|
haha yeah funny observation what's your home address, i have a gift i want to send you
|
#
?
Dec 8, 2023 22:28
|
|
|
|
| # ? Jun 8, 2024 00:14 |
|
|
GhostofJohnMuir posted:b)it will definitely turn up unknown bastard children and leave big questions about a generation that's all dead and can't answer them Same for my coworker. They found out her 84 year old father never knew his real father and they've just kept it hidden for fear of breaking his heart. He has a secret half brother that they avoid even though he is really gung-ho to meet his half brother.
|
|
#
?
Dec 8, 2023 23:07
|
|
|
Antifa Spacemarine posted:Brave is the worst browser I've ever used, irrespective of the crypto poo poo. It's the only browser where regularly it seizes up and crashes. Avoid that poo poo like the plague. Can’t speak to android or desktop but the iOS version has been fine for me. Does have ads for the aforementioned crypto bullshit on the new tab tab but the inbuilt adblocker is worth it.
|
|
#
?
Dec 8, 2023 23:18
|
|
|
That’s because the iOS version is just Safari with a new skin. I do have Brave installed on my phone because it lets you play YouTube videos in background with the screen off.
|
|
#
?
Dec 8, 2023 23:25
|
|
|
Crain posted:EEehhhhh, I think they are in some of their branding for it. Watching that video I assumed that it was just a bunch of separate videos stitched together to create the narrative they showed, with all of the processing time and incorrect answers removed and all. Well it turns out I gave Google too much credit. https://www.bbc.com/news/technology-67650807 quote:In the video's description, Google said all was not as it seemed - it had sped up responses for the sake of the demo. The video being sped up and trimmed to just showing the good stuff is understandable. But lmao the entire video was fake altogether. It sounds like they recorded a bunch of video with a disembodied hand playing with objects on the table, and then they just took still images from the video and prompted their ai with those jpegs and plain text and recorded the answers. And then finally, they hired a voice actor to narrate and then just used a basic text to speech tool to do the ai portion of the conversation. The demo where the guy squeezes the duck and it makes a sound and the AI goes “oh it squeaked, yes it’s a rubber duck then”? quote:
Then there’s also the fact that in Google’s own comparison against openai a lot of the metrics were like just barely “better”. How are they so bad at this? E: In the video they showed a segment where the narrator “shows” the ai a map of the world and asks it to invent a game, and it invents a guess the country game. Completely fake as well. quote:Instead, the AI was given the following instructions: "Let's play a game. Think of a country and give me a clue. The clue must be specific enough that there is only one correct country. I will try pointing at the country on a map," the prompt read. They’re so bad at this. Boris Galerkin fucked around with this message at 00:16 on Dec 9, 2023 |
|
#
?
Dec 9, 2023 00:12
|
|
|
Boris Galerkin posted:It sounds like they recorded a bunch of video with a disembodied hand playing with objects on the table, and then they just took still images from the video and prompted their ai with those jpegs and plain text and recorded the answers. And then finally, they hired a voice actor to narrate and then just used a basic text to speech tool to do the ai portion of the conversation. More likely they had concept stills first, ran those through the AI to get responses, then wrote the script on which they made a video that matched all that. It wouldn't make sense any other way from a production point of view. The idea that there could be a truly intelligent AI that can recognize from a video feed with slight clues what you want in real time seems extremely far off based on this.
|
|
#
?
Dec 9, 2023 00:29
|
|
|
Nenonen posted:More likely they had concept stills first, ran those through the AI to get responses, then wrote the script on which they made a video that matched all that. It wouldn't make sense any other way from a production point of view. That’s what I thought too, but the bbc article explicitly says Google clarified they recorded the videos first, and took the screenshots from it. quote:Google clarified that the demo was created by capturing footage from the video, in order to "test Gemini's capabilities on a wide range of challenges". So, it was prolly: 1. Write a narrative screenplay 2. Record all the videos, but nobody needs to talk because it’ll be added later 3. Use screenshots from video to ask Google a few hundred/thousands of way to get cherry picked responses that are good. 4. Voice actor to record the human portion, text to speech to record the ai portion (or hell, maybe the human portion is synthesized too) And the thing is, the human wasn’t even reading the prompts that were fed to the Google ai as is clearly shown by their fake map game example. So who even knows if the responses presented as coming from the ai are even actual responses or if it was written by a script writer. Boris Galerkin fucked around with this message at 00:39 on Dec 9, 2023 |
|
#
?
Dec 9, 2023 00:35
|
|
|
So it's just another project natal then.
|
|
#
?
Dec 9, 2023 05:46
|
|
|
Tech Nightmares- The Dark Secret Behind Open AI's Serious Issue When you log into your Open AI account (which is a several week wait due to demand), your account page URL works for 7 days. Meaning ANYONE has read AND edit access to your account. So if you have a web crawler or automatic program, you can generate lots of urls till you find one that works, and steal all the customer info, delete the account, change the details, whatever, go wild. And the token expires after 7 days for "security reasons". Great security there chief. Open AI was advised of this flaw, and said it's working as intended.
|
|
#
?
Dec 11, 2023 08:56
|
|
|
I am once again asking anyone who complains about Google's results being trash "lately" to run the same search in an Incognito session and report back on any differences in results. I am begging for this to happen. I want to know, please tell me what happens, because my results are still useful and there's obviously something to be learned here. e. At the very least tell me a search term you've used where the results are clogged with trash so I can run the comparison myself.
|
|
#
?
Dec 11, 2023 09:49
|
|
|
mllaneza posted:I am once again asking anyone who complains about Google's results being trash "lately" to run the same search in an Incognito session and report back on any differences in results. If Duck Duck Go a ... go
|
|
#
?
Dec 11, 2023 10:01
|
|
|
there was that thread on HN lately from a goog search employee saying that they never actually changed how quotes work, they still search for the phrase verbatim (punctuation ignored) and if you get all SEO crap on the first page it's because they put it in there which I can believe, since in the Olden Days SEO pages would just be loads of common search term variants dumped into a blogpost, makes sense that they would now do the CV white text trick and have one version of the page for your eyes and another for the search. doesn't make the results better but they're excused I guess
|
|
#
?
Dec 11, 2023 11:28
|
|
|
Comstar posted:Tech Nightmares- The Dark Secret Behind Open AI's Serious Issue How about an actual article? Google turns up nothing and I’m not gonna sit here and watch a dumb video.
|
|
#
?
Dec 11, 2023 12:10
|
|
|
Ruffian Price posted:one version of the page for your eyes and another for the search
|
|
#
?
Dec 11, 2023 14:22
|
|
|
mllaneza posted:I am once again asking anyone who complains about Google's results being trash "lately" to run the same search in an Incognito session and report back on any differences in results. Biased here for aforementioned reasons but I find their results page too cluttered. A lot of time their auto-answers don't actually give me the answer I want (similar to GPT), and keep me from seeing the URL results. I feel like I spend more time sifting through the SERP than I should when I have to fall back to them.
|
|
#
?
Dec 11, 2023 15:05
|
|
|
Comstar posted:Tech Nightmares- The Dark Secret Behind Open AI's Serious Issue Duh that's why they call it OpenAI
|
|
#
?
Dec 11, 2023 15:06
|
|
|
Boris Galerkin posted:How about an actual article? Google turns up nothing and I’m not gonna sit here and watch a dumb video. It's a common web vulnerability, they don't invalidate or bind dynamic urls and the expiration period is very long. How "normal" sites work: you log in. In the background you take the servers public key and initiate a public private key exchange. You use your password hash (a pre shared secret) to sign a challenge from the server. You staple this to a temporary public key, send it through a SSL bridge back to the server through the servers public key, and the server validates this by finishing the SSL session. You get a url and a locally generated session cookie. This cookie just reminds your computer of its private key, it's expiration period, and what url to send to the server to tell it what temporary public key it should look back up. You can comfortably establish the shared secret through SSL, you get your email validated by a one time use token url. This is that 15 minute code. Chat GPT is alleged to just use that token url with no secondary validation that the person poking it knows the shared secret, the password. And keep it valid to connect for 7 days. This is bad because URLs are technically guessable. This is a critical security vulnerability if validated, and a violation of their PCI agreement (failure to secure data at rest). They could be in actual legal jeopardy over this. Depending on how dumb their web indexing is it could be really bad.
|
|
#
?
Dec 11, 2023 15:12
|
|
|
That's impressively terrible. At my last job we've built a web app that also involves user management and authentication, and despite having only all of 3 developers with not too much experience in that regard we apparently managed something a whole lot more secure than that, including email opt-in as well as revokable session tokens. And not because we're particularly clever or anything, but rather because that's extremely well-trodden ground with a whole host of established best practices and available packages. And yet here is OpenAI with practically infinite budget and they're loving up something this basic 
|
|
#
?
Dec 11, 2023 15:35
|
|
|
TACD posted:this part should not be possible Google's bot has a unique user-agent string if I remember correctly? So it's a theoretical attack vector, at least.
|
|
#
?
Dec 11, 2023 16:44
|
|
|
TACD posted:this part should not be possible Changing what gets sent to the user based on the request headers happens all the time, especially with malware. Gootloader, for example, will only show the fake forum page with the malware download if the headers on the incoming click match a specific set of conditions. That said, you can also do stuff like have a download page check the User-Agent header to decide what version of an application to default to.
|
|
#
?
Dec 11, 2023 17:00
|
|
|
I don't think they're even changing anything, it's probably stashed in css that won't render on a normal browser but can be seen in the downloaded source, that's why I compared it to CV stuffing, the process parsing the 1px white text with additional keywords you added is getting the same pdf a human recruiter would
|
|
#
?
Dec 11, 2023 17:14
|
|
|
Boris Galerkin posted:How about an actual article? Google turns up nothing and I’m not gonna sit here and watch a dumb video. Every time someone makes a video about something that would be better explained via an article you can thank Facebook for outright lying about video vs article views years ago and changing news for the worse.
|
|
#
?
Dec 11, 2023 17:34
|
|
|
https://www.bloomberg.com/news/articles/2023-12-07/ai-fast-food-drive-thrus-need-human-workers-70-of-timequote:Checkers and Carl’s Jr. are among US fast-food chains hailing AI-powered drive-thrus as labor-zapping wizards that speed up service. But a popular provider of these systems recently revealed a crucial part of how it gets so many orders right: humans. "AI" company actually a stealth off-shoring company.
|
|
#
?
Dec 11, 2023 23:41
|
|
|
Neito posted:Google's bot has a unique user-agent string if I remember correctly? So it's a theoretical attack vector, at least.
|
|
#
?
Dec 11, 2023 23:45
|
|
|
golden bubble posted:https://www.bloomberg.com/news/articles/2023-12-07/ai-fast-food-drive-thrus-need-human-workers-70-of-time Lmao they're loving outsourcing drive thrus to avoid paying a livable, what a loving world we live in
|
|
#
?
Dec 12, 2023 02:00
|
|
|
A lot of AI involves substantial outsourcing. OpenAI's operations employs a ton of Kenyans in vetting training materials for ChatGPT viewing all kinds of nasty poo poo to keep it out of the data set.
|
|
#
?
Dec 12, 2023 02:23
|
|
|
Kwyndig posted:A lot of AI involves substantial outsourcing. OpenAI's operations employs a ton of Kenyans in vetting training materials for ChatGPT viewing all kinds of nasty poo poo to keep it out of the data set. There are some bad poo poo stories about Youtube content moderators in Morocco.
|
|
#
?
Dec 12, 2023 12:34
|
|
|
Is there a conspiracy theory about big companies pushing disturbing content to third-worlders in order to zombify them or just destroy their moral standards yet
|
|
#
?
Dec 12, 2023 13:47
|
|
|
No it’s just cheaper and the countries in particular have populations that speak English. That’s all really.
|
|
#
?
Dec 12, 2023 14:17
|
|
|
busalover posted:There are some bad poo poo stories about Youtube content moderators in Morocco. I assume that is for the French and Arabic speaking moderation. My understanding is that the Philippines is where a lot of the English traumatic content moderation occurs.
|
|
#
?
Dec 12, 2023 14:42
|
|
|
Boris Galerkin posted:No it’s just cheaper and the countries in particular have populations that speak English. That’s all really. It's this. Most work in AI and machine learning (and moderation, now that I think of it) involves labeling data. Enormous amounts of data. It's tedious, poo poo work that powers basically every algorithm that does anything meaningful. You need an army to do it at scale. My understanding is that it's peanuts for the firms paying for the work, but it's meaningful money for folks in these countries. It's also genuinely an opportunity to work in the tech sector in places where there may not be many opportunities otherwise. But there's also a lot of poo poo that is deeply traumatic for a lot of people doing the work, and my understanding is that they're not well supported when they experience really awful things. It's ethically very murky and I don't personally know of a good solution that doesn't fall into extremes of "stop using labeled data" or "eradicate traumatizing content from the internet and also the people who create it"
|
|
#
?
Dec 12, 2023 15:07
|
|
|
The bigger companies typically provide some mental health resources or limit hours spent reviewing toxic content, although it’s questionable whether it has worked. Startups like OpenAI, I’m pretty sure they’re just going for the cheapest vendor.
|
|
#
?
Dec 12, 2023 15:12
|
|
|
https://restofworld.org/2023/kenya-content-moderators-battle-meta
|
|
#
?
Dec 12, 2023 16:08
|
|
|
Electric Wrigglies posted:I assume that is for the French and Arabic speaking moderation. My understanding is that the Philippines is where a lot of the English traumatic content moderation occurs. India too and South Africa's becoming more popular from the sound of things.
|
|
#
?
Dec 12, 2023 18:44
|
|
|
mllaneza posted:I am once again asking anyone who complains about Google's results being trash "lately" to run the same search in an Incognito session and report back on any differences in results. When I use DuckDuckGo, the page isn't clotted with SEO spam. Do a search for "chinotto fruit" on Google. Then do the same search on DuckDuckGo. Notice that the top of the Google results page is a synthesized textbox, plus a lot of drop-downs for commonly-asked questions. Google has decided what I want to see, and it's not my search results.   Now try just "chinotto". Again, note the real estate taken up by infoboxes.
|
|
#
?
Dec 12, 2023 19:28
|
|
|
Arsenic Lupin posted:When I use DuckDuckGo, the page isn't clotted with SEO spam. your second screenshot has the exact same number of search results as your first and the same result
|
|
#
?
Dec 12, 2023 19:34
|
|
|
(splitting this up to provide page break) Now search for [red dog]. Scroll past the top level of infoboxes, but stay on the first page. Google:  Note the Red Dog Saloon and Red Dog Pet Salon and Spa. DuckDuckGo:  Notes: I've never searched for "red dog" before, and I don't live in either of the states with the saloon and salon.
|
|
#
?
Dec 12, 2023 19:35
|
|
|
Jose Valasquez posted:your second screenshot has the exact same number of search results as your first and the same result
|
|
#
?
Dec 12, 2023 19:38
|
|
|
hot take, in addition to account creation fees, i think micro transaction fees uploading content should also be a thing. mods shouldnt have to risk some of the worst infohazards for modding. if bad actors had to add a digital fin paper trail they wouldnt pollute the upper surface levels of the internet.
|
|
#
?
Dec 12, 2023 20:18
|
|
|
|
| # ? Jun 8, 2024 00:14 |
|
|
PhazonLink posted:hot take, in addition to account creation fees, i think micro transaction fees uploading content should also be a thing. If SA is any indication a $10 fee for posting hosed up poo poo is not going to stop people
|
|
#
?
Dec 12, 2023 20:26
|
|