God damn it people. I just want to search the web. You aren't doing anything meaningful by giving me yet another fuzzy robot question machine. Just let me search the web. Queries in, web pages out.
I just don't want to search the web. I have questions and I want answers- searching the web is an intermediate step, not the end solution.
I have Question X and the answer isn't to surface "Blog Post Y"... it's the content of Blog Post Y I need, and only the subset of the content that answers my question, perhaps synthesized with the content of 5 other blog posts, a couple scientific papers and a recent press release.
We're now in this temporary zone where "just searching the web" is still the optimal way to get answers, while search engine companies are piling into other stuff.
But very shortly that other stuff will become our go-to.
Sometimes, often, I really just want to search the web. I just want to see what's out there! I have some vague ideas that I am trying to make more coherent and give sharper edges too, and I'm exploring the space those ideas fit into to see what others have done with them. Often I don't have a specific question in mind, and trying to frame my query as a coherent question will be limiting.
As another poster commented, don't take away my agency by trying to predict my intent! Better yet, don't try to force intent on me.
Arguably Google has been taking your agency away and predicting your intent for the last decade at a minimum. The only difference here seems to be that you’re noticing it.
You're not wrong in terms of the average user's intent (they mostly don't care about the source, they care about the content), but I see this as an ethical gray area when search engines effectively offer up all this content created by bloggers without the need of the user to visit the site itself.
Phind tries to ameliorate this a bit by providing a set of links where the summarized data comes from, but most people likely won't even click on these links.
It's going to be rather demotivating when your site's traffic suddenly gets cut in half.
At this point, anybody here likely has had the privilege of being educated to prefer primary sources of information over secondary, tertiary, etc.
The indexing feature that search engines for years provided was a side-effect and benefit of the internet’s success.
There have always been challenges with this as it is not immediately clear what should be treated as “primary source” information — it takes time and skill — the user in this model is accountable for their own exploration and consumption of other peoples’ knowledge.
AI seems to be purposefully at odds with that. So, for me, the distrust and avoidance is well-placed.
That said, it’s a powerful tool that has a place in the world, but I hope it doesn’t come at a detriment to the colloquial understanding of what it means to “search the web”
My main fear with AI becoming our go-to information retrieval platform is how can we be sure that the answer is objective truth without any slant? Shouldn't there be an ethical obligation that a model should not only fact check but also provide alternative theories? We lost that somewhere in journalism but we could regain that if we prioritized that the thinking and deciding still needs to occur human-side.
Go try to ask any of these hosted LLM bots about sexual topics to find out how far from objectivity we are. I earned a stern warning after trying to write a dirty pun with ChatGPT
Imagine trying to do research on sexually transmitted diseases, or trying to self-educate about sex. Unfortunately the limits put in place are quite coarse today.
This (DuckDuckGo AI Chat) is currently separate from search results, and in any case we plan to offer settings to turn off any AI search integrations, similarly for how we do for instant answers and ads.
I really like the way this has been added to the UI. It's off to the side so I can just use DDG as God intended if I want to, but also is distinct enough that I discovered it prior to seeing it on HN. It makes it clear this is done in a DDG private way (which is one of the things I'm wary about with ChatGPT on its own) and offers multiple models/vendors in one place.
Why ddg is not focusing on improving or innovating the search ?
DDG dosen't even have lens/goggles yet . Can't ddg just try to innovate like marginalia/kagi/Occam.
If you could stuff all webpages to a vector database and give us a nice RAG that just responds to questions about anything that could be found on Internet while generating links to the source, that would be even awesomer!
A lot of the time you aren't searching for "web pages". You are searching for answers, to get to which you'd have to dig through piles of web pages for an hour. Technology already exists to give you a direct answer in 95% of the time (with references), it just results in fewer ads shown to you, so search engines aren't interested in integrating it properly.
This you-know-better-than-me mindset is why all technology is converging to garbage. I can't speak for OP, but for me, you are half correct: I am indeed ultimately searching for answers, but I want the piles of web pages for an hour, and that is how I will find my answer. I don't like it when systems remove my agency by believing they can correctly predict my intent. (They can't.)
But there is no agency being removed here. You are being given more options by adding a button to chat if you so desire. It’s very small and unobtrusive
Ads, sponsoring and SEO spam were also very unobtrusive for a while. A few decades passed and we have to sift trough all the garbage to find anything relevant in our searches, it's very likely that search engines will push harder and harder for these AI assistants while making finding organic content and authoritative sources even harder, resulting in it being nearly impossible to do research by yourself rather then trough an AI that has god knows what bias and perhaps subtle ad driven bias into it.
In a lot of cases today’s AI systems can easily predict your intent once they accumulate enough context about you. Not in all, but when they predict incorrectly it’s easy to correct them using natural language. And nobody is removing the option of digging in the pile, it’s just that I don’t see why anyone would want to, in most cases. I certainly don’t - I got better things to do with my time.
There is an inverse relation between sites with ads, and sites that contain useful information that require time to look through. Open access journals, StackExchange, product support messageboards....none of these have ads.
In fact, it's the sites peddling trivial BS that are chock to the brim with ads.
This is an intermediary step to what will ultimately upend search: LLMs parsing web search results and giving you succinct results without you having to read them yourself.
If you haven't tried Arc Search's "Browse for Me" feature, you should. It isn't perfect yet, but I think it's clear that's the direction of search in the future.
The issue is LLM’s also generating the web results. Garbage in, garbage out. The real problem is maintaining the ability to find the ground of real knowledge.
Garbage content on the web existed long before LLMs were the new hot thing on the street. There are ways to weight or even exclude results that are poor quality, and that'll still be necessary.
Agreed. Additionally, I don’t actually want to talk to my computer for most use cases. When I do, then I’ll use a dedicated service like ChatGPT, but for most other take, I really really don’t want a chat interface.
I wish these companies would focus on making their core product better instead of spending resources on AI chat interfaces.
I'm quite a fan of how Kagi does it: they'll only give you back AI answers if you end your query with a question mark. For when I'm looking for a fact, it's very helpful and never gets in the way.
I'm a current Kagi subscriber and agree the search is excellent. Unfortunately Kagi is also wasting their time on AI bullshit, so I suspect it's a short-lived oasis.
What's the business model on this? GPT3.5 and Claude 1.2 aren't as expensive as other models, but they're not cheap. Is this going to get pulled later, monetised, etc?
It's concerning in general to see so many companies throwing so much money at this without business models. The big names mostly have either premium tiers or paid API usage, that's fine, but DDG/Brave/BingChat/Arc/etc are all just giving away money, and when the music stops it's possible consumers will be left high and dry.
The plan is to maintain free anonymous access to base models, with eventually search better integrated as well. We'd also like to experiment with providing anonymous access to more cutting-edge models as part of Privacy Pro.
This feels like a corpspeak non answer to the serious question above. This move will increase operating costs by a considerable amount. Is the company sufficiently profitable that it can eat the loss or is this a gamble that culminates in a PE takeover? I get that duck duck go is in a tough position since LLM search is increasing popular among users, but without a business model that balances out this feels dubious. E.G this is the marketing campaign before everyone is going to be herded into the paid tier Privacy Pro?
Yes, we are sufficiently profitable and have in fact been profitable for a decade. Our mission is to provide easy access to privacy, and offering things free where we can is key to doing that well. Additionally, we expect the forthcoming search integrations to end up at least break-even in any case.
Considering the speed at which models are improving, and more importantly, the speed at which models have begun improving per unit compute, I would bet that AI models at least as good as ChatGPT 3.5 will be free to use like this forever.
Right now there is so much interest/demand, and so much money sloshing around to provide it, that it's easy to find.
By the time the AI market cools off a little bit, I would expect that the cost to run a model meeting the current quality of ChatGPT 3.5 will be next-to-free.
Will near-state-of-the-art models be free forever? Absolutely not. But state-of-the-art circa late 2022 models are probably the quality basement, and free models will be at least that good.
Being DDG, with its reputation, I’m sure these chats really are never saved - but the wording in the sidebar:
never saved by us, and not used to train AI models
could make the more paranoid believe that they could be saved by DDG’s partners - from very directly (Anthropic saves chats with Claude) to indirectly (all channeled through a shady arms-length subsidiary).
Additionally, all metadata that contains personal information (for example, your IP address) is obfuscated from underlying model providers (for example, OpenAI, Anthropic).
If you submit personal information in your Prompts, it may be reproduced in the Outputs, but no one can tell whether it was you personally submitting the Prompts or someone else.
We have agreements with model providers to further protect your privacy.
As noted above, we call model providers on your behalf so your personal information (for example, IP address) is not exposed to them. In addition, we have agreements in place with all model providers that further limit how they can use data from these anonymous requests that includes not using Prompts and Outputs to develop or improve their models as well as deleting all information received within 30 days."
> (for example, your IP address) is obfuscated from underlying model providers
> so your personal information (for example, IP address) is not exposed to them
Is this not two different things? The first one saying that the IP address is changed in some way to make it unidentifiable. The second one is saying it's just not sent.
That’s a pretty disingenuous framing. Obviously the content of the communication- what you ask the chatbot - can totally be personally identifiable, and is stored.
You aren’t providing communications content privacy, you’re providing a meta data proxy which is not remotely the same.
If you submit personal information in your Prompts, it may be reproduced in the Outputs, but no one can tell whether it was you personally submitting the Prompts or someone else.
True, but it's also the best possible thing until they can run an LLM in your browser (or maybe we figure out how to do homomorphic encryption of neural networks).
What else should they call it? I suppose "anonymous" could work (i.e. they only know about you what you tell them).
Correct, that would be full communications content protection. This is not that, this is a pretty half assed product with a misleading set of promises.
Either use one of the many open models that beat GPT-3.5 and you have a real, and better!, product.
I literally have no idea what the point of this is, or why not make the actual step of rolling out a real private LLM. It’s literally inferior to what Mixtral provides right now, under Apache license.
DDG never pretended to be unbiased. They have all the same policy decisions around artificial boosting/throttling as google without the tracking.
In their golden era search engines were essentially super sophisticated keyword matching. Users used to find novel websites from passionate people. Now it's becoming more of a social media feed of the same 100 website where you train it what you like.
This is exactly why search is so vulnerable to disruption from LLMs.
>to hiding websites in search due to political bias
This is a gross mischaracterization. Every search engine down weights domains and pages with bad or inaccurate information. Their entire purpose is to find good information, not penis enlargement 4 u .ru pages.
You're getting downvoted pretty heavily. Do you have sources for that? I did a quick search on Kagi and couldn't find anything to support what you say.
"Welcome to DuckDuckGo AI Chat! This chat session is powered by OpenAI’s GPT-3.5 Turbo. All of your chats here are private, and are never stored by DuckDuckGo or used to train AI models."
Wait, they are sent to OpenAI but they are also "private"? What am I missing here?
>Additionally, all metadata that contains personal information (for example, your IP address) is obfuscated from underlying model providers (for example, OpenAI, Anthropic).
>In addition, we have agreements in place with all model providers that further limit how they can use data from these anonymous requests that includes not using Prompts and Outputs to develop or improve their models as well as deleting all information received within 30 days.
Everyone has the same agreement. Duck duck go is grasping at straws here because their business model is completely crushed if they don't have a privacy upper hand or a competitive product. They will be forced into a pivot they can't succeed in.
The messaging could be more clear, but is this substantively different from how search queries are handled when DuckDuckGo forwards them to Bing? Bing doesn’t know who made the search but could log the query itself.
Well it's a big deal to me. OpenAI have not conducted themselves in a way that makes me want to share my search intents with them, and I don't care if they or DDG tell me that what I tell them "won't be used to train AI models".
Interesting they use the word 'private' in their strapline, after all they frame themselves as private search albeit they don't have their own search engine and market themselves as a privacy search engine. Their index is essentially Bing and they're hosted on Microsoft IPs. So not exactly ideal.
But none of your personal details are sent to Bing? Sure they get some SERPS from them, but DDG abstracts in front of Bing's API using your query without passing your data.
That's one way of looking at it. Another way is that your PII like your IP hitting a MS server and searches being returned for that searcher are heavily correlated. They can't serve results like https://duckduckgo.com/?t=ftsa&q=shops+near+me&ia=web and return shops within miles of me without passing that info on.
IIRC they pass the /24 of your IP, but all the same, not much better.
Does DDG provide Bing with your location (/24 or some other proxy) always, or only in the case of the local search in your example? Which seems like a minimal privacy violation (Can’t have “What’s nearby” and “I don’t want to tell you where I am”, but I get your beef; you want DDG to localize results instead of M$).
No, your IP is not "you". It's most definitely not your PII, per se. It is PII when used in conjunction with other data(even under fairly strict GPDR rules).
The party storing the IP must have means of associating that IP with you, not simply "a city".
I mean... I can write an anonymous letter, drop it into the mailbox of my local post office and there will be a stamp of where and when that letter was mailed from. Is a post office stamp now my personal information?
When you have the /24 of a searcher, and assets on X% of websites where that same IP accesses clicking through from a search, I mean, it's well known that this is not very private.
And also doing xx searches a day. The idea of anonymity becomes a fantasy.
Sure, in the case of this specific tool. But the point of something being demonstrably 'private' is diminished based on your usage of the word in other contexts.
And again, hitting Microsoft IPs. Microsoft have assets on many websites, so the correlation is trivial over time, no specific IP needed, no cookies needed.
No. That's utterly wrong interpretation, according to ECJ.
IP is not PII per se(on it's own, if you don't understand that phrase)
If it's hard to understand, it's the same with phone numbers. Your phone number is PII, yet in every country with an option to move your phone number to another phone line provider - they are all very much not private... And telco can check if your phone number is in use, and not only telco. Phone numbers are significantly more private than IPs.
I don't buy into a future where I'm spoon-fed an answer with no way to verify it. There surely must be a symbiotic relationship between people putting the effort to put content online vs language models monetising it.
I didn't know Claude until I saw it as an option for DDG chat model. What caught my attention was the tagline "instant". Interesting. I decided to give it a try with expectations to see answers coming up almost instantly. And they really did. I'm enjoying it so far.
For what it's worth, Claude instant is part of Anthropic's first generation of models. There have been two more since then. Today (3rd gen) they have haiku, sonnet, opus. Worth checking out. Opus is competitive (many say better) than GPT-4
Sounds great. I'll keep an eye on it. I'm assuming Haiku performs better than Instant, thus the upgrade. I guess I'll be super impressed in the coming weeks.
I tried this, and it seems like this is just raw AI chat, no search?
I do extremely like the Kagi fastgpt feature, which lets you end your search query with a question mark, and above the search results you'll get an AI answer, that is based on the content of the top 4-8 results.
It's very fast and provides citations to the links it used, so it feels quite trustworthy (or at least as trustworthy as the sources it cites), which means I often use the feature and am happy with the response I get.
The first three, usually. PVC is extremely common in plumbing and drainage, and there the type of pipe is usually specified by material type, not use case. PVC cloth is less common, and I'd guess you'd qualify with a use case. ("pvc cloth for sewing"; it's often called "vinyl" instead).
Anyway, it's an intentionally hard query. I tried "tourism in mountain view, ca" first. They both do fine (though DDG doesn't provide links), since there's enough on the internet about that for the base GPT-3.5 model to recall it without additional context.
While Brave Search does include AI search results, they also include Leo in their browser which is a full chat interface as well. Baked right into the browser and sidebar.
Who, exactly, wanted to ask a search engine questions that would be answered by bullshit?
I quote:
what is the stoichiometric ratio between 2NO2 and N2O4
The stoichiometric ratio between nitrogen dioxide (NO2) and dinitrogen tetroxide (N2O4) is 1:1. This means that for every one molecule of nitrogen dioxide (NO2) that reacts, one molecule of dinitrogen tetroxide (N2O4) is formed. The equilibrium reaction between nitrogen dioxide and dinitrogen tetroxide can be represented as:
2NO2 ⇌ N2O4
In this reaction, two molecules of nitrogen dioxide can form one molecule of dinitrogen tetroxide, and vice versa.
AI Chat is separate from the search engine, providing anonymous access to popular models. On the search engine we have DuckAssist, which is currently grounded in Wikipedia (more sources coming): https://duckduckgo.com/?q=what+is+the+stoichiometric+ratio+b...
>Additionally, all metadata that contains personal information (for example, your IP address) is obfuscated from underlying model providers (for example, OpenAI, Anthropic).
They are acting as a proxy to increase privacy, which is pretty much their entire business model.
I have been using DDG as my primary search engine for several years. Their search results is bad now. Not sure if it is deteriorating or it was always bad but I put up with it. DDG could have adopted the Kagi way, let users rank their result, or just genuinely improve the search result. And no, LLM won't help us as far as factual information is concerned.
Have you been using the feedback button, located (unfortunately) in the bottom right corner of the results page? I've been told they are read by actual humans
I'm with you that the results are sometimes just comically bad, but there are a lot of 80/20 products in my life, and they do offer the bangs to rerun the search on an alternate engine
Pretty disapointed that this is just a proxy in front of Claude and ChatGPT, I thought it would be some kind of DDG-hosted llama3 8B since the quality is now good enough with a 8B model.
My complaint isn't about more models in fact, it's about not depending on OpenAI (and not feeding them with data and money), so I would rather have less models (just llama3) rather than more
I stopped using DuckDuckGo long ago until they stopped being what they were supposed to be. Obviously, I'm not alone in this, as they are looking for financial means to stay afloat!
It really does show the absurdity of the current discourse around sex and gender when the definition of woman is considered to be "inflammatory political speech".
People specifically use DDG for the sense of privacy they have established, and that works be shattered by a buyout. People who use DDG are savvy enough to change their default search engine, and therefore would not have trouble doing it again.
Give me another one then? It’s a search engine with no search index. It’s an ads supported site without an ads network. What is it then if not a wrapper?
I’m saying it’s not worth anything because it is a Bing wrapper. The other “products” are just to drive more traffic to Bing ads. There’s no there there.
Come on, this is the kind of project that could be done by an intern in a day or two. Not much of a waste and it's nice to be able to access models without login or history.
Most people use DDG to escape the general search enshittification of Google and Bing, so this feels like a very bad idea. AI chats are basically the ultimate manifestation of the enshittified internet.
Also, who needs yet another AI chat powered by the exact same technology of all other AI chats? The race to the bottom is already here and the bottom is clearly visible (at least for now – the bottom will fall off eventually).
Aside from the potential for privacy issues anti-thetical to DDG's own stated goals, another problem is these chat AIs can just be inaccurate, if not just outright incorrect, on the simplest topics. Something which would seem to be a risk to DuckDuckGo's reputation as a search provider.
