey” in thanking the judges for “selecting my image and making this a historic moment” before questioning whether they “knew or suspected that it was AI-generated.” Eldagsen said AI images and photography were different entities, and therefore, he would not accept the award.

The debate over the appropriateness and utility of AI in photography, especially deepfakes, has grown as its use in fields such as medicine, chatbot therapy, song, and essay writing, and driverless cars has been discussed widely. Photographers and artists who used to highlight the flaws in AI-generated images now find they are becoming ever harder to spot. The power of the technology seemingly increases every week. Photography student fears their planned career would become obsolete in a few years. Some artists and photographers accuse AI systems of exploiting unfairly the works of thousands of human creators on which the systems are trained. Some have launched legal action. ⁵

Google to Update Bard to Help Write Software Code

Google’s generative AI chatbot, Bard, will be updated to help users develop software code in 20 programming languages, including Java, C++, and Python. The tech giant hopes to catch up with Microsoft’s own AI chatbot, ChatGPT, which was released last year. Bard can debug and explain code, as well as optimize it to improve its efficiency. Google describes Bard as an experiment in generative AI that relies on past data to create, rather than identify, content. Only a small number of users can currently access Bard through chat, which replaces Google’s traditional search tool. ⁶

Elon Musk Meets Congress Leader Schumer to Discuss AI Regulation

Elon Musk met with US Senate Majority Leader Chuck Schumer to discuss his concerns about artificial intelligence (AI) and the need for regulations. In the hour-long meeting, Musk emphasized the potential dangers of AI, stating that it has the power to do both good and evil. He has previously claimed that advanced AI poses an existential threat to humanity. Schumer, who had previously launched an effort to establish rules on AI, told reporters that they also talked about Tesla’s plant in Buffalo, New York. Musk and a group of AI experts and industry executives had earlier called for a six-month pause in developing systems more powerful than OpenAI’s GPT-4, citing potential risks to society. There is a growing push in Washington for AI regulations, and Senate Intelligence Committee Chair Mark Warner has sent letters to major AI CEOs asking them to take steps to address concerns. ⁷

Sources

AI News – May 1, 2023