As I’m writing this, a new Chatterbox release is available on GitHub. As with the initial release, you can either download the source and recompile it or grab an executable (Windows, 32-bit) from the release.
(If you have never heard of Chatterbox, you should probably read the initial release article.)
The biggest change since v0.1 is support for DeepSeek AI.
Other changes include:
- Better network connection handling (queries can be stopped, and timeouts can be configured per engine)
- The system prompt can be enabled/disabled inside a chat tab
- New toolbar buttons to clear all chat histories and to jump to the previous or next question or answer
- The close chat button has been moved to the main toolbar
While you can use smaller DeepSeek engines through the ollama engine, the online version gives drastically better answers than small local models. On the other hand the local models ar private while the online version doesn't care about the data privacy.
The only official statement about the privacy is that the owner of DeepSeek servers (Hangzhou DeepSeek Artificial Intelligence Co., Ltd., and Beijing DeepSeek Artificial Intelligence Co., Ltd.) will collect and potentially use all the data they can get from us, including the content of the queries:
User Input. When you use our Services, we may collect your text or audio input, prompt, uploaded files, feedback, chat history, or other content that you provide to our model and Services.
While this is also true if you are using browser chat with other engines, other providers guarantee the privacy of your queries when you are using paid API access. With DeepSeek we have no such guarantie. I could find nothing related to the paid API privacy in the documentation and they had also not answered my question regarding this matter. I will therefore think three times before sending any fragment of my code to DeepSeek servers.
If we skip the privacy issue (simple: don't send any code in the queries), DeepSeek performs very well. See the last installment in the "Delphi and AI" series for an example.
I think, when it comes to privacy, any data sent to Internet could be leaked or used by third parties, even if providers "guarantee" to protect such data. Some of examples are "keyboards" apps on smartphone, sending all input to their owners for generating context adds and other. As for me, I would not be so worried about code fragments (they are not useful without wide environment of other stuff). What could be more sensitive - commercial information, business secrets, insider info, inner (private) business information, contained in reports and so on. If online LLM will be used to build customer support chat, containing RAG system, more data could be exposed to LLM owners, what looks a kind of risky.
ReplyDeleteFully agree, but I still feel a little bit better when sending code to the service that at least on the paper guarantees the privacy of the sent data. You never know when you'll be careless and posted some private data.
Delete