“Ready or not, here I come.” That was from a children’s game.
Now, it’s the reality of AI. So ready or not, here it comes. With AI, we have a rocky road ahead.
If I were given the choice, I would probably stop the AI train because I think it can do more harm than good. Feel free to tell me that I’m wrong.
NEWS: Robocall (AI deepfake) of Joe Biden tells NH Democrats *not* to vote on Tuesday.
We’ve been working with US officials on the expected surge in deepfakes.
This is just the beginning. https://t.co/LHg71ctqUE
— Miles Taylor (@MilesTaylorUSA) January 22, 2024
The NYT gift link below takes you to a test of 10 images that you determine are either real or AI. I wonder how well people can do with voices.
The most interesting thing in the article, to me, is that results showed that higher confidence in the person’s answer correlated with a higher chance of being wrong. Yikes!
Anyway, take the quiz, or just read the article at the link.
(New York Times gift link) h/t dnfree
Distinguishing between a real versus an A.I.-generated face has proved especially confounding.
Research published across multiple studies found that faces of white people created by A.I. systems were perceived as more realistic than genuine photographs of white people, a phenomenon called hyper-realism.
Researchers believe A.I. tools excel at producing hyper-realistic faces because they were trained on tens of thousands of images of real people. Those training datasets contained images of mostly white people, resulting in hyper-realistic white faces. (The over-reliance on images of white people to train A.I. is a known problem in the tech industry.)
The confusion among participants was less apparent among nonwhite faces, researchers found.
Participants were also asked to indicate how sure they were in their selections, and researchers found that higher confidence correlated with a higher chance of being wrong.
“We were very surprised to see the level of over-confidence that was coming through,” said Dr. Amy Dawel, an associate professor at Australian National University, who was an author on two of the studies.
“It points to the thinking styles that make us more vulnerable on the internet and more vulnerable to misinformation,” she added.
A.I. systems had been capable of producing photorealistic faces for years, though there were typically telltale signs that the images were not real. A.I. systems struggled to create ears that looked like mirror images of each other, for example, or eyes that looked in the same direction.
But as the systems have advanced, the tools have become better at creating faces.
The hyper-realistic faces used in the studies tended to be less distinctive, researchers said, and hewed so closely to average proportions that they failed to arouse suspicion among the participants. And when participants looked at real pictures of people, they seemed to fixate on features that drifted from average proportions — such as a misshapen ear or larger-than-average nose — considering them a sign of A.I. involvement.
What do you believe when you can’t believe your own eyes?
Open thread.
Baud
Also, a song lyric.
trnc
You are not wrong, WG.
ETA: One of my fears is how it might undermine real video evidence in a trial and create reasonable doubt.
WaterGirl
@Baud: I am about to send you an email message.
Baud
@WaterGirl:
I hope it’s about extending my car warranty.
Mr. Bemused Senior
@Baud: how do we know you’re not an LLM?
trollhattan
@Baud: “The moon landing was fake but this extended car warranty is very real–trust me.”
Gutter guards=the new extended car warranty.
Leto
I was told, in a previous thread, that anything technology related (Bitcoin/AI/what have you) is a red herring and that our leaders don’t need to spend time worrying about them. I’m going with the advice from the BJ knowledge base, and just letting this run it’s course. I’m sure it’ll be fine if we do nothing about it.
trnc
The flip side is, I can’t believe how bad the AI looks for the planned Kiss virtual tour, announced right after their last live show. Uncanny valley circa 2005.
wjca
@Baud: Or perhaps an awesome investment opportunity from the League of Legitimate Nigerian Businessmen.
SiubhanDuinne
Like any technology, AI itself is neutral but can be put to horrible uses by evil or unscrupulous people. It’s those people who terrify me, not AI per se.
(Also, I wish we had a serif font as the default, because every time I see the acronym for Artificial Intelligence, my first thought is Jolson? Bundy? Hirt? Capp?)
trnc
@Leto: Exactly. In fact, Dean Phillips call to not regulate AI is what makes him so presidential, and is an especially bold move that his biggest campaign donors just happen to agree with. I mean, what are the odds?
wjca
As Baud’s note highlights, we are still, a couple of decades later, struggling with the imact of email advertising. For that matter, snail mail scams have only faded because enail scams are so much cheaper to run. And note that TV scams (think My Pillow) are still going strong.
So it’s a good bet that we’ll take a while, like maybe a generation or two to adapt to AI-based scams. With the added fillip that they won’t be just financial ones.
trnc
“AI doesn’t create deep fakes, people do. OK, yes, with AI, but still …”
NotMax
Newest Samsung phone attaches a watermark to photos processed with its inbuilt AI as “insurance” they can be identified as having been manipulated.
Said watermark, however, is in no way permanent and can be removed the same way a photo can be altered by erasing any element in its composition.
WaterGirl
@Baud: No, I would have called you about that! Those calls work better by phone.
WaterGirl
@Leto: ooh, I haven’t had time to read BJ this morning. Which thread? Any suggestion for a nym to search for?
Azhrie139
Regardless of whether the pseudo-AI technology itself can help or do harm, the bottom line is that it is being developed, funded, and advocated for by the most unethical low lives in the already abysmally low standard tech. industry. For that alone everyone should have visceral negative reaction to its proponents.
WaterGirl
@SiubhanDuinne: I have the same problem. It reads as Al, short for Alan or Allen.
NotMax
@WaterGirl
Go old school for clarity: A.I.
;)
Hob
@Leto: This isn’t a complaint about you in particular, but about a type of comment I find especially pointless & annoying from anyone: the sarcastic “But I’ve been told…” that refers very vaguely to some other person’s comment you didn’t like, and makes it sound like it was somehow an official consensus position (“the BJ knowledge base”). If you want to argue with whoever it was, just argue with them! This kind of thing (with minor variations like “But I’ve been reliably assured that…”) now makes up roughly 25% of all comments at Lawyers Guns & Money, and I hope it doesn’t become equally popular here– I’ve never understood why anyone thinks it’s a fun way to make a point or score points or whatever.
Baud
@Hob:
LGM ruins everything.
Urza
Need AI that is trained to detect what other AIs create. Photo, voice and text. There will be false positives as many humans are definitely not much more than a large meme model regurgitating.
WaterGirl
Did anyone else take the test at the NYT link?
My score for guessing right was 60%.
Brachiator
I only caught one AI image.
I have heard the voice of the AI George Carlin. It didn’t sound quite right. And it lacked a certain something with respect to wit and delivery. It reminded me of the scene in The Fly, where the characters note that a teleported steak looks right, but tastes like crap because it lacks everything that makes steak savory.
I don’t fear AI, but I am deeply ambivalent about it and see that, once again, tech enthusiasts insist that we must jump into AI as fast as possible and worry about controls and regulations later.
It also astounds me that some AI enthusiasts ignore, accept or are as pleased by AI failures as they are by AI successes.
cmorenc
@SiubhanDuinne:
Kelly
A.I. will completely undermine the Porn industry. No need to find a website with images and videos of real people. Buy an app to generate your fantasy right there on your PC.
Mousebumples
@NotMax: not sure if you (or others) have techy input, but I’m debating what to get for my new/next phone. (my current Galaxy S20 series phone isn’t holding a charge for long at this point)
Debating S24 Ultra vs Pixel 8 Pro, mostly, but open to other (Android) options.
Apparently the Pixel Pro will sit on hold for you, lol, and answer unknown numbers with Google Assistant. That’s a use of AI that I didn’t know I needed.
SiubhanDuinne
@WaterGirl:
Or Albert, or Alfred, or Algernon….
Urza
@Brachiator: AI lying to them makes it closer to human.
Suburban Mom
Since this is an open thread, does anyone have experience adopting or fostering a geriatric (15+ year old) cat? I have a family member who needs to find an alternative home for a pet, and I’d like to understand what I’m signing up for if I do it. I’ve had cats before, including cats that old, but they were always with me from the time they were kittens.
Kristine
I got 5/10. Looking at the overall scores, not surprised.
I’m wondering if some of the “Real” photos I got wrong had been photoshopped.
bjacques
There’s always the possibility that people are now so firmly ensconced in their worldviews that they’ll believe anything that aligns with it, no matter how sloppily constructed, and will disbelieve what doesn’t, however well sourced–so AI arrived a bit late
EDIT: I got so many of the faces wrong I just gave up. But I don’t care about any of those people.
Chris
@Hob:
But I’ve been told that it is a fun way to make a point or score points!
Urza
@Mousebumples: I like the Fold phones from Google and Samsung. Not cheap, but option for larger screen/mini-tablet if you use it frequently is pretty nice.
Mr. Bemused Senior
@cmorenc: a computer is a device for making errors faster than humanly possible.
WaterGirl
@Brachiator: They are attention whores. Any attention will do.
teezyskeezy
In general, yes, AI disinfo is a scary thing, but on this particular issue, if we get into an arms race of robocall deepfakes of Biden and Trump each telling the other side not to vote, then what shakes out is the most gullible group most beholden to a cult of personality will be the most affected and…I like our chances.
WaterGirl
@Kelly: will A.I. generated kiddie porn still be prosecuteable as kiddie porn if they can’t point to a real child?
teezyskeezy
@Brachiator: I don’t fear AI, I just fear stupid/evil people giving it responsibilities it should not have, or using it for stupid evil purposes.
That said, if a robot dog shows up at my house, I’ll probably fear that. So, I guess maybe I have to admit *some* fear directed at the tech and not just the people who made it.
teezyskeezy
@Kelly: The uncanny valley aspect might even be a selling point to some.
patrick II
I stopped at a McDonald’s this morning. They had a new robot voice taking orders in the drive-thru lane. I ordered a #1 with a coke. The order screen showed a #1 with a coffee. I said no, that’s with a coke. So The screen showed a #1 AND a coke. I said no that is a #1 with a coke. It showed another #1 this time with a coke. I said no I don’t want the second #1, I just want one #1. The screen showed another #1 with Coffee. . By that time I had ordered three #1s for $25. I gave up and circled around again and got behind one other car giving an order. It seemed to take awhile. I hope it went alright for him. Then my turn again and I started over to very carefully say that I wanted a #1 (no pause) with a coke. This time it correctly got my order. I drove off with a new optimism about where this world is headed.
WaterGirl
@Mr. Bemused Senior:
That’s good! Is that an original or just something I hadn’t heard yet?
trollhattan
Know your memes. If Stanley cups are fad-o-the-day, then by gawd, get you some Stanley cups!
Roseville exists primarily to host shopping and chain restaurants, so this is very on brand.
TriassicSands
There may be no failing of human beings greater than their inability to choose wisely between “can we” and “should we.” When you have a society and economy centered on maximizing profits without regard for the well-being of the society or individuals, bad choices will be common.
NaijaGal
@wjca: Hey!
WaterGirl, even though I use machine learning/AI in my work, I share some of your fears and there have already been news articles detailing wrongful arrests of black men based on facial recognition software trained mostly on white faces.
Unrepresentative datasets used to train AI that’s then marketed as one-size-fits-all produce biased results that can cause real harm. This is also a problem for healthcare-related data.
prostratedragon
@SiubhanDuinne: I knew a set of brothers who were Alfred (Jr.), Albert, Alvin, Allen, Alex, and Aladdin. Mother’s name was also an “Al.”
What Have the Romans Ever Done for Us?
Not surprised that the people most confident in being able to spot the A.I. fakes are the ones least likely to be able to spot them…there are a lot of morons out there who are to stupid to see their own intellectual fallibility. As a result they have an inflated view of their own intellectual abilities that NOTHING can dent.
Another Scott
Haven’t clicked the link.
We should remember that even “real” images are so manipulated by software these days that I’m not sure the linked “test” is meaningful even if carefully done. And that’s before one gets to .jpeg artifacts and the like.
E.g. PetaPixel.com – Galaxy S24 Ultra Review for Photographers:
(Earlier versions of the Google Pixel’s “portrait” magic would erase appendages, also too.)
Basically, don’t trust any image / video / audio / etc. you didn’t take yourself, unless you really, really trust the source.
Cheers,
Scott.
WaterGirl
@NaijaGal:
I have been seeing that as a plot in TV shows for at least 3 years, so it has to have been a thing for much longer than that.
Why are they not working on that problem before its in use a= nd hurting real people?
Ken
@WaterGirl: Unclear, see this article. In 2002, the Supreme Court said that if no actual child is involved, the images are protected speech. The technology has changed, and laws have been passed banning the more-realistic images now possible — but they haven’t been tested against the 2002 ruling.
trollhattan
@Mousebumples:
Don’t have a suggestion but since you mention it, my S20 is entering hitherto unexplored regions of reliability. I’ve never had a smart phone last more that two years and this one is in year 4, I think.
Swelling battery, popping the case apart, has been the downfall of every other smartphone, and their “feature” of no battery access is their achilles’ heel.
Mr. Bemused Senior
@WaterGirl: I’m either quoting or paraphrasing. On my phone at work, no time to look it up.
TBone
https://m.youtube.com/watch?v=Ngeb5lSA8eM
Kelly
@Mousebumples: I bought a Pixel Pro 8 in November. It has a neat feature to navigate phone trees. It builds a touch menu so I don’t have to remember did I want option 2 or 5.
WaterGirl
@Ken: Ugh.
marklar
@NotMax: “Said watermark, however, is in no way permanent and can be removed the same way a photo can be altered by erasing any element in its composition.”
I wonder if it would be possible to legislate that a file’s metadata (for readers who don’t know, file metadata is information that describes or relate to a file, such as its name, size, type, date, author, location, and tags) include a non-editable indication that it was generated with AI.
MisterForkbeard
@NotMax: Even better – you can literally ask the ai to remove the watermark for you and it’ll do it.
Mr. Bemused Senior
@Another Scott: the Circus always insisted on original negatives.
Brachiator
@SiubhanDuinne:
There is something about AI that is not neutral. Maybe this says something about the people who are developing AI and how they relate to it.
In one discussion in a tech podcast interviewing AI developers, one host noted that when AI was used to research and answer questions about various subjects, it would consistently but not randomly come up with answers that were just flat wrong, but not nonsensical. This occurs so often that there is already a term for it: hallucinating.
There is currently no way to predict or to anticipate when it how this happens.
This also reminds me of issues involving facial recognition technology. Despite failures correctly identifying non-white or female faces, this technology continues to be deployed without correcting identified problems. People in control don’t care or have decided that it’s good enough, so it’s just tough luck if people are harmed by erroneous applications of the technology.
Another Scott
@Mousebumples: I was debating on either one of those as well, and just ordered the S24 Ultra over the weekend. (I’ve got an S20+ 5G now.) I’m on Google Fi and they have a special where the Ultra is $650 off and substantially cheaper than the P8Pro.
Plus, I’ve read some horror stories about the P8Pro’s cellular radio having reception troubles in some situations and eating the battery then. My previous phone (LG V35) had issues like that and my J often was furious at me when she wasn’t able to reach me when she wanted, so it was too risky for me to be back in that situation. ;-) (It’s still hard to beat Qualcomm’s cellular radios.)
So, the choice was easy for me.
I expect I’ll still be somewhat annoyed by Samsung’s default camera tuning (way over sharpened), but in every other respect the Ultra seems to be a big upgrade for me.
It’s supposed to ship at the end of the month or early February.
HTH a little.
Cheers,
Scott.
Wave Function Collapse
I guess I’m just a curmudgeon, but a lot of the discussion about AI lately gets an eye-roll from me.
There is real stuff happening in AI, has been for a while, but a lot of the current buzz is because it is the current darling of the venture capital pump and dump crowd. There is a new trend almost every year, in reverse chronological order:
AI will take over the world very soon
Web3 is the future of technology
The Metaverse is coming soon
Cryptocurrencies will replace all normal currencies any day now.
All of those things have an element of reality at their core, but each giant hype train was more about bilking investors than anything else.
Back in the early ’90s there was a big slogan posted in our techy workplace that always gave me a laugh.
ARTIFICIAL INTELLIGENCE NEEDS MINDS LIKE YOURS
Hob
@Baud: I mean, I assume it’s a common thing in lots of forums, LGM is just one that I’ve looked at the most (because there are still some commenters I like, & also because I’m a creature of habit). Internet comment culture spreads around all the time, people see a turn of phrase or a joke structure they like (or even just a word that stands out because it had fallen out of common usage, like “feckless”) and they reuse it in some other community and eventually it’s just another thing for cranky people like me to be sick of. I guess the bigger the community in general, the more likely it is to be an incubator for fads like that.
RepubAnon
@Suburban Mom: I took in an older cat whose human could no longer do so. Giving the cat lots of space and places to hide helped – as did bringing in the cat’s old litter box and other familiar items.
TBone
@NotMax: good rule of thumb. Sometimes the old ways are best (common sense isn’t all that common).
Brachiator
@TriassicSands:
Ian Malcolm speech, Jurassic Park.
dnfree
@WaterGirl:
I got 70% on the NYT quiz, and I’m handicapped because I have face blindness (prosopagnosia). My granddaughter who deals with these images tells me that the clue is in the pupils of the eyes or the teeth not looking right.
dnfree
@TBone: My father used to say of common sense, “If it’s sensible it’s not common, and if it’s common it’s not sensible.”
TBone
https://m.youtube.com/watch?v=WjcYVcoRYdg
TBone
@dnfree: Murphy’s Law!
WaterGirl
@dnfree: I work with someone who also has that. So were you just giving random answers? And you got 70%
Tony G
@teezyskeezy: “AI”, as it exists now, is not “intelligent” and is full of bugs. Whether it will be possible to make it reliable enough to be useful — and more dangerous — remains to be seen. In the meantime, garbage software that is full of bugs can still be very dangerous. I plan to steer clear of it unless I’m forced to use it.
Kelly
@WaterGirl: No idea but I’m gonna read Ken’s link
Sure Lurkalot
I’d like AI used to create some Jetson style Rosie to clean up after dinner and wash windows ie make more free time for humans not mimic the creative endeavors that people want to do in said free time. But no, we get AI “making art” and “writing short stories”.
CaseyL
@Suburban Mom: Good on you for giving the old kitty a home!
I’d introduce the new old kitty the same way I’d introduce a new young kitty: For the first few days, keep them in a smaller space (like a spare bedroom, if you have one) so they have a “home base” they can explore at first and feel secure in. Make sure their water/food/litter pan is in there with them.
I would also take some items – linens, pillows, clothing – from their old home and their original human and put those in the smaller space so there are familiar items they can sniff and snuggle into. Also, have any fave toys from their previous home.
You need to give the kitty some space (i.e., don’t hover over them all the time) while also being ready to give strokes, scritches, and verbal reassurance whenever they seem to need it.
Once the kitty is comfortable in their space, leave the door to it open so they can come out and explore the rest of the house. I’m not sure how to shift their food/water/litter pan – maybe others can chime in on that one.
Other than that, just be aware that geriatric kitties tend to develop geriatric health issues. Thyroid, heart, and kidney functions need to be tracked. (My own old boy has a thyroid issue that affects his heart, and he’s on medication for it. So far, knock wood, his kidneys are doing just fine.)
They might also need specialty food, though I’m assuming (hopefully) that if so, you already are aware of this and maybe already have a supply of it.
Best of luck!
VFX Lurker
@WaterGirl – in the past, an actor could have spoofed Joe Biden’s voice. Or, an email from “jobyden dot com” could have spoofed “a message from the President.” An anonymous text message or mailers from Trump supporters could also have spoofed Joe Biden.
AI may make these shenanigans easier, but these deceptions are not new. Mister Rogers sued the KKK in 1990 because they spoofed his show to deliver racist propaganda over the phone.
I listened to Carl Sagan’s 1995 book, The Demon-Haunted World: Science as a Candle in the Dark last week. He argued that we need to arm the public with skeptical, critical thinking and an understanding of science to handle our changing world. Fraudsters will commit fraud, but we can diminish the number of gullible marks.
prostratedragon
@Brachiator:
Favorite word on the subject from Catherine Russell
Hob
@Azhrie139: Yeah, this is basically where I’m at too. “The technology itself is neutral” is the kind of statement that can be literally true but also irrelevant, because we’re never just dealing with an abstraction of the technology, we’re dealing with what people do, and in this case that’s not hypothetical at all– we already have tons of examples of the problems.
If we’re arguing in favor of, say, gun control… it’s not hard to understand that we don’t mean “propelling a projectile via a small explosion is an inherently evil concept.” Most people who favor gun control don’t think literally all guns should be banned everywhere; some do; but either way, they’re responding to a situation right where the things are dangerously out of control and pushback is clearly necessary, because terrible people are pushing to make it worse. There is no good reason to encourage massive private arsenals. And there is no good reason, other than short-term profit, for machine-learning systems to be rushed into production in every possible niche of existence. They are not solving urgent problems. If they’re so massively useful, they can survive some caution and pushback and supervision.
TBone
I’m unable to post verbose, coherent thought today because I had to go to urgent care for Covid test and the staff all had their chins masked and my flu test is negative but my strep test is positive and I have to take amoxicillin for strep on a gut already ruined by ABX and I just got off the fucking steroids and am waiting for the Covid test results so I am letting music speak for me today. Rage doesn’t cover how I feel today. At least hubby has agreed to mask again. Jiminy Cricket!
trollhattan
@TBone:
As a lousy multitasker I wish you good luck on all that mess!
Mousebumples
@Urza: thanks. I’ve had fold phones before (not in a few years, so obviously new technology), but I think I’m more looking phone vs tablet. The screen size on these is sufficient for my needs.
Lol, yeah, I know the feeling. But almost 4 years! (I got this one juuuust before Covid lockdowns) My husband and I used to be on alternating yearly cycles of who needed a new phone, lol.
Thanks for the positive review!
Definitely, thanks! I was sold on the Pixel this weekend, read some of those reviews, and am now waffling…
TBone
“JANE, LET ME OFF THIS CRAZY CONTRAPTION!”
azlib
@Wave Function Collapse: As a retired IT professional, I agree. The AI hype is to attract investment dollars. I know the cow has already left the barn, but I wish we wouldn’t call all these technologies AI. For example LLMs are really very complex and sophisticated search engines trained on a massive amoutn of data.
One of the biggest problems I see with these systems is the massive amounts of data needed for them to work well and only a few corporation or governments have the means to tap into and store all that data. Of course the data harvesters are running into all kinds of tricky copyright and trademark issues as they harvest data off the Internet.
The old sayin in IT of GIGO (garbage in garnbage out) still applies to these systems. The data a neural network is trained on makes a big difference in how the system performs.
UncleEbeneezer
@Kelly: AI porn is a very long way from replicating the real thing. Or so I’ve heard…
TBone
@trollhattan: thank you, I am feeling like if a frog had wings, he wouldn’t bump his ass a-hoppin’.
surfk9
@trollhattan: My S10 is five and a half years old and is starting to fail. Although the battery is still strong
JML
@VFX Lurker: It is easier than ever to get the lie out in front of a large audience really fast…and with institutional distrust higher than ever (whether it’s media, government, teachers, etc) it’s harder for it to get knocked down.
On the other hand, there’s fewer and fewer persuadables every election anyways, IMHO. So most of the time the only people getting suckered are the ones who already believed the lie?
blegh.
TBone
@prostratedragon: ❤️👏
Tony G
@Tony G: For example: The day when I call a corporate “800” number and get an automated voice-recognition system that provides actual useful information instead of being a goddamn waste of time until I can get a human being on the line will be the day that I think that they’re made some progress with this technology. That day has not yet come, and might never come.
rikyrah
anyone_want_chips (@anyonewantchips) posted at 9:51 AM on Sun, Jan 21, 2024:
When Neil Gorsuch refuses to recuse from a SCOTUS case that would directly benefit billionaire Philip Anschutz who’s been showering Gorsuch with dove shooting weekends at his sprawling mountain resort – the Supreme Court ethics code is a fucking joke. https://t.co/rBnRcnLApC
(https://x.com/anyonewantchips/status/1749097436826525960?t=JxcwipaoRCT2ZlC-M0KFPg&s=03)
Hob
@VFX Lurker: I agree that the choice of a prerecorded audio fake of Biden as the first example undermined the point of the post a little, because it’s the least new problem. With video, there’s more of an issue of people not expecting it to be fakeable as well as it can be now.
That doesn’t mean that audio fakes aren’t also a problem that’s getting worse, but I think a better way to look at that problem would be in terms of non-famous people and lower-level scams. That is, if I’m a political operative and I want to fake a Biden robocall, I have plenty of options— lots of people can do an OK Biden impression, there’s a ton of real recordings to be re-edited, etc.— and I’m probably well-funded enough to get it done no matter how it’s done. But a scam like “target 500 random people and try to make each of them think their mom left them an emergency voicemail” isn’t really feasible unless most of those steps can be done without human intervention. And if it can be done without human intervention in real time— like, making a customized Fake Mom or for each of those people that uses a LLM to be semi-interactive— that’s significantly worse.
Another Scott
@azlib: Brad DeLong loves playing with the latest things, in the hope of enhancing his teaching. He’s been trying to see if “A.I.” can make an automated teaching assistant for his classes.
His experiments have been, er, interesting. (Many have been buried under a subscription on his Substack, unfortunately.) E.g. even with all his books and his decades of writings, the LLMs have a really hard time giving a coherent summary of his teachings when asked a specific, relevant question.
He calls the “A.I.” LLMs “autocomplete for everything”. It’s impressive, but it’s not an example of understanding.
Cheers,
Scott.
mrmoshpotato
@patrick II: I have no optimism for your GI tract.
mrmoshpotato
@TriassicSands: You just say that because you picked the name TrassicSands and not Jurassic Park!
VFX Lurker
Good question. In 2010, Canada brought criminal charges against a comic book reader who had comic books on his laptop. No photos, just drawings. (The link points to the Comic Book Legal Defense Fund summary of the case, and it does not contain the artwork that brought criminal charges).
TBone
The “I” in A.I. is a scam.
Another Scott
@Tony G:
But, but, don’t you want your car dealer phone chatbot to solve the Navier-Stokes equation??
Cheers,
Scott.
sab
@Suburban Mom: We adopted my dad’s elderly cat when Dad went into assisted living. We really enjoyed having him around. We took him once to visit Dad and neither he nor Dad were much interested in each other. He developed hyperthyroid so he needed expensive surgery but lived four more years. He went blind, in one eye from cataracts and in the other from toxoplasmosis (note to cat: don’t eat squirrels), but he mostly got around fine. We made him stay inside and he adjusted well to that. We had two dogs and a bunch of other cats at the time.
We never knew how old he was, but we knew him for at least twelve years.
rikyrah
@Suburban Mom:
I have no advice, but, thank you for stepping up and giving this Senior cat a home :)
UncleEbeneezer
I couldn’t agree more, but man, talk about a big political lift considering how many Americans (Conservatives mostly, but #NotOnly/NotAll) are very much opposed to the prospect of children learning critical thinking skills. It’s not just the “race” but also the “critical” part of Critical Race Theory, that they are so afraid of. Critical thinking unavoidably leads to people seeing through the bullshit of Conservatism and the white-washed history it pushes. It’s one of the reasons my conservative-leaning Dad didn’t want me going to a strong liberal arts college. And it’s really at the center of the whole fear of “liberal college campuses.” Colleges aren’t just the places where we start to meet, befriend, love, understand people from other demographics, it’s also where many of us first start to question all the bullshit that our families, churches, and communities led us to believe through critical thinking. Conservatives HATE that and will fight it at every turn.
Kelly
@UncleEbeneezer: A lot of live actor porn stars are a very long way from the real thing. Or so I’ve heard…
Brachiator
@UncleEbeneezer:
But unfortunately, good enough to cause harm and distress. From October…
I suppose that Photoshop and other tools let you do this before, but I don’t know if AI was faster or more efficient. And some young people have a lot of time to experiment with these tools.
Urza
@marklar: All data can be edited. Even if this were a thing, the day after someone would have a hack freely available online.
sab
@WaterGirl: The people using it (police) don’t care if it only hurts black people.
RSA
This is a nice idea, but so far it hasn’t worked out, for maybe understandable reasons? Large language models, for example, generate the next word in a sentence by figuring out what’s most likely to come next (roughly speaking), based on billions of training examples. It’s not obvious how one might judge that a sentence is artificial, when every word seems to fit.
One of my AI professors in grad school said that we used to think playing chess was hard and playing football was easy, but we learned that it was the opposite. Lately I’ve been hearing that we might see a comparable revolution in robotics to that we’re seeing in language and images, but if so I think it will also come as a surprise.
The field has been called “AI” since the 1950s, so that’s the name we have for its products.
I hadn’t thought of that way of thinking about an LLM, but I don’t know… I think “artificial neural network” gives better intuitions, with the idea of tiny, local, weighted computations being combined on a massive scale. You’re entirely right about the importance of training data, though.
Dorothy A. Winsor
Via Steve Vladeck at BlueSky:
Scout211
@Dorothy A. Winsor: I came here to post that. Finally, an Abbott loss. And some lives saved.
WaterGirl
@TBone: That sounds overwhelming.
Hob
Also, at the risk of leaning too hard on my gun control analogy above, I think that analogy also applies somewhat to the “we need to educate everyone to be more savvy about deepfakes” arguments I’ve seen. I mean, yes, it would be good if more people were less gullible. And it’s also good for people to be aware that there are a lot of nuts with guns out there. But there’s a limit to how much people can reasonably be expected to protect themselves against an enormous mass of danger coming from unknown directions. If I have to wear body armor to the drugstore and the kids have to do mass shooter drills at school, or if I have to come up with a secret way for my mom to prove to me that it’s really her whenever she calls me with a question about some trivial thing (in case it’s a step in a phishing scheme to acquire some other more important information), something has gone very badly wrong. In both cases the threat isn’t unprecedented in itself, it’s just become worse in degree; but it’s worse in degree in a way where the tools have made it possible for a small number of malicious people to do harm to a MUCH larger number of people, MUCH faster than before. When the threat is as asymmetrical as that, an incremental increase in individual defenses can’t possibly keep up.
Leto
@WaterGirl:
comment #61 is a good example.
@Hob: I’ll be sure to file that complaint right next to the one about people cursing too much, and how the youngs don’t respect the hiring process and don’t show up for low paying job interviews.
Also regarding “internet points”: where are these kept? Is there a tally? Can I use those points to redeem them for, say, a Tupperware set? Pyrex bowls? Maybe a new Insta-pot? Or are those points similar to those in Whose Line is it Anyways?
rikyrah
BrooklynDad_Defiant!![]()
(@mmpadellan) posted at 5:30 AM on Mon, Jan 22, 2024:
Make no mistake: No Labels is just a group pretending to be anti-trump that actually does want to help elect him, but they know they can’t get money from moderates by openly admitting that shit.
A vote for No Labels is a vote for trump.
Want to defeat trump?
VOTE BIDEN.
(https://x.com/mmpadellan/status/1749393976170258770?t=obUMNbblffQ53YImfZL8zg&s=03)
prostratedragon
Moreover,
CliosFanBoy
Well,, that was humbling. I went 1 for 10. OUCH
Anoniminous
@WaterGirl:
“A computer is a device for making errors faster than humanly possible” and its many variations has been floating around the tech industry since the mid-70s
Urza
@RSA: It ‘should’ be similar to training it to find cancer cells or the fresh bread at a bakery. Spot the inconsistencies or to perfect images.
UncleEbeneezer
@Brachiator: Ugh. Yeah, I remember guys doing this with physically cutting/pasting the faces girls from yearbook pix onto porn images as a joke among their friends. This was long before Revenge Porn had become such a common thing. AI and other technologies make this sort of thing a much bigger problem.
Leto
@Hob: other nations are able to educate their populace on how to spot/ID misinformation, yet just like with gun, we can’t do anything. This comes back to the fact that conservatives have helped to eradicate critical thinking in our education system, which in turn helps make the populace more susceptible to misinformation. Which is part of their goal: keep the populace dumb/ignorant, continue to push the message of “the other” taking everything, as well as culture war bullshit. We can fix this, but it’s another thing on the never ending list. But it absolutely doesn’t hurt to do what we can, now, to address some of this. In whatever fashion we can.
sab
I got 60%. I used eye level symetry as my method.
Kelly
@WaterGirl: “It’s all one big guess but our program calculates it out to 6 decimal places.”
rikyrah
Jessica Valenti said months ago that their new tactic would be to try and make BAN not a BAN. And, that they were trying to bully the MSM into not saying BAN anymore.
Lauren Windsor (@lawindsor) posted at 6:27 PM on Sun, Jan 21, 2024:
NEW: At Friday’s March for Life breakfast, @RepAndyHarrisMD told my colleague “Rose” how Republicans could pass a federal abortion ban if they win majorities in Congress this year.
One key: Don’t ever say the word ‘ban.’ https://t.co/oehKMX21J7
(https://x.com/lawindsor/status/1749227167756570790?t=Ds-oI5oXOE55KYoCnKk1dQ&s=03)
Leto
@Brachiator: @UncleEbeneezer: here you go:
The Aftermath Of Twitch’s Deepfake Porn Scandal Speaking with Sweet Anita and internet lawyers on the somewhat unexplored territory of deepfake porn
And here’s the thing: this shit is never off the internet. It’s always somewhere. Maybe not in the top hits, but you can eventually get to it. The women involved here are some of the top paid streamers, so they have some more resources to throw at this. But for like the high school students? Just regular people? I know that the post’s article is how they’re using this tech to fool voters, but this was around at least a year ago because the Obama Bros’ podcast did a series on this (a year ago) where they paid for a Biden/Obama voice to read silly shit. It was really funny, but at the same time they said this was coming down the pike. And now here it is.
Anoniminous
Technology is never neutral because technology doesn’t exist until it is instantiated – made real – and then put to a use. Shower facilities are good. The cyanide-based pesticide Zyklon B was used for good. Put shower facilities and Zyklon B together and you get genocide.
AlaskaReader
One of the worst aspects of what we call ‘AI’ is that it isn’t artificial intelligence at all.
It has acquired a label, popularized in the press, that doesn’t properly define what it is.
Machine learning within preset parameters, what we now call ‘AI’, is not an artificial intelligence.
So much misunderstanding drives both those who fear this machine learning as well as those who ascribe properties to it that it cannot and will not produce.
It’s likely too late to put that paste back into the tube, but it isn’t quite like words that change in meaning over a period of time, ‘AI’ started out being imprecise and inexplicit right from the start.
That ‘AI’ is not what we are calling it only confuses it’s impact, how it will be used, and how badly society understands it.
Ruckus
@TBone:
The common part of common sense wasn’t what one has but what one should have. The sense side of that is what determines how large the common side gets.
Anoniminous
@azlib:
OpenAI & etc. flat out stole Intellectual Property such as copyrighted material. When challenged the dickheads flat out said they couldn’t make money if they had to pay for it.
geg6
You are not wrong. It’s going to be a total fucking disaster.
TBone
@UncleEbeneezer: nailed it again.
Tony G
@UncleEbeneezer: Critical thinking is hard work and, with rare exceptions, people tend to avoid hard work. It’s much easier to believe (or pretend to believe) “information” that corresponds to what you want to believe. This tendency long pre-dates “AI”, but “AI” makes it easier to generate false information that’s custom-tailored to each person’s preferences.
Hob
@azlib: LLMs are not search engines. They don’t work in the same way at all, and they’re not well suited to that task. I’m not that kind of software person, I have no expertise in the technology, but I think it’s safe to say that among people who are more deeply familiar with it the consensus is that the only way to make an LLM work well for web search (as opposed to the horribly bullshit-prone “AI search” products that are being pushed right now by companies that ought to know better) is to use it as a front end for an actual search engine, i.e. to figure out what combination of search terms might work well to answer this person’s question— not to do the search itself. It’s fundamentally different from the task they’re designed to do, which is to produce sentences that behave similarly to the sentences it’s seen in the input data, not to reproduce the input data accurately and definitely not to remember where the input data came from.
As for the loose use of the term AI, comment 104 by RSA is correct: that’s nothing new. Within the field, it’s never been used exclusively to mean “a computer that’s intelligent in the same sense that a human being is, able to really understand things and solve unfamiliar problems of any kind.” It’s a field of research that has always included subjects as diverse as natural language processing, image recognition, optimization of specific administrative tasks, and chess. Some subspecialties in that field have focused on trying to really emulate how humans think, but a lot more of it has been about trying to achieve “intelligence-like” results for types of problems that we know humans can do but that computers were traditionally very bad at, as opposed to other kinds of data processing tasks that computers were always better at than humans. Machine learning is somewhere in between: the ability to make the computer do some of the work of devising a problem-solving strategy based on examples, rather than requiring a programmer to figure out the details of that strategy, is a massive advance just in practical terms of what can be done, regardless of whether it really “understands” anything or is “thinking” like a person… and some researchers also feel that it’s a useful step toward real understanding/thinking, while others don’t. Either way, it’s well within the bounds of what AI research has always been about. I agree that it’s still misleading to talk about ChatGPT being “an AI” because people will think of robots in movies etc., but there’s not really a firm distinction like you’re implying.
TBone
@WaterGirl: thank you. Having had long Covid already, it’s scary and enraging – not knowing whether or where I can get Paxlovid here in the hinterlands, since it depends on my “GP’s judgement” according to urgent care doc, is very concerning since I got sick Saturday and am still waiting for PCR test results to come in. A mad scramble for that drug is not optimal, if I’m allowed to have it. Just getting through to GP’s office is a Herculean effort. Thank you for listening, it helps.
WaterGirl
@TBone: Just reading about that is enraging.
TBone
@Ruckus: from yer lips to whatever intelligent spirit has ears!
TBone
@WaterGirl: that’s (partially) why I said earlier my uterus is firing bullets today!
raven
The Delfonics – Ready or Not Here I Come
TBone
@raven: 😎❤️
TBone
https://m.youtube.com/watch?v=aIXyKmElvv8
raven
@TBone: What the hell
The Delfonics – La-La Means I Love You
raven
@TBone: Lauryn kills it.
UncleEbeneezer
@TBone: Ugh, so sorry to hear it. Just got over my rebound Covid. Wishing the best in your effort to procure Paxlovid. It really helped reduce symptoms in my initial case, to a more manageable level. Before that, my throat from Covid was almost as bad as Strep (which I’ve had several times and is the WORST!). Rebound case after Paxlovid was very mild.
raven
@TBone: Shit, I’m going to Costa Rica in five days and trying to dodge it!
Manyakitty
@Mousebumples: I have a Pixel 5 I’m generally happy with. Following, as I’m also getting ready to replace it.
bluefoot
@Hob: When molecular biology techniques (think cloning) became more accessible and reduced to practice, there was a worldwide agreement on a moratorium on human cloning because it was obvious that its ethical framework and consequences needed to be worked out. Even for applying straightforward new technology (like say, mRNA vaccines where the benefit of vaccines in general is well established) there is a huge regulatory framework and burden on the producers before the tech can legally be applied to humans in the real world.
Coming from that community, the difference between the 90s era geneticists/molecular biologists and today’s tech professionals is striking to me. Which is to say, I agree there should be a pause so that ethical and legal frameworks, and exploration on consequences can be done. People tend to respond with “you can’t halt progress” which is true, but we can be more deliberate about how new technology is applied/deployed.
I am realistic about whether or not that’s ever doing to happen.
hw3
Great book on the issue was published in 2020 as a warning, and so far is spot on.
Deepfakes: The Coming Infocalypse
by Nina Schick
Her term Infocalypse is so accurate.
We don’t know how to even teach ourselves how to distinguish from dis- and mis-information. Throw in compelling “evidence” hallucinated wholesale for our pleasure by our “AI” tools, and it’s going to become the societal norm to live entirely in a closed bubble of comfortable information.
TBone
@raven: (((hugs)))
Hob
@Leto:
Look, I know that any complaint about any kind of commenting behavior runs the risk of getting labelled as “old man yells at cloud.” I did at least make an effort though– I said why the “I’ve been told” thing is annoying: because it would’ve been so fucking easy for you to just say who the fuck you’re arguing at from some other thread, instead of getting pissy about “the BJ knowledge base” in a thread where pretty much everyone is fucking AGREEING with you about the issue already. What the fuck ever, you do you, double down on the pissiness if you must, I just wanted to put that out there as someone who’s been hanging out here for 20 years & is accustomed to plenty of snark & swears but would just rather not see it turn into a non-stop passo-aggro sarcasm-fest like some places. I promise I will never mention it again, and I do appreciate that you’re also talking about the actual subject now.
Leto
@Hob: I’m still waiting on where I can redeem the internet points. Good rant though.
artem1s
This didn’t start with AI. This is the result of decades of airbrushing and photo shopping the ‘imperfections’ out of models and actors in fashion magazines and more recently CGI in films to make them appear younger or prettier or more handsome. And of course there is p0rn. bigger tits are better tits! It’s no wonder that people see symmetry and idealized white ‘perfection’ as more real than reality.
TBone
@UncleEbeneezer: thank you, I remain hopeful as is my nature (or I’d be a sad, sorry sack of worthless). Attytood is half the battle!
TBone
@raven: At the risk of overkill (haha) I recommend a constant N-95 mask AND goggles! Face shield! Don’t forget your anti-gravity chin strap and flak jacket!
frosty
@azlib: I prefer GIGO: Garbage In Gospel Out. Goes back to the old days – if it comes out from a line printer, then it’s correct.
The Lodger
@SiubhanDuinne: Remember the nym “Sentient AI from the Future”? I didn’t know if it was Franken or any of those other guys.
TriassicSands
@mrmoshpotato:
I’ll add this: My writing it in a blog will have no more effect than a movie character’s speech had when it was uttered when that film came out. Nor when others have said or written it before or since. It never sinks in and never changes anything despite the fact that with every passing year the repercussions of that failure grow.
Note: For the record, my choice of nym had nothing whatever to do with dinosaurs or movies or anything in popular culture. What it is from always evokes a fond memory for me, although I was not yet born during the Triassic period. That was a few years before my time.
cckids
@Suburban Mom: I know I’m really late to the thread, but my son and daughter-in-law have adopted several senior cats (it’s kinda their thing); their best advice is to have a separate space for the kitty to get adjusted; don’t expect too much interaction at first; let them retreat as needed. It will depend a lot on what kind of life the cat has lived; full of love & attention or having to fend for itself.
Good luck! All three seniors they’ve adopted have turned into absolute lovebugs, regardless of their earlier years & experiences.
The Lodger
@mrmoshpotato: Then again, TriassicSands AFAIK hasn’t begged us for any money.
cain
@patrick II: I hope your drink didn’t taste piss poor. :-)
Glad no #2 was involved!
BruceFromOhio
If I hunker down in the sludge of the contemporary “conservative” mind, it doesn’t matter.
Out here in the real world, same as it ever was: know what motivates, know who to trust, know where your boundary is willing to bend, know where it can’t break ever for any reason. The places I get my info and the people who help me find it will be the first to refute something as a sham. Since the printing press could crank out the words of the printer, there’s been this crap, the press just moves faster now.
Suburban Mom
@rikyrah: Thank you to everyone who gave advice and encouragement. This isn’t a sure thing because it involves a cross-country flight for the cat and everyone involved would prefer a home on the west coast, but if they don’t find one I’m willing. It’s reassuring to know that some of you have done this kind of thing successfully.
Other MJS
I suspect that the ultimate rule will be: don’t trust any digital media unless vouched for by a trusted and verifiable source. Too many people are willing to believe anything that supports their preferred worldview. How much worse deepfakes will make it, I don’t know
Edit: from 1985.
WaterGirl
@Other MJS: Mainstream media has an agenda these days, and I no longer consider them a reliable source. So who would a reliable source be? For a normie who isn’t engaged like we are here.
dnfree
@WaterGirl: My case of prosopagnosia is not as severe as some people’s, but if there’s a movie with two dark-haired men in it, I can’t tell them apart if one doesn’t have a mustache or some other distinctive characteristic. But with this quiz I didn’t have to identify who the person was, just whether it seemed to be a real person.
I had two blonde friends in college whom I could tell apart if they were together, but if I met just one of them I wasn’t sure which one it was.
kalakal
As someone who makes AI images for a bit of relaxation (click on my nym if you want and you can see a few on my redbubble page, use the explore designs tab, it’s quicker ) I do worry about deepfakes. Personally I don’t do photorealistic stuff but I can see how it’s possible. Most of the popular image generators eg Dall -e3 and Stable diffusion tend to produce obvious fakes ( they’re too perfect) but anyone competent with an image editor could work on that.
Having seen how much AI generators can screw up prompts I wouldn’t trust any text generated by them
Another Scott
@Other MJS:
Obligatory, “Obama” (1:12 from 2019).
Cheers,
Scott.
Perilous
I just started using Chat.openai because I am desperate to find a job. 11 months with only a few big breaks that fizzled. My generation of white-collars are really taking a hit right now. (I acknowledge anecdote vs. mass data. But a but it seems a larger-than-usual job seekers are mid-level professionals. I could be wrong.)
Anyhoo, it’s hard to write 8 coherent cover letters a day. Damned if AI doesn’t put out the purest cover letters I’ve ever seen, perfectly tailored to the job. You have to give it the right prompts, but the results are nearly perfect.
In other words written by HAL. While I 100% agree that AI is going to be a problem in the now/future, my immediate needs take precedence. I’m 54 and rebuilding my life. As liberal as I am, at some point I gotta look out for the one tribe.
OzarkHillbilly
Funny how the only choices were Real or Fake? They never give you the option of saying “How in the fuck would I know???? You stupid fucks you…”
Paul in KY
@WaterGirl: I was 40%. blasted on the AI faces 1/5. Didn’t really study them that closely, though.
WaterGirl
@Paul in KY: I just went on gut feeing.