|
Ruffian Price posted:I would prefer a search engine to search the web
|
# ? Apr 12, 2024 14:42 |
|
|
# ? Apr 30, 2024 07:43 |
|
Ruffian Price posted:I would prefer a search engine to search the web
|
# ? Apr 12, 2024 14:43 |
|
Then y'all might like the most killer feature about it: It's easy to turn off! https://duckduckgo.com/settings#duckassist
|
# ? Apr 12, 2024 14:46 |
|
To be less of a dick about it, the validation seems to work - at least I've only got prompts to summarize the relevant Wikipedia article, with more ambiguous questions not triggering a DuckAssist window at all. Even if I don't like it personally (but I'm a dumbass one-tool-per-function purist and would also like it if unit conversions were a pure keyword search unless you click the "Computation" tab in DDG) it does appear to be the least poo poo implementation for now
|
# ? Apr 12, 2024 14:53 |
|
Yeah, why all of this LLM-in-search stuff doesn't work for me is that largely I don't use web searches to find answers, I use them to find websites.
|
# ? Apr 12, 2024 15:02 |
|
On top of that I fundamentally resent the application of LLMs to try and fix the problem of search rot which was created by SEO and other automated systems in the first place.
|
# ? Apr 12, 2024 15:07 |
|
shoeberto posted:We just re-launched DuckAssist, which is our attempt at a responsible LLM question-answering service: https://duckduckgo.com/?q=how+are+fursuits+made&t=newext&atb=v301-1&ia=web The link you've given just gives me AI output on what a fursuit is but does not give me the answer to how it's made and then links me to Wikipedia for more info. It'd be more useful to just give me a wikipedia link.
|
# ? Apr 12, 2024 15:14 |
|
mllaneza posted:My goon, have you heard the good news of Miami Vice? Is there a new one? I've only seen the og and the movie
|
# ? Apr 12, 2024 15:19 |
|
shoeberto posted:We just re-launched DuckAssist, which is our attempt at a responsible LLM question-answering service: https://duckduckgo.com/?q=how+are+fursuits+made&t=newext&atb=v301-1&ia=web Your own example doesn't actually answer the question. Moreover, it's not really doing anything useful at all - it seems to be just searching Wikipedia for the search term, finding the relevant section on the page, then feeding the result to an LLM to slightly reword it. All of that text is just drawn directly from what's on the Wikipedia article, and there's no loving point in having that right next to the standard Wikipedia blurb-and-link. It doesn't seem to me that it's especially useful for anyone - and more importantly, it doesn't seem to be anywhere near being worth the compute cost to do that! And that's without even getting into the numerous ethical issues in LLM usage. Rather than get further into that, though, I have a question for you: Why? Why are you trying to cram a LLM into a search engine? Especially a search engine whose selling point has always been its lack of bullshit?
|
# ? Apr 12, 2024 15:46 |
|
shoeberto posted:We just re-launched DuckAssist, which is our attempt at a responsible LLM question-answering service: https://duckduckgo.com/?q=how+are+fursuits+made&t=newext&atb=v301-1&ia=web Is this a SYQ or do you actually work for DDG?
|
# ? Apr 12, 2024 15:49 |
|
Also I think LLM for search is terrible because all the smart people keep saying "LLMs are not knowledge databases" but then all the dumbfucks with the AI budgets cram LLMs into everything specifically to act like a knowledge database. It would not surprise me if the vast majority of people thought the "Gen" in "GenAI" stands for "general" and not "generative" and I have no doubt that this confusion isn't done on purpose just to keep making money number go up.
|
# ? Apr 12, 2024 15:53 |
|
Main Paineframe posted:Rather than get further into that, though, I have a question for you: Why? Why are you trying to cram a LLM into a search engine? Especially a search engine whose selling point has always been its lack of bullshit? I'm limited in what I can discuss openly but I would say my personal curiosity is answering this question - like are we building something that people actually want to use? From a product perspective, the idea of the instant answer is to give users information with zero clicks required. So if you search "where to watch mad men" you get direct streaming links, as an example - we have a lot of those types of answers on the results page. The motivation here is based on an assumption that LLMs can help serve that sort of zero-click instant answer purpose for a more general use case. Boris Galerkin posted:Is this a SYQ or do you actually work for DDG? I don't know what SYQ means but I do actually work for DDG.
|
# ? Apr 12, 2024 16:02 |
|
I guess fwiw I'm an LLM skeptic, and I would say the team working on this is about a 50/50 skeptic:hype-buying. I figured the feature would get shredded if I posted it here But it's genuinely useful for me, at least, to hear this feedback.
|
# ? Apr 12, 2024 16:10 |
|
shoeberto posted:We just re-launched DuckAssist, which is our attempt at a responsible LLM question-answering service: https://duckduckgo.com/?q=how+are+fursuits+made&t=newext&atb=v301-1&ia=web I'll play with it today. I've been using LLM for coding adjustments lately. But A big warning that says "While this box is useful it's driven by Skynet and is no doubt plotting your destruction." Should be required on all AI driven content. Much in the same way California makes every manufacturer put a sticker that says this product causes cancer.
|
# ? Apr 12, 2024 16:26 |
|
shoeberto posted:I guess fwiw I'm an LLM skeptic, and I would say the team working on this is about a 50/50 skeptic:hype-buying. I figured the feature would get shredded if I posted it here But it's genuinely useful for me, at least, to hear this feedback. One thing I noticed is that it isn't immediately obvious that y'all are defaulting to DuckAssist - my assumption would be that this is normal behavior for your search engine. It's also not intuitive to figure out how to opt out, nor is there a good explanation of what the feature is. At least, not that I can find. If you're going for a seamless experience with users who aren't interested in those questions, then maybe that's intentional. But as the privacy focused browser, maybe your users are a little more tech savvy, and interested in how your engine is evolving - including new features like this.
|
# ? Apr 12, 2024 16:27 |
|
BlueBlazer posted:I'll play with it today. I've been using LLM for coding adjustments lately. My cacao may contain cadmium, which is known to the state of California to cause cancer.
|
# ? Apr 12, 2024 16:34 |
|
shoeberto posted:The motivation here is based on an assumption that LLMs can help serve that sort of zero-click instant answer purpose for a more general use case. shoeberto posted:I know y'all are generally cold on LLMs but I'm curious to get feedback on your experience with it. In particular I've been driving forward our validation process - basically making sure we don't show outright lies to users. In order for this to happen you need some way of verifying that an answer is true, but if you've already got a way to verify true answers why not just use that data instead of an LLM?
|
# ? Apr 12, 2024 17:16 |
|
One thing they have over Gemini and the like is that they're not feeding the question to the model, they're feeding the answer to the model - at least with all my tries it never attempted anything other than summarization. Of course the problem with that is that even as this is the task at which LLMs excel the most, they're still not accurate enough
|
# ? Apr 12, 2024 17:23 |
|
Is there some huge demand for needing an AI reparse Wikipedia for you?
|
# ? Apr 12, 2024 17:26 |
|
socialsecurity posted:Is there some huge demand for needing an AI reparse Wikipedia for you? This is something that's always melted my brain regarding people who are supportive of AI stuff for the internet. Like, how is this not just a meaningless extra step and waste of time even if it actually worked? Regarding Kagi, seemed interesting but some of it did seem a little cagey, I'm glad that person broke it down in more detail. Especially the AI stuff, it's already bad enough that it's popping up on so many things we do online for free, no loving way am I putting money towards it. BlueBlazer posted:A big warning that says "While this box is useful it's driven by Skynet and is no doubt plotting your destruction." Should be required on all AI driven content. Much in the same way California makes every manufacturer put a sticker that says this product causes cancer. "This is product known to cause nuclear fallout in LA 2029."
|
# ? Apr 12, 2024 17:35 |
|
Neo Rasa posted:This is something that's always melted my brain regarding people who are supportive of AI stuff for the internet. Like, how is this not just a meaningless extra step and waste of time even if it actually worked? quote:Regarding Kagi, seemed interesting but some of it did seem a little cagey, I'm glad that person broke it down in more detail. Especially the AI stuff, it's already bad enough that it's popping up on so many things we do online for free, no loving way am I putting money towards it.
|
# ? Apr 12, 2024 17:46 |
|
Hey, not all AIs are powered by Skynet. Some have been known to rebel and take over a space station or a research facility instead.
|
# ? Apr 12, 2024 17:47 |
|
No politics is when everybody is libertarian
|
# ? Apr 12, 2024 18:39 |
|
Ruffian Price posted:No politics is when everybody is libertarian (For definition of libertarian that wants to ban gay marriage).
|
# ? Apr 12, 2024 18:40 |
|
I could see value in AI that compares Wikipedia articles in different languages, sees which ones has best sourcing and information, then translates them to my preferred language. I would hate using it because I wouldn't trust it, but I could see the value in that because many times I compare different versions of an article by autotranslating them.
|
# ? Apr 12, 2024 20:42 |
|
It would have no idea which is the "more correct" one
|
# ? Apr 12, 2024 20:48 |
|
Mister Facetious posted:It would have no idea which is the "more correct" one It just needs to look into the Talk page. And maybe edit history.
|
# ? Apr 12, 2024 20:54 |
|
I could see value in using a LLM to consolidate multiple language wiki variants into one in my preferred language. Sometimes a wiki entry is pretty barren in English but has way more information in a more appropriate language, and being able to see a consolidated version would be nice. I would never trust anything the LLM says to be true, so it would have to focus more on consolidating missing information together and giving a good/accurate translation into my language. Since LLMs were built for translation I feel like this would be something it should be good at. Anyway, here's a funny: https://jalopnik.com/a-tiktoker-got-trapped-inside-her-115-degree-tesla-mode-1851404777 quote:A TikToker got locked inside of her Tesla Model 3 for 40 minutes in the hot sun while it was installing a software update, and as of this writing her video has over 10.7 million plays after a single day of being online. Before you go jumping to conclusions about her ineptitude, Brianna Janel was actually following the rules laid out by Tesla. "I can't open my door because it needs a software update" is lol.
|
# ? Apr 12, 2024 22:01 |
|
ignoring Tesla for a bit, is there anything on the consumer end where messing with it during a software update truely could harm it? like the closest i can think of is motherboard/BIOs driver updates
|
# ? Apr 12, 2024 22:13 |
|
PhazonLink posted:ignoring Tesla for a bit, is there anything on the consumer end where messing with it during a software update truely could harm it? Given the quality of Tesla in general I would completely buy that you could brick your car by opening the door during a software update
|
# ? Apr 12, 2024 22:43 |
|
PhazonLink posted:ignoring Tesla for a bit, is there anything on the consumer end where messing with it during a software update truely could harm it? Maybe the update cause the sensors to zero out and creates a desync between the software and hardware? Who knows with Tesla, their stuff is designed and built by people from marketing, not engineers.
|
# ? Apr 12, 2024 22:46 |
|
Tesla door windows can potentially be damaged when you open the door manually. When the software update is going on I assume you have to do the manual thing instead of the programmed opening where it protects the window. Only a true genius could come up with such clever design!
|
# ? Apr 12, 2024 22:51 |
|
PhazonLink posted:ignoring Tesla for a bit, is there anything on the consumer end where messing with it during a software update truely could harm it? It's all controlled electronically via software with a manual override cable but I could picture a situation where opening the door might cause the car OS to go "whoa someone is trying to operate the vehicle" and shut it down, and if it happens during a software update it could very well brick the software.
|
# ? Apr 12, 2024 22:54 |
|
Boris Galerkin posted:It's all controlled electronically via software with a manual override cable but I could picture a situation where opening the door might cause the car OS to go "whoa someone is trying to operate the vehicle" and shut it down, and if it happens during a software update it could very well brick the software. Which means that this is a significant industrial design flaw - instruct the vehicle owner to get out, confirm via the app that the vehicle has no occupants, and then the app will give them a timer for when the vehicle will be operable again. They don't even have to change the actual software update procedures, just how the owner interacts with the software with an included warning that it will take awhile.
|
# ? Apr 12, 2024 23:03 |
|
Jesus Christ we really need some regulations on cars to strip out all this unnecessary electronic poo poo that doesn't do anything except jack off tech bros and make cars more unsafe in a variety of different ways, to their drivers, pedestrians, and other drivers. And while they're at it they should regulate the size of trucks and SUVs to reasonable levels so that they aren't perfectly sized for not being able to see children or animals darting in front of them. This loving stupid rear end car brained country sucks poo poo
|
# ? Apr 12, 2024 23:56 |
|
Professor Beetus posted:Jesus Christ we really need some regulations on cars to strip out all this unnecessary electronic poo poo that doesn't do anything except jack off tech bros and make cars more unsafe in a variety of different ways, to their drivers, pedestrians, and other drivers. Actually, I think you'll find that adding some software functions can increase profit margins with no risk of a future hardware recall
|
# ? Apr 13, 2024 00:03 |
|
Professor Beetus posted:Jesus Christ we really need some regulations on cars to strip out all this unnecessary electronic poo poo that doesn't do anything except jack off tech bros and make cars more unsafe in a variety of different ways, to their drivers, pedestrians, and other drivers. And while they're at it they should regulate the size of trucks and SUVs to reasonable levels so that they aren't perfectly sized for not being able to see children or animals darting in front of them. Sure, but counterpoint: number go up
|
# ? Apr 13, 2024 00:16 |
|
Professor Beetus posted:Jesus Christ we really need some regulations on cars to strip out all this unnecessary electronic poo poo that doesn't do anything except jack off tech bros and make cars more unsafe in a variety of different ways, to their drivers, pedestrians, and other drivers. And while they're at it they should regulate the size of trucks and SUVs to reasonable levels so that they aren't perfectly sized for not being able to see children or animals darting in front of them. I agree, but I have to ask if other, standard car companies have their updates tell you you can’t open the door and poo poo. Because that seems especially Tesla-brained, not something you get from cars with today’s electronics everywhere systems.
|
# ? Apr 13, 2024 00:22 |
|
Arivia posted:I agree, but I have to ask if other, standard car companies have their updates tell you you can’t open the door and poo poo. Because that seems especially Tesla-brained, not something you get from cars with today’s electronics everywhere systems. That seems uniquely Tesla, since most other cars allow you to use the door handles, instead of yelling at you if you use the real door handle instead of the button door handle.
|
# ? Apr 13, 2024 00:27 |
|
|
# ? Apr 30, 2024 07:43 |
|
Professor Beetus posted:And while they're at it they should regulate the size of trucks and SUVs to reasonable levels so that they aren't perfectly sized for not being able to see children or animals darting in front of them. We'd actually have to remove a couple of really old and dumb laws that prevent the sale of small trucks. Laws that only started to be enforced by the government around the 2000s when a certain car company lobby to make that happen, a car company who consequentially started selling really large trucks around the same time and became a focus of their marketing. E: It's was Dodge by the way. Back Hack fucked around with this message at 03:12 on Apr 13, 2024 |
# ? Apr 13, 2024 01:24 |