What is you guys opinion on DuckDuckGo AI?
from CodenameDarlen@lemmy.world to privacy@lemmy.ml on 26 Jan 14:30
https://lemmy.world/post/42194478
from CodenameDarlen@lemmy.world to privacy@lemmy.ml on 26 Jan 14:30
https://lemmy.world/post/42194478
Do you think it really doesn’t train on your data?
I’ve been using it and it looks good so far, I just ask simple questions and never let the context get too big.
It’s good that it doesn’t require login, just open and ask something.
threaded - newest
Interesting to see comments on this. I’m a bit suspicious, understanding the costs involved.
I don’t want AI anywhere near my daily drivers. If I want to use AI it needs to be siloed and unable to access any of my data unless I explicitly feed something into it.
Hijacking your comment to remind others of noai.duckduckgo.com The more that is used, the faster DDG may remove AI from the main page. (Hopes and dreams…)
Neat, in IronFox, that is already one of the search providers to choose from to set as the default. I had to manually add it to KISS, bit it is working great
<img alt="" src="https://quokk.au/static/media/posts/Dq/H0/DqH0qnLG3Pn0V1G.webp">
Great idea! Thanks for showing how to do it!
👏 Thanks.
I switched entirely after they renamed their app to AI. Hopefully that helps even more.
any ideas how to add that engine to Vanadium on GrapheneOS?
I’m talking about the duck.ai not the integrated AI on search results, I don’t use that one, I let it disabled.
I’m real sorry! Not to be a complete ass. But the sooner people stop using the “daily driver” phrase the better! Oceans of cringe. Please, please delete this from your vocab! Its like throwing up in the mouth every time I see this.
What would ypu suggest they use instead?
This is a ‘you’ problem
Good news: “AI” doesn’t actually exist.
True, but nowadays, and for foreseeable future, AI means what the tech bros mean by it.
From ArsTechnica:
So there is some trust involved, but I’m inclined to believe DDG.
It’s fairly bare bones in all honesty. It does sum up initial searches fairly well but beyond that you’re not going to get much, if anything, out of it. Asking it follow up questions regarding a search is more miss than hit.
It has provided me decent initial results when I’m searching for some random linux question/solution but clarifying or expanding on initial results is useless.
That has been my experience as well. I use it extensively for searches, not for chatting.
Everything you type in the chat box is sent to the LLM provider but they get Duck Duck Go’s IP instead of yours.
So if you type personal things its mostly just like typing them directly to ChatGPT. However, with duck.ai your IP, Browser info, Location (if shared), etc is seen by Duck Duck Go instead of OpenAI.
I don’t think DuckDuckGo is lying when they say that they don’t use your chats to train models. However, that leaves plenty for OpenAI and Duck Duck Go to do with your chats, like building shadow profiles.
I suggest that if you want to be anonymous to Duck Duck Go, then use duck.ai via vpn or tor. Always assume the content of your chat session is being logged by the LLM provider.
What do you mean by “shadow profile”?
Like a dossier which identifies you only by some ID which is used to compile data about you but never includes your direct personal info (name, email, home address, mobile number) so they don’t have to tell you or ever delete it, even under laws like GDPR.
How can they tie it to me tho? Or use it against me? Especially if behind ddg “proxy”
They can tie it to you by cross-referencing all the signals they have about you with data collected from other data collectors and aggregators. With enough data they can connect things like browser fingerprints and so-called anonymous ad IDs with your real identity.
Keep in mind that there is a good number of technical people whose job it is everyday to continuously figure out new ways to track everything they can about every person they can. These data collector and brokers have demonstrated time and again that they don’t really care about following the rules either. Here’s a good resource for more info https://noyb.eu/en
In terms of how they use it against you, this is some good info and it applies even if you aren’t American https://epic.org/issues/consumer-privacy/data-brokers/
Oh well all of this explains why I sometimes get instagram ADS which are relevant to me on my work phone even if I looked for stuff on my PC browser or smartphone (all using adblocks, deegogled android, private DNS, tracker blockers, private browsers and other preventive measures).
I didn’t realize they could literally track you probabilistically or they could tie different devices to you…
So essentially if I use my amazon account on the same PC (or on a device tied to me) that I use for looking at “cat food” I am screwed and they will know I have a cat and amazon will start advertising cat stuff to me?
I did know they tracked you, but I thought it had to be a consistent set of datas: accounts, unprotected browsing, keeping cookies for a long time etc etc… I didn’t know they could probabilistically try to catch you nor that they could so reliably tie devices togheter… I would like to know more about what they can actually do and what are preventive measures that actually works…
For example: can they (and how) get over tracker blockers? VPNs? Proxies? Private DNS? Degoogled devices/Linux?
What I’ve got to do if I want to be on the internet preserving my privacy? Should I literally stop using the internet? Should I use devices on which I do not login on a normal account ever and just use my self hosted stuff or the federated web?
And what about all the data they already have on me, it will be their’s forever??
This is getting ridiculous, we need new solutions, the internet as we now know it is completely screwed.
If you want to be extreme, you only use devices that you control all the software on and you are very careful about what software you run. You always use a killswitch vpn. You choose carefully the websites you use and you use all the standard counter measures (ublockorigin,DoH,uMatrix,pihole,etc) at all times. You keep another laptop that has whatever you need to install to be able to use your bank and the online shopping you can’t live without. You use it for nothing else.
Even then its not perfect. I’m pretty sure that all Android, iOS, and Windows devices track the wifi access point name and mac address of all the ones you use. They also track the location of all the access points as seen by everyone elses location enabled devices. Easy for them to combine that to basically know where every visible wifi in the world is.
Don’t want it, don’t need it.
It’s a frontend. Of course they don’t, it’s not their AI.
The GPT OSS can be self hosted tho.
I guess if you can trust ddg using ms search w/o sending information back, you can do the same with ai?
*Oh, ya mean the conversational AI.
I've totally switched over to noai.
i try not to use AI too often, but when i do have a question i like the interchangeability of the models, so when I feel that one model is too lobotomised to get my question i change it for another and compare it. i trust what ddg says about respecting privacy, i don’t trust the backend llm suppliers though, i don’t have any illusions about that.
i also think the premium through duck.ai option could be interesting if you are a professional user because it still offers the interchangeability option.
but still, it should not be integrated on the main search page, and search should be revamped. normally the first results i get are for some social media presence of a restaurant somewhere halfway around the world instead of the wikipedia article on the main subject that made the name famous in the first place.
also the domain owners of duck.au, duck.si and others are getting a lot of traffic
How do you check if a domain is getting traffic? I thought only the owner could see that
my current sample size of: 01 person(s) indicates that a 100% of users that participated in the study have at least once wrongly entered the web-domain address and erroneously navigated to duck.si and duck.au
I honestly just type “duc” then let the browser auto complete it. After you enter “duck.ai” correctly a few times
They got all “i am a app now!” since yesterday and don’t work on my firefox profile anymore. That’s all.
It’s working fine on IronFiox + uBO on Lineage to me
I actually like the search overviews, they yap a lot less than Google and usually give the correct answer within the first sentence
I hate AI being shoved into everything. I’ve been using noai.duckduckgo.com and am very happy with it.
Is that an official noAI version of DDG?
Yes it is. No AI results by default.
I like it. There are some questions I need a quick answer for. Code syntax for example. I don’t need to read stack exchange. I just need the one quick thing.
No thx.
If you are talking about free ai, the cost is your data.
Just use 4get. It’s a privacy focused proxy search engine that supports multiple backends. No telemetry, no tracking, no bloat.
Repository: git.lolcat.ca/lolcat/4get
I’ve been using Duck.AI because you don’t need to sign in or anything. Works well
GenAI is cancer.
It’s good sometimes when queries aren’t getting þe right results. Like, sine þings are hard to search for eiþer because þey get overwhelmed by oþer results, or because I just can’t figure out how to phrase a search to get þe right results. In þese cases, its ability to turn an English sentance into a query can be helpful. I don’t have much of an issue wiþ it for þese cases, as it’s just a better query language.
It’s terrible for answering questions. It is regularly simply wrong. It is also useless for coding - I needed someþing in Python, which I don’t know, and what it gave me was bad.
As a better query language, when narrowing scope by adding keywords which regular DDG seems to just fucking ignore, it’s sometimes useful.
It needs to PISS OFF.
It should be off by default. As it’s set up right now, DDG settings don’t keep on privacy respecting browsers due to cookies being cleared regularly. Since their AI is on by default, that means it regularly gets shoved in your face.
For you and whomever else needs to hear it, you can solve the second part by going to duckduckgo.com/settings and saving the bookmark with all the options included. Then you can clear all the cookies you want and it’ll always load the same settings.
If you’re using cookies. In general noai.duckduckgo.com is the way to go otherwise
I’d prefer this route too, but my phone browser doesn’t allow users to add custom default search engines
Which phone browser? You could always set it as your homepage to it if that’s where you usually start out
Sorry, I wasn’t as clear in my original post.
At the bottom of the settings page you can generate a “bookmarklet” that includes the config options in your URL. So duckduckgo.com becomes duckduckgo.com/?kae=t if I want the Terminal color scheme. And this works without cookies. And this applies for all possible settings.
Interestingly I set my home page to a string that should disable all the AI features and I still got summaries sometimes so not sure if that’s a bug, user error, or hostile UI. In either case, something like you suggest with noai.duckduckgo.com/?kae=t&etc=… should give you the best of both worlds.
I like this, thank you for the insight
There is no such thing as “AI”.
But I appreciate the generated summaries of search results. Sometimes they miss the point of my search, but they’re often quicker than searching through the webpage results.
I don’t know why people get so mad about it.
Because of rampant hallucinations, yet people taking them as gospel. Not to mention the energy cost for no real benefit
The generative summaries have the potential to take clicks / visitors away from the sites they’re from. I’ve seen reports of smaller sites being at risk of closing down as a result, and if there are no sites to summarise…
.
I use it reluctantly and only to un-fuck the fucked up search. if there were a good search engine I wouldn’t use the thing.
I don’t trust DDG, really. I run LLAMA locally. It serves my purpose.
How much params can you run on your hardware?
3B to 4B on phone.
They constantly measure DomRect using javascript, which is a unique hardware-based metric that can be used to track individual users.
Imagine the cost of running duck.ai. What exactly is the revenue that it brings in?
Of course, if it were some honeypot, using DomRects to track users (and DomRect is not protected by Tor Browser or Mullvad Browser etc), well then it doesn’t really matter if it’s not bringing in much revenue since it’s value is in being a honeypot.
Yes, DomRect can be used legitimately in coding without tracking users… but why does ddg need to use this when they know that it CAN be used to track users and users have no way to audit the servers?
It’s really interesting they measure DomRect and not Canvas when privacy-aware users often block canvas fingerprinting but don’t block DomRect.
It’s sus
Duck.ai is a saas, I can never be 100% sure that information sent to it is private, the only way to use an LLM privately is to run it locally.
That is very unlikely, duck.ai doesn’t brew it’s own in-house AI, they run models made by third parties like Mistral, Facebook and openAI.
As far as non-local LLM inferencing goes, I think duck. Ai offers the most privacy-friendly service.
While it’s impossible to warranty privacy, you can warranty anonymity, because duck.ai is accessible over tor.