Anthroponic says it doesn’t use your personal data to train its AI.

Leading generative AI startup Anthropoc has announced that it will no longer use its customer data to train a large-scale linguistic model (LLM) and help defend users against copyright claims.

Founded by former researchers at OpenAI, Anthroponic has updated its terms of service to describe its ideas and objectives. By capturing the personal data of its customers, Anthroposic strongly differentiates itself from rivals such as OpenAI, Amazon and Meta, which use user content to improve their systems.

“Anthroponics may not train models on Customer Content,” according to the revised agreement, adding that “as between the parties and to the extent permitted by applicable law, Anthroponic agrees that Customer shall retain ownership of all products and shall not waive any rights it may have in receiving Customer Content under these Terms.”

The terms go on to say that “Anthroponic does not expect to acquire any rights in Customer Content under these Terms” and “neither party grants any rights, of any kind, to the other's content or intellectual property.”

The revised legal document appears to provide greater protection and transparency for Anthropic's business customers. Companies own all AI outputs, for example, avoiding potential IP disputes. Anthroponic also promises to defend its customers against copyright claims arising from any infringing content created by Cloud.

The policy aligns with Anthropic's mission statement that AI should be useful, harmless and trustworthy. As public skepticism grows over the ethics of generative AI, the company's commitment to addressing concerns like data privacy could give it a competitive edge.

User Information: Essential Food for LLMS

Large scale language models (LLMs) such as GPT-4, LlaMa or Anthropic's Claude are advanced AI systems that understand and generate human language trained on large amounts of text data. These models use deep learning techniques and neural networks to predict word sequences, understand context, and understand the subtleties of language. During training, they constantly refine their predictions, improve their speaking skills, compose text, or provide relevant information. The effectiveness of LLMs depends on the variety and amount of data they are trained on, making them more accurate and context aware as they learn from different language styles, patterns, and new information.

And that's why user data is so valuable in training LLMs. First, it ensures that the models are up-to-date with current language trends and user preferences (eg, understanding new words). Second, it allows for personalization and better user engagement by adapting to individual user interactions and styles. However, this creates an ethical debate because AI companies do not charge users for this critical data that is used to train models that cost millions of dollars.

As reported by Decrypt, Meta recently revealed that it is training the upcoming LlaMA-3 LLM based on user data, and the new EMU models (which generate photos and videos from text queries) are also trained using publicly available data uploaded by users. social media.

In addition, Amazon said LLM, which powers the upgraded version of Alexa, is being trained on user conversations and interactions, but users can opt out of training information designed to assume users agree by default. to share this information.”[Amazon] “We've always believed it's important to train Alexa with real-world questions to deliver an accurate, personalized, and constantly improving experience to customers,” an Amazon spokesperson told Decrypt. But at the same time, we give our customers control over how Alexa voice recordings are used to improve the service, and we always respect our customers' preferences when training our models.

With tech giants racing to roll out the most advanced AI services, responsible data practices are key to earning public trust. In this regard, Anthroponic aims to lead by example. The ethical debate over access to more powerful and convenient models of handing over personal data is as prevalent today as it was decades ago when social media users were promoted as products instead of free services.

Edited by Ryan Ozawa.