DuckDuckGo Launches DuckAssist
spreadprivacy.com
spreadprivacy.com
It can still be wrong for a variety of reasons mentioned in the post, but this is a different approach and in our validation testing (using banks of questions with known/unknown answers in Wikipedia) we know it works reasonably well, and have a lot of ideas of how to make it work much better over time. This is a beta after all, and we have plans both for improvements and additional features.
For example "what is most attractive about men?" Produces an answer, but not if asked about 'women'.
And "why do men earn more?" Doesn't activate duckassist but "why do women earn less?" Does.
Well it did load must be overwhelming at the moment.
How can you ask the LLM whether the answer is in the sentences without leaking the user's question?
Let's take a concrete example: if I ask "Is ADHD caused by non-genetic factors?", what does the LLM see? It's not clear how OpenAI does not learn from this example that user X is more likely to have/know someone with ADHD.
This cannot be asserted and dismissed so easily.
I would not be surprised if it returns content that it thinks would have or should have been in Wikipedia.
The more you constrain it (e.g., "only answer truthfully") the more it bristles and tends to react in the opposite way, because it can't actually do that.
> The multi-store model (also known as Atkinson–Shiffrin memory model) is a memory model in computer science that describes the interactions of threads through memory and their shared use of the data. It allows a compiler to perform many important optimizations.
> More info in the History and significance section of the Memory model (programming) Wikipedia article.
It seems to think that the Atkinson–Shiffrin memory model is somehow related to computer science, which it is not. It's a model of human memory. And that article it references does not once mention the Atkinson–Shiffrin memory model. At least it's easy to verify.
I really do not see any wisdom in turning search engines into answer engines. Yes, they're bespoke toys today, and you can still search the web alongside asking a LLM for an answer. But the day this becomes the default, it just becomes a countdown to the day "search" is depreciated entirely. And then rather than "doing your own research", you get to pick which Big Tech silo you trust to be the objective arbiter of truth. What a great future we're hurtling towards at breakneck speed...
hurdle: to leap over something, or to overcome an obstacle.
If you hurdle towards things with great speed I admire your dexterity. I'll stick with hurtling, thanks.
Edit: can I assume that "you're" entire comment was deliberate?
> A person is smart. People are dumb, panicky dangerous animals and you know it.
Can't get too much worse than thousands dying from a virus that has a vaccine because they're off drinking bleach and horse dewormer.
And that horse dewormer? That is just ivermectin in a paste base (e.g. hydroxypropyl cellulose and castor wax) where the human version is ivermectin with a filler-diluent, probably lactose, microcrystalline cellulose or starch. There is no special 'veterinary ivermectin factory', just some factories which use the same ivermectin as used for humans which they then mix with a paste base. I'm not convinced ivermectin actually is effective as a prophylactic against or treatment for SARS2 but it is known that it shows antiviral activity. I am fully convinced that it has far fewer and far less serious side effects than the experimental vaccines which were pushed past the traditional safeguards. I am reasonably convinced those vaccines were not effective against the later strains of SARS2 - Delta and Omicron and whatever came/comes after - so it is likely that the 'boosters' actually caused more damage than they have avoided given the fact that these were administered when the bulk of the infections were caused by Delta and Omicron.
It will take a long time for the narratives (plural, both from the 'follow the leader' as well as the 'my body my choice' side of the population) to subside so that something resembling the objective truth may be discerned so we´ll just have to be patient until then.
In other words, when you can outsource the part of cognition where you need to understand the problem, go through the ranked sources, find the right answer and digest it in right way.
This is unbelievably powerful, comparing it to horse carriages vs cars does not do it justice, we are talking about Jarvis in real life!
P.S. I have some colourful opinions about the kind of thinking thats leads to your comment but I'll keep that to myself since its likely to attract scrutiny or censure
no sign-up required!
Having to make an account is annoying. Having to install native software is a no-go for me.
MS: "IT IS OPEN FOR EVERYBODY TO TRY!"
Me: Ok, let me try it.
MS: Ahem, ok, just give us your email
Me: What? Ok, dammit, here it is.
MS: Ok, now install our browser.
Me: What?? Ok, done.
MS: Ok, now allow us to track your location.
Me: What??? No!
MS: Sorry, something went wrong.
Me: What?
MS: Sorry, something went wrong.
Me: What about the chat thing?
MS: Sorry, something went wrong.The only reason to deny access to a website for other browsers is to get your own native software on the machine of the user.
If Google would deny access to a Google website with browsers other than Chrome, everybody would be mad at them.
There are a lot of angry posts about Googles websites being less optimized for Firefox than for Chrome.
Outright banning alternative browsers is what is happening here.
The former is much more likely to find a home on my devices.
I've found it limiting to not use WhatsApp, but not that much. It's kind of like if I knew someone who is known to stalk people, but we also have many friends in common. Maybe some friends won't come to my party, but I'm still not letting them into my house.
DuckAssist answers questions by scanning a specific set of sources — for now that’s usually Wikipedia, and occasionally related sites like Britannica — using DuckDuckGo’s active indexing. Because we’re using natural language technology from OpenAI and Anthropic to summarize what we find in Wikipedia, these answers should be more directly responsive to your actual question than traditional search results or other Instant Answers.
If you enter a question that can be answered by Wikipedia into our search box, DuckAssist may appear and use AI natural language technology to anonymously generate a brief, sourced summary of what it finds in Wikipedia — right above our regular private search results.
The model is still trained on a broad corpus like ChatGPT3, i.e., the Internet. The corpus of Wikipedia alone probably isn't enough training data.
It's also the type of search where I'm sure this addition is least valuable. If I'm searching for a discrete fact that I know exists, stuff that you find on wikipedia, that's the type of search that still works fine. Instant answers are convenient but not necessary.
Often when I'm trying to solve a problem I need to break it down into more generic, smaller components, discrete pieces of information that I know exist, then search for those and recombine/recontextualize the information I got. Or branch out into new searches based on anything that was an "unknown unknown" before the initial search.
I'm hoping LLMs can help with both of these things. Allow me to search for something closer to the problem I'm actually solving rather than requiring me to break it down into components; do the recombination for me and do some of the branching for me.
If you can't do the job right and make sure the answer is right, then you shouldn't be publishing anything, or you are just damaging people and eroding trust. Returning search results in a fuzzy way I can't understand or fault you for is one thing and gives you plausible deniability, but not this. I'm all for Google alternatives (I use DDG), but this is definitely a turn off. Put the time into finding the right search result that actually has an accurate answer and just point me to that as the first result. Hell add a sublink that points to the actual answer on that page below the main result, but get rid of the gimmick that is AI.
Less hallucination is still hallucination.
Though that very well could be trademarked by Disney.
However, it's going to take me awhile to come to terms with the fact that AI makes ownership obsolete.
I wonder if their licensing allows training their own models while using Anthropic and OpenAi? Seems like a conflict of interest, but I hope they can. I have mixed feelings about instant answers keeping users from clicking through to the sites where the content was originally posted, but this is going to help a lot of people.
"Do cows sleep standing up?" gave an answer of some sort, but very unnatural and not very helpful.
It is using inference somewhat at least.
I expect their AI to be heavily lobotomized to the point of being completely useless.
Try, for example:
where is bill watterson?
in both and compare for yourself!
....proceeds to say that ducks commonly fly together.
Thanks for the non-answer. Surprised that was their example.
Maybe concentrate on not trying to censor websites