I’ve been leery of doing much with AI. ChatGPT, Gemini, Claude, Copilot and the rest seem to be well-trained workalikes owned by Google and Microsoft and therefore are probably as leaky as sieves.
Just yesterday I discovered Duck.ai, which seems to be as independent and private as DuckDuckGo. It just might be the safer way to go.
It looks safer on the surface.
It would be unwise to put personal info into any AI engine. They remember anything you input.
That’s why I asked the group about Duck.ai. My search moved on to Proton, which has fielded Lumo. It might be even safer than the Duck product. I asked it my test question and its answer was very much the same as Duck’s. Both organizations far exceed the security available from the usual suspects.
Who owns ChatGPT?
Open AI is the parent company. I don’t know what it’s security level is. I think I’ll stick with Proton and Duck.
I asked DDG about the security of OpenAI ChatGPT
“All user chats are completely anonymous.
OpenAI does not record any user chats.
Identifiable metadata (like IP address) is removed before sending prompts to OpenAI, making chats anonymous. OpenAI may store chats briefly for system checks, but there is no risk of de-anonymization.
Personal information entered in chat cannot be connected to any individual.
OpenAI will not use DuckDuckGo chats to train or improve models.
Chats are sent anonymously only to OpenAI, not to other third parties. DuckDuckGo sends prompts without identifying info, such as IP address.”
That seems OK, but what can you believe?
I don’t seem to attract any ads based on searches I do in DDG. I trust it more than Google or Bing.
That is a clever test.
DDG works great on my Android tablet
After reading everyone’s posts here, I’ve come to the conclusion that I’m less concerned with privacy than most. When I started using Garuda Linux, I installed Firefox because it’s what I’d been using in other distributions as well as Windows. The FireDragon browser is installed by default in Garuda Linux, and I’ve recently become aware that it’s maintained and developed by Garuda, so I decided to give it a try, and I like it. It’s a fork of Floorp, which in turn’s a fork of Firefox. As with Firefox, I enabled AI in the side bar and chose Gemini as the agent because it worked so well when I had it help me create my Show-It.py script for use in both Garuda and Windows, but I have kept the default search engine - Garuda’s searX, also privacy oriented. In both Firefox and FireDragon I use both the address and side bars to ask questions and perform searches, etc., but I still, and probably always will confirm the AI results I get by checking the sources attached to the search as well as any search results listed in the main window, because just as I don’t blindly trust any news item I see or read on TV or across the Internet, I don’t implicitly trust AI. Incidentally, I haven’t noticed any uptick in advertisements I get via email or on websites I visit that are related to my recent activity with Gemini. As for trusting claims of privacy from the developers/owners of any AI agent you use, if you can’t trust their claims on their sites, you can’t trust their AI agent enough to use it either.
The above is a description of how I use AI as well as some of my thoughts on the technology in general,
Ernie
I rarely use any AI, but when I do, I use the one offered by the Duck search.
I recently used it for a complicated debugging exercise. duck.ai only got me part way there, I had to ( with help) go to ChatGPT …
That got me closer, and it seemed to understand the complex situation better.
So now I dont know what to think.
AI engines differ in ability as well as in privacy.
Hi Neville,
About security on ChatGPT…
A while back, after chatting with GPT for a few hours, I made a mistake and copied and pasted a local IP address (and I think a password).
I was immediately warned not to do that and that it wouldn’t save any sensitive data.
I came up with a solution - I’m not sure if it’s viable - but I told it to always use the fictional IP and password I created as a reference for that specific project, in case I ever make a mistake again, or whenever GPT needs to specify an IP and password in a particular code.
In any chat I use, even though GPT says it doesn’t have access to the content of other chats, if it needs to reference the IP and password for the project I’m working on, it always mentions the fictitious ones.
I’ve asked it several times at different times if it saved my real data, and, in its own words, it says I can rest assured that it wasn’t saved.
Whether that’s true, I don’t know…
Jorge
But is nt all input from users traind on with most ai chats ?
And if it is the data is not secure or am i wrong
Lumo says it’s confidential. Duck.ai says mostly the same thing. I would doubt that the big AI enterprises would share what they get, not when they can sell it.
That is what I assumed … it adds all user input to its training data. So its knowledge grows but it could be fed fake information.
also
It has to get its initial training data from somewhere … anything that has ever been digitised may have been used . What is a serious omission from its trainig data is older publications that have never been digitised.
It at least saves your input during a session… because in answering it refers back to previous inputs.
I also like DuckDuckGo for cloud AI usage, it also has a feature that allows you to delete completely any of your queries, which is nice. The important thing is that they say they do not train their model on your queries. I have a Proton account, but completely blanked on Lumo, need to try that out.
I also host a local instance of Ollama to use when I am coding. Since it is hosted locally, it is free and private, and I usually use Codestral with it. Resource intensive to run an AI locally though, it can take a while to return an answer depending on your hardware. I haven’t tried it yet on either of my older linux laptops, but only because their hardware is not the best. Seems to work decently on my m3 macbook, which it better, I spent waaaay too much money on the thing a couple years ago…
Hi @james3k ,
That makes me feel happier . I think I trust DDG more than most .
You are using AI to help with coding. I would like to try that… probably in R or C.
Welcome to the forum
Regards
Neville
AI can help a whole lot in learning, or expanding, your knowledge of a specific coding language.
Word of caution, since most LLMs can code out whole modules, or even programs if given the correct prompts, I have seen (and heard of) of many software developers ability to build out and engineer decent code wither on the vine (so to speak) when they use AI to do the work for them. And, LLMs are still (and likely will be for the foreseeable future, in my opinion) not that smart. Not really anyway. They have been shown to make all sorts of hard-to-find errors, and even hallucinate and make up incorrect rules and documentation for languages.
I have been out of the workforce for about a year, and am now trying to get my skills back up to snuff. To help me out, I have been using LLMs to answer questions about various aspects of the languages as I go, and to give examples of functions or whatever. I am trying not to let the AI write the code for me though, because that kind of defeats the entire purpose.
That said, here in the US, a lot of corporate jobs REQUIRE you to heavily use AI, and track your usage. I hear some CEOs are finally starting to realize how stupid that is, but we’ll see.
I zm old and retired . my C skills are fading from lack of use.
From what you ssy, I should use AI for tiny code snippets to refresh my ability … a bit like looking up examples.