Post B0iDsG6APExZPoPXCS by parker@dsmc.space
(DIR) More posts by parker@dsmc.space
(DIR) Post #B0i9uY4KAFi3uaNxJo by 7666@comp.lain.la
2025-11-28T17:31:02.117824Z
1 likes, 1 repeats
i dread the day AI companies figure out that they are sitting on a gold mine of advertising and product placement injection. all they gotta do is respond to prompts with product recommendations for all their free models and make people pay not to get sold a product."hey <ai product> how do I get this stain off my counter""That sounds tough! Have you tried OxyClean? OxyClean is the toughest stain remover! Use OxyClean!"
(DIR) Post #B0iA4N4jtMtmiEGO12 by meso@new.asbestos.cafe
2025-11-28T17:33:09.750355Z
0 likes, 0 repeats
@7666 they're basically on "dicksuck any solutions that make corpos money and are corporate-safe" mode right now. you ask it "should i do this" and it's like "It's not recommended to do this!!! you should buy <whatever>", it sounds like it treats shilling products like it's a politically correct thing yknow? like that's the other PCness of AI.
(DIR) Post #B0iADmpLW7aXBbYL6O by PositiveC@cawfee.club
2025-11-28T17:34:49.405467Z
0 likes, 0 repeats
@7666 kek, that'd be horrible, but there's already something somewhat as sinister as that. There's a company that injects product placements in existing media using AI
(DIR) Post #B0iAFpS0jXNpu0bzrU by 7666@comp.lain.la
2025-11-28T17:34:57.819083Z
1 likes, 0 repeats
@meso I don't know of AI explicitly recommending products, or AI companies taking money to steer their model responses towards brands. That's the breakthrough I expect they'll figure out.
(DIR) Post #B0iATloD4SVwf3dvPs by mint@ryona.agency
2025-11-28T17:37:43.509092Z
1 likes, 1 repeats
@7666 The SEO niche for poisoning LLM datasets with advertisements probably already exists. Generate enough sites with articles praising your product and there's a non-zero chance it would get recommended to whoever prompts for that product niche.
(DIR) Post #B0iB3sAatkCSoZKP1E by meso@new.asbestos.cafe
2025-11-28T17:44:16.629066Z
0 likes, 0 repeats
@7666 yeah, they don't explicitly steer them towards certain things. but the most popular thing generally gets free advertising and people are told to use it instead of some less shilly solution
(DIR) Post #B0iBAwG462pPWNFxM8 by 7666@comp.lain.la
2025-11-28T17:45:15.103534Z
2 likes, 0 repeats
@mint I don't think there's specific weighting being done at the LLM layer, maybe just from the dataset itself, but in this case I don't think it has any affect because everyone SEOs the shit out of their products. Just prompting "Can you recommend some soda brands" gave me this (image). Maybe it lists these because it defaults to American brands because that's probably what most of the training was done on (or it's using a language switch, or it's just using my estimated location).Can you imagine if they fucked with the ordering of the brands for money though? Bolded or highlighted certain brands for money? Excluded brands at the request of other brands for money? That's the crazy shit I'm talking about. If they really wanted to manufacture a reason to do this, they could make a "trusted products" list and make companies essentially pay protection money to be on it to then be referenced in models.
(DIR) Post #B0iBI7HtUHhbkuXCUK by RustyCrab@clubcyberia.co
2025-11-28T17:46:50.329665Z
4 likes, 1 repeats
@mint @7666 :niggakagi: already suffers this issue with their quick answer feature. If you ask it about any product it will use the top results to uncritically summarize what the pages say. That's what it's "supposed" to do so it's not "wrong" but when you ask about a product you are presumably not asking for manufacturer propaganda
(DIR) Post #B0iBhbDbLJ4pp8g59k by phnt@fluffytail.org
2025-11-28T17:51:26.950993Z
2 likes, 1 repeats
@RustyCrab @7666 @mint I've noticed a decline in results quality on :niggakagi: 2-3 months ago and after the addition of continuation prompts to the quick answers, quality went completely downhill.
(DIR) Post #B0iC31nJjEikKTbMCO by RustyCrab@clubcyberia.co
2025-11-28T17:55:19.126846Z
2 likes, 1 repeats
@phnt @7666 @mint that's because the quality of googles results is dropping due to mass spam. Kagi doesn't crawl their own web but rather re-ranks the results of other services. One criticism I do have is that their image results quality HAS dropped and it does not appear related to google. For the first time in several years I sometimes have to go to google to find an image. However curiously their text results will often have the image I want in the first page so there is some disconnect between their image result set and their general web ranking I might make a bug report about that soon because it's pretty repeatable
(DIR) Post #B0iCKsCO8xwPmVpzpQ by phnt@fluffytail.org
2025-11-28T17:58:33.186433Z
2 likes, 1 repeats
@RustyCrab @7666 @mint Which is somewhat weird, because I switch to Brave/DDG at the start of November to try to recalibrate my sense of reality and it hasn't been that bad. But also I didn't search for anything technical which I used to do a lot on Kagi.For image search, I almost always turn to bing or yandex.
(DIR) Post #B0iCa2hlMfZMi1kKEi by RustyCrab@clubcyberia.co
2025-11-28T18:01:17.379093Z
4 likes, 1 repeats
@phnt @7666 @mint DDG is laughably bad to the point that I cannot understand why anyone uses it. Brave is pretty alright but has very bad image search. Bings interface gives me a fucking aneurism I literally want to punch my monitor looking at it
(DIR) Post #B0iD0FT4XCMfp2N2Rc by RustyCrab@clubcyberia.co
2025-11-28T18:06:01.362536Z
3 likes, 1 repeats
@phnt @7666 @mint but honestly I rarely use web search anymore as it is largely unusable everywhere equally. 9/10 the information you want is on page 4 past 50 incomprehensible spam blogs or buried 4 screens down in a random Reddit comment and then you have to read Reddit comments
(DIR) Post #B0iDExve1j79lkkjb6 by phnt@fluffytail.org
2025-11-28T18:08:41.347026Z
2 likes, 1 repeats
@RustyCrab @7666 @mint Yeah, I miss the assistant and translator more than their search :D
(DIR) Post #B0iDM7SF0bPZL8PWPQ by RustyCrab@clubcyberia.co
2025-11-28T18:09:58.793998Z
2 likes, 2 repeats
@phnt @7666 @mint I wish that LLMs were better at citing sources of where they got info from so you can verify. They're very good at relaying information but very bad at pointing you to where they got that information
(DIR) Post #B0iDXaqc4AN1gcHWbo by parker@dsmc.space
2025-11-28T18:12:03.107587Z
3 likes, 1 repeats
@RustyCrab @7666 @phnt @mint Google's Deep Research is pretty good at that. I've asked it medical questions and it gave a bunch of sources.
(DIR) Post #B0iDfnUuuwSFEj2sxU by RustyCrab@clubcyberia.co
2025-11-28T18:13:30.481974Z
5 likes, 1 repeats
@parker @7666 @phnt @mint well medical sources may be an easier case in general because it's going to be looking at peer reviewed journals. Much of the knowledge from programming comes out of random comments some graybeard seer left on GitHub issues 10 years ago
(DIR) Post #B0iDsG6APExZPoPXCS by parker@dsmc.space
2025-11-28T18:15:46.915206Z
4 likes, 1 repeats
@RustyCrab @7666 @phnt @mint Well that's very human of it>how do I do X in C?Like this>Where did you learn that?I dunno
(DIR) Post #B0iDvQJ2uVinwVwaUC by lord_nougat@nicecrew.digital
2025-11-28T18:15:37.609947Z
0 likes, 0 repeats
They made it up.
(DIR) Post #B0iDvRq3DHiUhAOSZM by 7666@comp.lain.la
2025-11-28T18:16:04.984772Z
3 likes, 1 repeats
@lord_nougat @RustyCrab @phnt @mint
(DIR) Post #B0iE6DAg5PXIOIAqp6 by skylar@misskey.yandere.love
2025-11-28T18:18:18.173404Z
3 likes, 1 repeats
@7666 they're gonna get filled with spam, and eventually the only way to get information will be posting online, then making a sock puppet account to respond with something totally wrong.people won't just reply to a question, but they'll show up to argue and call the sock puppet guy a retard.
(DIR) Post #B0iE6tCZok0ikLNxc8 by RustyCrab@clubcyberia.co
2025-11-28T18:18:25.626507Z
3 likes, 1 repeats
@parker @7666 @phnt @mint unironically LLMs do actually behave a lot like humans in practice, and will often make the same mistakes which they will then catch if you ask them to proofread their own work. Sometimes Claude has gotten snarky with me if I ask it to do a task in the wrong mode and do malicious compliance like leaving a new file in my repo with "hey dumbass you asked me to do auditing work in code mode but here's your answer anyway"
(DIR) Post #B0iElXp0DI5UJGkEN6 by RustyCrab@clubcyberia.co
2025-11-28T18:25:45.044832Z
2 likes, 1 repeats
@Red_Hat @7666 @parker @phnt @mint yeah obviously. They just make similar logic mistakes that humans would miss, which clearly is not intentional but I found it interesting that they run up against the same cognitive challenges that an otherwise smart human would
(DIR) Post #B0iEwrDiQjNTeuhMiO by RustyCrab@clubcyberia.co
2025-11-28T18:27:49.375227Z
1 likes, 1 repeats
@Red_Hat @7666 @mint @parker @phnt and I don't think the explanation is that they are "trained on mistakes" because they then catch and correct those mistakes upon audit. Clearly they "know" that's not how it's supposed to be
(DIR) Post #B0iF1vtOUDI1b6c5Gy by mischievoustomato@tsundere.love
2025-11-28T18:28:43.241200Z
0 likes, 0 repeats
@7666 it's gonna happen soon. nothingburger really
(DIR) Post #B0iFBUNJeMyr9LAL6O by mischievoustomato@tsundere.love
2025-11-28T18:30:27.439147Z
2 likes, 0 repeats
@RustyCrab @7666 @phnt @mint I remember when people shilled DDG, I tried it out, sucked ass, back to google 30 minutes later.
(DIR) Post #B0iKMKKbpeIGxKt3aK by 7666@comp.lain.la
2025-11-28T19:28:04.236570Z
0 likes, 0 repeats
@flisk Well it hasn't happened yet so they haven't found the right combination of justification and revenue forecasts to make it happen
(DIR) Post #B0iKh5bnlXl7NLpSYy by scathach@stereophonic.space
2025-11-28T19:32:13.159381Z
0 likes, 0 repeats
@7666 I'm just surprised they aren't doing this already