DuckDuckGo proves that it's indeed a reliable privacy browser with the new "AI Chat" service it offers to the users. Unlike other AI chatbots that keep your data and require you to sign up or install the app, the company will allow you to chat anonymously with OpenAI, Meta, Anthropic, and Mistral's mid-range large language models (LLMs).
With this, you can converse with the AI with peace of mind since your data are anonymized on the platform.
What Chatbots Can You Access on DuckDuckGo's AI Chat
(Photo : DuckDuckGo)
AI chatbots are on the rise, but users are skeptical as to how their data is used. DuckDuckGo will soon allow users to converse anonymously with four LLMs.
DuckDuckGo's AI Chat includes access to OpenAI's GPT-3.5 Turbo, Anthropic's Claude 3 Haiku, Meta's Llama 3, and Mistral's Mixtral 8x7B. This variety enables users to test different AI capabilities without needing to install software or create an account.
The service is currently free, with daily usage limits, and can be accessed through the DuckDuckGo search engine, direct links, or using "!ai" or "!chat" shortcuts.
Related Article: Apple Safari: Initial Plans of DuckDuckGo Switch, But Google Remains for Private Mode-Why?
Privacy and Anonymity Features
According to DuckDuckGo, all chats are anonymized, with metadata and IP addresses removed to protect user identity. The company asserts that user inputs are not used for AI model training, in line with their privacy policy. However, when using GPT-3.5 and Claude Haiku, user inputs are sent to remote servers for processing, which could pose potential privacy risks if extreme inputs were provided.
"We have agreements in place with all model providers to ensure that any saved chats are completely deleted by the providers within 30 days and that none of the chats made on our platform can be used to train or improve the models," says DuckDuckGo.
Performance and Limitations of AI Models
While the service works smoothly, its practical utility is debatable. GPT-3.5, though initially impressive, often produces inaccurate information. More advanced models like GPT-4, Claude Opus, or Google's Gemini Ultra are absent, likely due to higher operational costs.
DuckDuckGo hints at future paid plans that may include these advanced models and higher usage limits, Ars Technica reports.
The other three models (Claude 3 Haiku, Llama 3, and Mixtral 8x7B) generally outperform GPT-3.5 in coding tasks with fewer hallucinations. However, they can still produce errors, making the service a novelty rather than a reliable tool for critical tasks.
Privacy vs. Accuracy: The Trade-off
DuckDuckGo emphasizes privacy in its AI Chat, but this comes at the cost of accuracy. As stated in their privacy policy, AI Chat generates text with limited information, which can lead to outdated or incorrect outputs. Users are advised not to rely on the AI's responses without verification, especially for professional advice in medical, financial, or legal matters.
DuckDuckGo's AI Chat offers an ideal blend of privacy and AI interaction, appealing to those concerned about anonymity. However, the accuracy of the AI models remains a significant challenge.
Users can enjoy experimenting with the chatbots, but they should be cautious and verify any important information through other sources. It's always important to know that not all responses we get from AI chatbots are accurate.
The latest DuckDuckGo AI service is a game-changer in integrating privacy-focused AI into everyday use, but it has yet to overcome the inherent limitations of current mid-range LLMs.
Read Also: Google's AI-Powered Note-Taking Assistant, NotebookLM, Goes Global in Over 200 Countries
