- Aug 11, 2023
- 2,440
- 1,459
- 24
- Country
- United States
- Faith
- Christian
- Marital Status
- Single
Hello folks. The year is 2024, and AI continues to grasp headlines in the news. Most AI models such as ChatGPT are run on servers, hundreds of miles away from the users. This means that data is not private, and AI could potentially be used to build a profile of you. There is a solution, known as Mistral 7B, an AI model that can be used locally on your own PC or Mac. It generates a bit of heat when running, but can answer basic to more complex questions (except math, cos AI is bad at math).
This is the AI explaining what a rosary is. It took about a minute or two to answer the question.
Article about ChatGPT privacy concerns: ChatGPT Has a Big Privacy Problem
Mistral 7B summary of the article (yes, my laptop got warm when summarizing this):
In July 2020, OpenAI released their large language model GPT-3, which was trained on millions of pages scraped from the web, Reddit posts, books, and more. However, this data inclusion has led to privacy concerns as it contains personal information of millions of individuals, including Italians.
On March 31, 2023, Italy's data regulator, Garante per la Protezione dei Dati Personali, issued a temporary emergency decision demanding OpenAI stop using the personal information of millions of Italians in ChatGPT. The regulator argues that OpenAI doesn't have the legal right to use people's personal information without their consent, which is a requirement under GDPR. In response, OpenAI has stopped access to ChatGPT for Italians while they work with the regulators.
This decision is the first taken against ChatGPT by a Western regulator and highlights privacy tensions around the creation of giant generative AI models. Similar investigations could follow across Europe, with France, Germany, and Ireland already contacting the Garante for more information. Tobias Judin, head of international at Norway's data protection authority, says that if a model is built on data that may be unlawfully collected, it raises questions about whether anyone can use the tools legally.
Under GDPR, organizations must rely on one of six legal justifications to collect and use people's information, including obtaining their consent or arguing legitimate interests. OpenAI's privacy policy mentions relying on "legitimate interests" but doesn't specify how they obtained the personal information used for training ChatGPT.
The Italian decision comes as scrutiny of large AI models increases, with tech leaders calling for a pause on their development due to potential future implications. The GDPR rules protect the data of over 400 million people across Europe, and OpenAI's actions could potentially breach these regulations by not having age controls, providing inaccurate information, failing to inform individuals about data collection, and lacking a legal basis for collecting personal information.
Experts argue that the issues raised by the Italian regulator are likely to apply to all development of machine learning and generative AI systems, as many data sets used for training have existed for years without considering privacy implications. The EU is developing AI regulations, but little action has been taken against the development of machine learning systems regarding privacy.
There's a lack of clarity on how to collect data lawfully for training data sets and how to delete or correct inaccurate information from large AI models, especially if the origins of the data are unclear. GDPR allows individuals to request deletion or correction of their information but may not be able to do so with large language models, as they don't have provisions for it.
The Italian regulator's decision has the potential to be the first of many cases examining OpenAI's data practices, as every individual European country can open complaints against it. OpenAI isn't alone in facing these issues, and many other AI developers may also need to address privacy concerns related to their training data and model development.
This is the AI explaining what a rosary is. It took about a minute or two to answer the question.
Article about ChatGPT privacy concerns: ChatGPT Has a Big Privacy Problem
Mistral 7B summary of the article (yes, my laptop got warm when summarizing this):
In July 2020, OpenAI released their large language model GPT-3, which was trained on millions of pages scraped from the web, Reddit posts, books, and more. However, this data inclusion has led to privacy concerns as it contains personal information of millions of individuals, including Italians.
On March 31, 2023, Italy's data regulator, Garante per la Protezione dei Dati Personali, issued a temporary emergency decision demanding OpenAI stop using the personal information of millions of Italians in ChatGPT. The regulator argues that OpenAI doesn't have the legal right to use people's personal information without their consent, which is a requirement under GDPR. In response, OpenAI has stopped access to ChatGPT for Italians while they work with the regulators.
This decision is the first taken against ChatGPT by a Western regulator and highlights privacy tensions around the creation of giant generative AI models. Similar investigations could follow across Europe, with France, Germany, and Ireland already contacting the Garante for more information. Tobias Judin, head of international at Norway's data protection authority, says that if a model is built on data that may be unlawfully collected, it raises questions about whether anyone can use the tools legally.
Under GDPR, organizations must rely on one of six legal justifications to collect and use people's information, including obtaining their consent or arguing legitimate interests. OpenAI's privacy policy mentions relying on "legitimate interests" but doesn't specify how they obtained the personal information used for training ChatGPT.
The Italian decision comes as scrutiny of large AI models increases, with tech leaders calling for a pause on their development due to potential future implications. The GDPR rules protect the data of over 400 million people across Europe, and OpenAI's actions could potentially breach these regulations by not having age controls, providing inaccurate information, failing to inform individuals about data collection, and lacking a legal basis for collecting personal information.
Experts argue that the issues raised by the Italian regulator are likely to apply to all development of machine learning and generative AI systems, as many data sets used for training have existed for years without considering privacy implications. The EU is developing AI regulations, but little action has been taken against the development of machine learning systems regarding privacy.
There's a lack of clarity on how to collect data lawfully for training data sets and how to delete or correct inaccurate information from large AI models, especially if the origins of the data are unclear. GDPR allows individuals to request deletion or correction of their information but may not be able to do so with large language models, as they don't have provisions for it.
The Italian regulator's decision has the potential to be the first of many cases examining OpenAI's data practices, as every individual European country can open complaints against it. OpenAI isn't alone in facing these issues, and many other AI developers may also need to address privacy concerns related to their training data and model development.
