Artificial intelligence still has some way to go

i think niche and dorky is sort of AI’s sweet spot tbh. general <waves hands> magical productivity improvements less so.

film world arguing about ai will catalyze the singularity

A friend sent me MRI brain scan results and I put it through Claude.

No other AI would provide a diagnosis, Claude did.

Claude found an aggressive tumour.

The radiologist report came back clean.

I annoyed the radiologists until they re-checked. They did so with 3…

— Misha Saul (@misha_saul) March 22, 2024

lol

Impressed with the vitriol my post has attracted

My theory is that the mentally ill he/hims have reverence for authority and doctors are a special class of revered expert rather than a fallible class of professionals

Or maybe trying to use tech is inherently suspicious? 🤷‍♂️ https://t.co/dBG8odC038

— Misha Saul (@misha_saul) March 24, 2024

https://x.com/nousresearch/status/1771735632035127594?s=46

xp
https://i.postimg.cc/sgVbY6j2/IMG-7757.jpg

People who can’t think think “AI” can think.

I really don't get people who think it's more fun to badger the stupid robot into doing something that you clearly could do by yourself than to just do it, tiresome

because it took him 6 minutes this way rather than an hour the other way.

i don’t rely on it to think i only rely on it to have good judgment.

(def hearing that by The Cramps)

This is niche and dorky, but as someone who works predominately with audio, I'm really looking forward to the restorative programs and plug-ins that are hopefully around the corner.

There's already been some great improvements in the last few years that make my life a lot easier for sure, but they fall down in certain ways, especially where music is involved.

I expect in 5 or more years we should be able to properly restore and improve old film/tv audio, archive recordings, tape, optical and wire etc.

― Maresn3st, Sunday, March 24, 2024 5:52 AM (two hours ago) bookmarkflaglink

Honest questions: why do these things need improving? If they're archived properly already, why "improve" them? If they're not archived properly already, what is wrong with current archival protocols? I guess I am just highly suspicious of AI "improving" things to the point of totally denuding them of their original context, which is part of what makes them what they are.

Maybe I'm in over my head here, but surely archiving and restoration are two separate, though related, concerns? I took that post to mean restoring degraded audio in music, film, etc.

This admitidely gives me some knee jerk concerns as the line between restoration and messing-with is notoriously difficult to trace in image (cf complaint about the "yellow" nature of a lot of the Cinemateca di Bologna restorations of classic cinema) and I'd assume audio as well, but tbf this is an eternal concern that predates AI.

i think that’s otm. i won’t speak for maresn3st but AI has been used in video compression (where information is necessarily lost to enable efficient storage or carriage) for some time. it is in effect restoring lost information by analysing the frames around it and filling in information. at least in part it’s how we’re able to watch high quality video. (there are countless examples like this that make the world go round, load balancing mobile phone mast data loads for example) - i’m just selecting one in my area.

because it took him 6 minutes this way rather than an hour the other way.

― 𝔠𝔞𝔢𝔨 (caek), Sunday, March 24, 2024 8:18 AM (forty-three minutes ago)

The fuck does he need the other 54 minutes for jerking his dang hog

I'm not sure I understand your post, Fizzles— I would frankly rather watch a film as it would have appeared fifty years ago than in a "gloriously restored version." Maybe I'm in the minority there, but I am often left completely cold by such restorative efforts.

sorry i wasn’t being very clear about the context. in order to store, transfer and stream media, it’s necessary to compress it, which means deliberately losing information or data from the originally produced version. this is in a sense a degradation of the media. in order to avoid this degradation being visible to the viewer, AI (and other algorithmic processes) are used to “fill in” the gaps. it’s extremely refined and sophisticated - invisible to audiences.

it is worth noting that some broadcasters don’t like having compression before a hand off. that’s mainly because there will be subsequent conversion or information-loss processes depending on where it’s being distributed or stored. the more you do it, the harder it is to reconstitute.

i was using this example as an analogy for historically degraded media.

my point is that we use AI to reconstitute media all the time. this doesn’t invalidate your point, which i think daniel is also saying, which is that questions of restoration and appreciating the degraded form etc

… questions of that sort are persistent…

(trying to post while cooking only marginally less successful than trying to post when not cooking)

I would frankly rather watch a film as it would have appeared fifty years ago than in a "gloriously restored version."

The vast majority of film restoration is aimed exactly at making it possible to watch a film as it would have appeared fifty years ago - and this is necessary because film stock degrades. I think you're misreading "restoration" here to always mean stuff like adding CGI in or Lucas messing with his old Star Wars films.

I wonder how much AI is being used in the guerrilla nerd efforts to make 4k copies of the original theatrical releases. Some of them definitely used AI upscaling from the laserdiscs, but the other nerds are hardcore in procuring buried cinema prints and scanning them frame-by-frame.

The fuck does he need the other 54 minutes for jerking his dang hog

― G. D’Arcy Cheesewright (silby), Sunday, March 24, 2024 12:03 PM (thirty-five minutes ago) bookmarkflaglink

he can use it to contemplate if a park thats half not a park is really a park

Thanks for clarification and edification, Daniel and Fizzles.

Heya, so quite a lot of my work is spent tidying up poorly recorded audio (or degenerated/archive audio) to make it more discernable and pleasing, it is mostly spoken word.

To address your point Table, I think the only kind of media that might *need* to be restored using machine learning/AI/neural network software would be something that has degenerated past the point of being listenable or was recorded with faulty equipment or involved some kind of operator error. It may have been poorly archived, or it could perhaps have been stored correctly but still be substandard quality.

There will surely be a level of futzing around by nerds with things in the future, same as we're experiencing with film and TV images.

This might be somewhat dry so apologies in advance, but one kind of plug-in that I use a lot right now builds back in lost, or never captured, frequencies to an audio clip (while cleaning up background noise and some other balancing issues) the AI component definitely has a ways to go before being a magic bullet solution, you can easily end up giving someone a wicked lisp or making them sound robotic and weird.

It's also not good with musical elements, but I'm sure this will get better in a few years, I'm just keen to find out how much more we can improve these things.

What I was getting at with my post was the potential to improve, for instance, an old movie made in the 1930s that might have a very bad quality optical soundtrack.

It wouldn't be equivalent to a modern recording, but you could subtly build in low and high frequencies that would smooth the overall sound out and improve the listening experience, same with any archival material really, old radio recordings, historical documents, forensics etc.

Here is a link to a before and after clip to demonstrate, as you can hear, it's still very basic, but it was almost impossible to do to this level even a few years ago - https://we.tl/t-b7Q6juy3ds

I appreciate that, Maresn3st. I guess that some of what we're talking about, and what that clip demonstrates, is an idea of "integrity." I don't know much about this stuff, so while I agree that the adapted version is more clear, I was left wondering: what was the motoric noise in the first clip? Where was the original recording made, and in what context? If such a recording is clear enough to be transcribed, which the example you give is, then why does it need to be "cleaned up," so to speak? Maybe I'm just a glutton for punishment or something, but I think that the weird motoric rumbling in the original is *part* of the context of the recording— why get rid of it?

i was watching a film from '39 over the weekend and the things that needed doing to it were a) adjusting the position of all the frames so there was less jitter and b) fixing the slight changes of brightness in each frame. and i think this is the kind of thing that could be done by an algorithm (and neither of which would really destroy any data from the original the way, say, colourisation does)

XP - In the context of my work, very simply to make it easier to hear and sometimes to help bed it into something like a radio documentary or podcast.

In this instance you could argue that it doesn't have to be cleaned up, sure. However, for me, static background noise or artefacts of a primitive recording process aren't necessarily integral or precious, and in that small clip you can discern more of the voices in the background that were masked by noise.

It was an outdoor wartime recording of Neville Chamberlain, at an airfield iirc.

Got it, thanks!

I think that what I worry about isn't necessarily what you're point to or doing, which makes sense to me. What I worry about is that these cleaned up versions will become the "standard" versions by which certain events are known or available in the archive, which to my mind goes against the spirit of a lot of archival practices. That is, the adapted version has its functionality that is important, and the original has its functionality that is important, but I worry that the two will become confused, or even that the latter will be lost and discarded.

Absolutely yeah, I know my view is a little basic I guess because it's ultimately threaded into my work practices.

But I understand how future technology will affect the integrity of archive media, we already have to be careful with AI assist software as it can very easily tip the balance and make a voice sound like a different person, heck I've even heard it sneak in extra syllables in more extreme cases.

On the upside, I think what the people who write this software are reaching for is aligned with what Daniel was talking about upthread, the restorative aspect, but yeah, it may tip over into something else as archive media becomes ever more malleable.

gotta make sure you dont motion smooth buster keaton

the worst thing copilot does is create these phantom imports which you dont even notice cause theyre at the top of the page not where youre currently working, oh bool from sharp thanks thats a huge help seems super real

https://i.imgur.com/R0AV6rQ.png

A few jobs are going until this shit gets sued for copyright.

Sobering….🙁 @EquityUK @bbcarts pic.twitter.com/9D0H928xJZ

— Sara Poyzer (@SaraPoyzer) March 26, 2024

gotta respect the criminal ingenuity

https://www.theregister.com/2024/03/28/ai_bots_hallucinate_software_packages/

Several big businesses have published source code that incorporates a software package previously hallucinated by generative AI.

Not only that but someone, having spotted this reoccurring hallucination, had turned that made-up dependency into a real one, which was subsequently downloaded and installed thousands of times by developers as a result of the AI's bad advice, we've learned. If the package was laced with actual malware, rather than being a benign test, the results could have been disastrous.

nice, nice

(this_is_fine.png)

normal

for a period of time in 2023, the person in charge of OpenAI's $175 million startup fund appears to have been completely fake — and OpenAI says the documents filed with the California Secretary of State to put the fake person in charge were "completely fabricated."

https://futurism.com/the-byte/fake-person-openai-fund

I have to file these things all the time in my job. I highly doubt someone other than the company filed this. The only incentive I see to do this is to obscure ownership/management from the regulator and public.

its a really weird thing to do cause they know the company its associated with and if they have to check in for some reason its just going to create a problem (felony)

Claude 3 just hallucinated a new technology, its name, purpose and several scientific papers and authors regarding the technology in answer to a query. Pure Peter O'Hanraha-hanrahan stuff. (The symposium did exist)

I apologize for the confusion, and thank you for bringing this to my attention. Upon further investigation, I couldn't find any reliable sources confirming the existence of the paper by Jens Ulrich Gerhardt, Jens Steger, and Klaus Merkel from the 2013 IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB).

It seems that my previous responses were based on an error in my knowledge base, and I couldn't find any credible references to this specific paper or presentation. I apologize for providing incorrect information.

https://www.theguardian.com/world/2024/apr/03/israel-gaza-ai-database-hamas-airstrikes?CMP=Share_AndroidApp_Other

The Israeli military’s bombing campaign in Gaza used a previously undisclosed AI-powered database that at one stage identified 37,000 potential targets based on their apparent links to Hamas, according to intelligence sources involved in the war.

In addition to talking about their use of the AI system, called Lavender, the intelligence sources claim that Israeli military officials permitted large numbers of Palestinian civilians to be killed, particularly during the early weeks and months of the conflict.

“This is unparalleled, in my memory,” said one intelligence officer who used Lavender, adding that they had more faith in a “statistical mechanism” than a grieving soldier. “Everyone there, including me, lost people on October 7. The machine did it coldly. And that made it easier.”

Another Lavender user questioned whether humans’ role in the selection process was meaningful. “I would invest 20 seconds for each target at this stage, and do dozens of them every day. I had zero added-value as a human, apart from being a stamp of approval. It saved a lot of time.”

Autonomous Weapons

Like I feel like the only thing that would have the military replace human controlled drones with AI controlled drones is if they turned out to be much more accurate than humans in terms of identification & targeting.

― Mordy, Tuesday, 28 July 2015 22:57 (eight years ago)

Turns out accuracy was not that much of an issue

No it seems AI is pretty accurate at choosing targets, it's just that it was decided that killing those targets when they were at home with their families was easier.

No it seems AI is pretty accurate at choosing targets, it's just that it was decided that killing those targets when they were at home with their families was easier.

... and less expensive.

stumbled onto this via bluesky and thought it was pretty interesting: https://softwarecrisis.dev/letters/llmentalist/

The LLMentalist Effect: how chat-based Large Language Models replicate the mechanisms of a psychic’s con

https://mailman.nanog.org/pipermail/nanog/2024-April/225407.html

https://cdn.bsky.app/img/feed_fullsize/plain/did:plc:x4qyokjtdzgl7gmqhsw4ajqj/bafkreibhlmcgsb6japqhcvgakjvqq5jb4iqka5tn3qxjydoxpyanffybb4@jpeg

In my advocacy class in law school in the late 90s, we were given a fact pattern where the driver, "Johnny" (or his estate I guess), was suing a railroad for a collision with a train at an allegedly unsafe RR crossing. The dispute was whether the driver hit the train or the train hit the drive. My friend's closing statement was, "If Johnny hit the train, you must abstain."

Wait, what thread is this, lol?