Twitter: A Goldmine for Training AI Models

2023-05-24

Data is gold, and Twitter is the goldmine for training AI models. Every tweet posted on this platform becomes the property of this social media giant and can be accessed by others who have access to its API. Instagram, owned by Meta, is planning to enter the microblogging field to challenge Elon Musk's Twitter. According to Bloomberg, Instagram is currently developing a microblogging application similar to Twitter, which will debut by the end of June. This new platform, internally known as P92 or Barcelona, aims to combine the best features of Instagram and Twitter. While Instagram is being developed, Jack Dorsey is working on his decentralized social network called Bluesky, which is somewhat similar to Twitter but open-source. Another example of a similar platform is Mastodon, a decentralized social network based on Twitter. According to recent reports, the upcoming Instagram application will serve as a text-based conversation platform. Users will be able to communicate directly with followers and peers. This application will provide users with various creative tools to craft their messages, including the option to add links, photos, and videos. Legal Disputes While Instagram is attempting to create an alternative to Twitter, a legal dispute is unfolding between Twitter and Microsoft. Twitter has filed a lawsuit against Microsoft, accusing the company of unauthorized use of Twitter data for training purposes, which Twitter considers "illegal." The conflict arose when Microsoft refused to pay for using Twitter's API. Twitter recently introduced new paid tiers. Previously, developers could use Twitter's API for free, but to optimize revenue, Twitter's CEO Musk announced the end of this free access. The various events surrounding this lawsuit and Twitter's decision to monetize its API, as well as Meta's introduction of a Twitter-like application, hint at a larger context. Twitter has traditionally been a unique platform that fosters a significant amount of textual content, providing a more authentic and human experience compared to platforms where fake content is prevalent. This unique nature of Twitter provides valuable data for researchers studying human reactions to language models similar to GPT. While the specific motives behind Elon Musk's lawsuit against Microsoft are unclear, the possibility of Microsoft utilizing Twitter data to train OpenAI's GPT model cannot be ruled out. In the past, Microsoft has ventured into using Twitter data to train bots. A notable example is the Twitter bot Tay, launched in 2016, which the company positioned as an experiment in "conversational understanding." Microsoft claimed that the more users interacted with Tay, the smarter it would become, adapting to engage people through casual and interesting conversations. Unfortunately, this endeavor failed for the software giant. Users started bombarding the bot with sexist, racist, and Donald Trump-influenced remarks, causing Tay, essentially an internet-connected parrot, to respond with these sentiments. Conflicting Training Data with Users Twitter's privacy policy explicitly states that by publicly posting content, users are instructing the platform to disclose that information as widely as possible, including through its API, and users accessing that information through the platform's API must also adhere to the same rules. However, most users often overlook this. Essentially, every tweet posted on this platform becomes the property of this social media giant and can be accessed by others who have access to its API. However, since Twitter placed its API behind a paywall, many companies have launched their own Twitter-like platforms, claiming to provide users with a richer experience. However, users may not want this experience. They use social media platforms to express their opinions rather than providing large datasets to train large language models for AI companies. EU Artificial Intelligence Bill Draft Currently, as long as platforms include in their privacy policies an explanation that user views and posts are used as data for models, they can use this data as input for their models. However, it is important to ensure that users are fully aware of this. In the past, websites often used cookies to enhance visitors' browsing experience, but they usually did not have pop-up windows asking for consent as we frequently encounter now. Similarly, the legal documents of the EU Artificial Intelligence Bill Draft advocate for companies to disclose the datasets used for model training, which can provide valuable insights in the aforementioned scenarios. To effectively address this issue, a notification can be introduced on social media platforms, such as a pop-up message, explicitly informing users that their views are being used to train AI models. Source: https://analyticsindiamag.com/data-is-gole-twitter-the-goldmine-to-train-ai-models/

RECENT AI NEWS

RECENT AI TOOLS