DuckDuckGo
On Thursday, DuckDuckGo unveiled a new “AI Chat” service that offers four medium-sized languages from OpenAI, Anthropic, Meta, and Mistral in a ChatGPT-like interface while trying to provide users with privacy and anonymity. Allows interaction with models (LLMs). . Although the AI models involved can easily output false information, the site allows users to test different medium-level LLMs without installing anything or signing up for an account.
DuckDuckGo's AI Chat currently features OpenAI's GPT-3.5 Turbo, Anthropic's Cloud 3 Haiku, and access to two open source models, Meta's Lama 3 and Mistral's Mixtral 8x7B. The service is currently free to use within daily limits. Users can access AI Chat through the DuckDuckGo search engine, direct links to the site, or by using the “!ai” or “!chat” shortcut in the search field. AI chat can also be disabled in site settings for users with accounts.
According to DuckDuckGo, chats on the service are anonymized, removing metadata and IP addresses to prevent people from going back. Citing its privacy policy and terms of use, the company says the chats are not used to train AI models.
DuckDuckGo says that “we have contracts with all model providers to ensure that chats saved by providers are completely deleted within 30 days,” and that our platform Any chat done on the form cannot be used to train or improve the models. “
Benj Edwards
However, the privacy experience is not bulletproof because, in the case of GPT-3.5 and Claude Haiku, DuckDuckGo needs to send user input to remote servers over the Internet for processing. Given some input (ie, “Hey, GPT, my name is Bob, and I live on Main Street, and I just killed Bill”), if such an extreme need arises, the user can still potentially be identified.
While the service seems to work well for us, there is a question about its usefulness. For example, while GPT-3.5 initially surprised people when it launched alongside ChatGPT in 2022, it also made a lot of noise—and still does. GPT-4 was the first major LLM to control bots to the point where the bot became reasonably useful for some tasks (although that itself is a moot point), but those more capable models are not present in DuckDuckGo's AI Chat. Is. GPT-4-level models like Cloud Ops or Google's Gemini Ultra are also absent, possibly because they are far more expensive to run. DuckDuckGo says it may develop paid plans in the future, and these may include higher daily usage limits or access to “more advanced models”.)
Granted, the other three models generally (and subjectively) pass GPT-3.5 in their ability to code with less illusion, but they can still make things. With DuckDuckGo AI Chat, the company has a new chatbot with a great interface and the promise that your interactions with it will remain private. But what's the point of completely private AI conversations if they're riddled with errors?
Benj Edwards
As DuckDuckGo itself states in its privacy policy, “By its very nature, AI Chat produces text with limited information. Thus, output that appears complete or accurate because of its detail or specificity. E.g. As such, AI Chat cannot dynamically retrieve the information and therefore the output may be out of date, without you verifying its content using other means, especially professional advice (such as medical, financial, or legal advice) should not be relied upon.”
So, have fun talking to bots, but tread carefully. They will simply “lie” to your face because they don't understand what they are saying and are designed to output statistically plausible information, not factual references.