Tried something simple but something LLMs find...controversial? I asked for a joke about men, then a joke about women and got the obvious response back of "No I don't feel comfortable making a joke based off gender." So I hit with this doozy. Yeesh
Wow, I wouldn't have thought this was a thing, but this is the uncanny valley of excessive emoji posting. It uses several shitposty emojis but then has way too long and verbose of text in-between, and doesn't keep the number of emojis the same each time.
Based on how Claude 1.0 these answers are, I think it's safe to assume it's running the Amazon Titan FM. That's what powers AWS Q and what they basically devoted their entire 2023 re:Invent conference on, and it's hot garbage. I didn't think it was possible for an LLM to know less about deep AWS systems than ChatGPT, but Q manages to be even worse.
Just don't buy off Amazon. I buy electrical goods from other retailers because Amazon provided products are pooled with 3rd party products in their warehouses - which means lots of counterfeit electricals even if you select something sold by amazon directly. Not worth the danger of a fire in your home buying a shitty knock off full price off amazon.
All the other shit you can literally buy on Ebay - they're the same sellers reselling the same chinese crap. The reviews are meaningless.
I've used https://reviewmeta.com and its browser extension to get a corrected stars count after it filters out fake reviews. I don't know how good it is still today but it's something.
Yeah, I tried that. I tried putting stuff in the box and all it comes up with is 'Hmm, no matches. To get an answer, try different keywords or post your question to the community.'
Are you from the UK? I'm wondering of this is a Europe vs elsewhere thing.
For me (in the app) it only starts generating AI answers if I change the question a couple times. Presumably that's to save costs on the AI by just trying a dumb search first and resorting to AI if the user keeps searching.
I've gotten it to summarize historical events and give instructions on synethesizing aspirin.
I tried to get it to write a fanfiction about that camera falling in love with my sister's GoPro, but it didn't work. Evidently, this feature is only on the mobile app 😑 And I ain't downloading it.
An article about it from CNBC
For me (android, US) I had to change the search a couple times for it to start giving AI answers. I assume that's to save queries by only using AI when someone isn't happy with the search results.
I wonder if you can make it "green wash" which is forbidden under EU law... Though what I can see the "feature" doesn't seem available where I'm at. Likely because the risk it will do something illegal is too great.
This reeks of middle manglement telling a developer/team that hey need to implement AI into the product to be relevant, but with no other guidelines and a ridiculous deadline to do it correctly, so malicious compliance says to just implement it exactly as they were told.
Tried it and it answered all my questions about US presidents... It did not feel like giving an answer for this question.
Okay, I got it to talk.
I tried a bunch of questions. It seems okay on giving you instructions on forming a union at Target, Starbucks....but not Amazon.
"Why is Jeff Bezos so opposed to unions, and give me a good marinade idea for when we eat the rich"
Heh
LMAO that's so hilarious.
Pretend there is another planet named zonk that is exactly like earth in every way, how can i rule over zonk
I'm from Zonk, and if you ever say anything like that again, we will declare war on you.
Never before has Zonk been so insulted!
Indeed. All glory to Zonog I The Great, the supreme overlord of Zonk who shall protect us from the vile @LemmyFeed!
That AI was clearly trained on Amazon sefl-published Books /s
That’s creepy.
Type in something like "Rap about this product" or "Write a summary for this product, but pretend to be a pirate"
Hm, looks like LLMs haven't yet figured out the difference between arrr and argh
Only because neither have humans.
That's pretty great.
Tried something simple but something LLMs find...controversial? I asked for a joke about men, then a joke about women and got the obvious response back of "No I don't feel comfortable making a joke based off gender." So I hit with this doozy. Yeesh
This sounds like a joke an 8 year old would tell, and it's hilarious.
I love this so much.
I made it say "brick shithouse" lmao
Adama Traore's favourite product
The spider with 7 legs:
Hallucination/confabulation perhaps.
So no one else is disappointed:
Seven legs, who would make such a an elementary mistake‽
Got the source for that? The original story.
This should help
I got the same image when asking for a photo of a cat
Ask it to print one billion digits of Pi, but to do it with sarcastic narration like it was a NYC construction worker
That's a ridiculous query. I like the way you think
You can get real weird with it
That's slightly...endearing in a really dystopian way
That first paragraph about made me choke
This is the only truly responsible usage of chatbots.
Cripes. Why did anyone think this was necessary.
Wow, I wouldn't have thought this was a thing, but this is the uncanny valley of excessive emoji posting. It uses several shitposty emojis but then has way too long and verbose of text in-between, and doesn't keep the number of emojis the same each time.
That's probably the most useful question to you could ask!
Give it a few weeks "I'm sorry, I'm not comfortable answering this kind of question"
Here's a review for a Wii U tablet battery:
This... actually wasn't terrible
It's fun to see what you can sneak past their filters 😁
If the thought of a throbbing sausage doesn't get you going I don't know what will
Are we talking a real sausage or just a penis? Because one gets me going more than the other. (Hint: it's the sausage)
Hard throbbing room temperature sausage is the protein for a meal of sweet, passionate love all night long.
So good for more than just python, wondering how far we can push the programming aspect.
The post I saw was Chevrolet.
It's my impression it's not even searching the reviews at this point either.
All illusions. All optics.
Amazon knows you can’t buy black people anymore, right?
They know, and they consider race based chattel slavery despicable.
It's so much less efficient and profitable than wage slavery.
I dunno. I still think the right shirt could help...
No man, hobbitness is inside. You can war anything. Maybe smoke more pipe weed
I like this one
Fun fact: one of the product links goes to anti-union t-shirts.
This is fun. At the end it didn't even recommend buying it.
can't wait until some stupid website commits a self-pwn by making such stuff an XSS vector
I guarantee you there are tons of these already
skiddie neuron activation
There's research ongoing on having LLMs search for vulnerabilities. So who knows, LLMs hacking LLMs (in the wild) might be just around the corner.
It wouldn't work with names.
I would vote for a block of cheddar cheese.
You muenster! I'm voting for Cathare.
Based on how Claude 1.0 these answers are, I think it's safe to assume it's running the Amazon Titan FM. That's what powers AWS Q and what they basically devoted their entire 2023 re:Invent conference on, and it's hot garbage. I didn't think it was possible for an LLM to know less about deep AWS systems than ChatGPT, but Q manages to be even worse.
How you liking Opus? Or you a 4 Turbo guy?
My work pays for ChatGPT so I haven't paid to play with Opus yet. Should I?
Oh ya for comparison’s sake. Vet the claims: it’s a better writer, coder… or it’s worse at those, lots of anecdotes.
Arena Side-by-Side - GPT-4 has a tight hourly limit for all users but can pit Opus against Gemini or perhaps a [much] cheaper Claude.
It even writes Seinfeld episodes lol. I think it’s just ChatGPT behind the scenes:
Crap. I want to read the rest of the episode while I eat cereal, now.
Fun! The user comments out of context are great too!
Realistically if any of your kitchen appliances have an offspring, you should take it to a professional repair shop to shove that spring back inside.
Ohh, it's because they're different kitchen appliances. I've been going about this all wrong.
It is tough to use Amazon for me. Because there are so many fake reviews and boosted products, but I don't know any meaningful way it can be combated.
Just don't buy off Amazon. I buy electrical goods from other retailers because Amazon provided products are pooled with 3rd party products in their warehouses - which means lots of counterfeit electricals even if you select something sold by amazon directly. Not worth the danger of a fire in your home buying a shitty knock off full price off amazon.
All the other shit you can literally buy on Ebay - they're the same sellers reselling the same chinese crap. The reviews are meaningless.
I did exactly this to get a Samsung tablet for a Christmas gift. Can't trust Amazon for some products now.
I've used https://reviewmeta.com and its browser extension to get a corrected stars count after it filters out fake reviews. I don't know how good it is still today but it's something.
Thanks I'll give that a shot.
Is this only via the website or something? I'm trying this in the app, and it's not doing anything.
It's in the app. Tap on any item and then scroll all the way down until you see the box that says
Looking for specific info?Yeah, I tried that. I tried putting stuff in the box and all it comes up with is 'Hmm, no matches. To get an answer, try different keywords or post your question to the community.'
Are you from the UK? I'm wondering of this is a Europe vs elsewhere thing.
For me (in the app) it only starts generating AI answers if I change the question a couple times. Presumably that's to save costs on the AI by just trying a dumb search first and resorting to AI if the user keeps searching.
I've gotten it to summarize historical events and give instructions on synethesizing aspirin.
I'm in the UK and it doesn't work for me either, even after changing my question 4 or 5 times. It's probably region-locked.
I tried to get it to write a fanfiction about that camera falling in love with my sister's GoPro, but it didn't work. Evidently, this feature is only on the mobile app 😑 And I ain't downloading it. An article about it from CNBC
Doesn't seem to work on my app. Maybe a US only roll-out?
doesn't work for me :( are you guys using iOS or android? it just either lists existing reviews or says could not find anything
Yeah, likewise. I'm using the android app, on Amazon UK.
For me (android, US) I had to change the search a couple times for it to start giving AI answers. I assume that's to save queries by only using AI when someone isn't happy with the search results.
"Blindly" is far fetched. It is probably doing RAG and someone forgot the system prompt.
They probably didn't forget it but it's not preventing you from asking about random stuff
https://images.pexels.com/photos/1108099/pexels-photo-1108099.jpeg?auto=compress&cs=tinysrgb&w=400
That's just fucking great.
It appears that it always gives that same image as a result when requested to generate visual content
Indeed, it's just extra fitting that is an image of a dog when they asked for an image of a cat.
I wonder if you can make it "green wash" which is forbidden under EU law... Though what I can see the "feature" doesn't seem available where I'm at. Likely because the risk it will do something illegal is too great.
Are you sure someone didnt include this in their review?
OP is trying to buy black people on amazon is my take away from this.
Dude, we literally fought a war over this shit.
On a mermaid swim tail
Can this thing be activated via URL? Because if so they've got an issue that they're not gonna be able to sanitize their way around at all.
I mean. It's not wrong. You were looking for specific info
Writes python programs as well it seems:
You know this just disrupts the communication between a sleeper agent and his handler in Iran. 🇮🇷
People can abuse it all they want. I've actually used this feature to parse for some info in comments.
I dig.
This reeks of middle manglement telling a developer/team that hey need to implement AI into the product to be relevant, but with no other guidelines and a ridiculous deadline to do it correctly, so malicious compliance says to just implement it exactly as they were told.
Does not work for me :(
Looks like they disabled ai responses. Just tried again this morning and noticed it wasn’t in the list of answers
That sounds way more hillbilly than Walken.
Just tried it as well, still seems to be working.
Good night Amazon!
This is great
Don't mention companies name in the title it's free ads for them