deepseek is open source, so you can just use it locally.
Quick answer: Don’t give any non-locally running non-opensource LLM’s sensitive info / private info.
and lie a lot
What does this means?
Tbh, if you don’t know what that means, you can’t trust it.
Though, it means that unless it’s running locally on your own hardware and not in the cloud and you haven’t verified the source code directly (or someone else you trust hasn’t) then assume it is nefarious and do not give it any personal or sensitive information you wouldn’t want anyone on the Internet to know.
Could you please explain what “personal or sensitive info” means?
Name, address, phone number, Bank info, nude photos of yourself, etc. If the info being released could harm you or in some way negatively impact your life, assume it would be sent to China or anywhere else on the world wide web if you ran it without following the previous guidelines.
Any data related to your person. (name, contacts, date of birth, etc.) Search “PII” or “personally identifiable information” if you want to read more about that.
‘locally-running’ means it is on your computer, will work without an internet connection
anything you access using the internet is not ‘locally-running’
The comment means don’t send information over the internet that you don’t want to share.
Thanks @[email protected] for filling OP in! I want to add a few things incase OP is unaware of more than just what you explained:
LLM = large language model, one of the types of AI. Examples: ChatGPT, DeepSeek, Meta’s LLaMA
Open-Source: the program code of the AI is available to look at, in its entirety
If you are not sure if you understand these terms and what frightful_hobgoblin said, then just assume whatever AI you are using is going to share your chat with the company behind it.
open source is also very tricky with LLMs: i’d argue if you can’t recreate it from scratch, it’s not open source… deep seek does not contain all the data necessary to recreate it from scratch: it’s open weights (the model itself can be downloaded and run) but not open source… i’d classify it as free (as in beer) software; not open source
Excellent addition, I agree!
That’s the criteria of many FOSS catalogue repositories: they won’t add any software that is not completely reproducible.
Depends on what you ask.
Go ask it about NATO or Tienanmen Square and see what happens. The data model is heavily redacted, filtered, suppressed, biased…
So if you ask it a question, it will always be pro-China/anti-America. It also changes responses on the fly to fit with Chinese law, which includes denying the Tienanmen Square massacre, and other historic events and even goes as far as to imply or outright say they never happened at all.
So can the content be trusted? Not really.
I personally just want to get chibese translations, but I don’t if it’s worth it anymore…
Refer to my other comments above. Self-hosting it removes censorship and bias. It’s only biased as long as it’s on Chinese servers and therefore following Chinese law.
DeepSeek has some of the most syntactically correct and accurate English to Chinese translations I’ve ever seen–so it’s super useful for that.
This is incorrect. This only applies if not hosted locally. I host it myself it has none of these restrictions. If you’re using it from their app or website it’s hosted in China and must follow Chinese law.
If you’re using it from their app or website it’s hosted in China and must follow Chinese law.
This is literally what I’ve just said…
It also changes responses on the fly to fit with Chinese law. You called what Is aid wrong, and then immediately exactly reiterated what I’ve said…
Why? What do you get out of it?
I suppose if that line is a catch-all, sure. Your message didn’t make it clear that self-hosting removes Chinese bias and censorship. This is an important bit of information for OPs question, and what I get out of it is a valid and important addition to the conversation. I genuinely don’t know why you’re defensive. Being incorrect, or I suppose in this case, lacking nuance, isn’t a character flaw. I do it all the time.
Anything that is not local AI cannot be trusted.
Have you ever thought to yourself, where the fuck do these corporations get the funding to make me use such a service for free? By harvesting your data and selling it.
From your other comment i saw you aren’t using a PC, i haven’t tested this out but you may be interested in it (local LLM and android only): https://github.com/Vali-98/ChatterUI
Best of luck to you.
👀 I was looking for something like this for ages, eventually I had given up and assumed Kobold.cpp on desktop was the only choice. Thank you for the link!
You can’t trust anything.
You always have to use trustless software.
‘Trusting’ is privacy-by-policy.
Trustlessness is privacy-by-design.
Deepseek’s models can be run truslessly locally, or can be hosted on a server.
Wait were you talking about privacy or fact-checking? LLMs don’t stick to the truth.
You can’t trust anything
Yeah, this is exactly why I put it in quotes. And I know that LLMs don’t know what is true.
Can you expand on your question a bit? What is it you’re trying to find out?
Does DeepDeek respects my privacy or will it track my every move?
Here’s what you can trust: https://lmstudio.ai/
Otherwise, only ask it generic coding questions that any student studying your topic would, and then there would be nothing to distrust over.
Looks very similar to “Jan”. I wonder if it’s a fork (or the other way around).
Probably not a fork as Jan is AGPL and LM Studio is mostly proprietary
Interesting, this is my first time learning about Jan, and frankly, it looks better! I think I’ll pivot…
Seems interesting, but I don’t have a computer to run it…
Then try: https://gpt4all.io/index.html
It won’t be strong, though. I’d just probably use DeepSeek but restrict questions to generic ones, like how to build a spreadsheet formula or something. Don’t ask for life advice lol; do that here in [email protected] or [email protected].
Then try: https://gpt4all.io/index.html
Thank you for the suggestion, but I don’t get it, does it runs on the website or has a downloadable version for Android?
Neither; both of these are downloaded for offline running on PCs only. Androids aren’t typically powerful enough for any meaningful offline AI.
