Artificial Intelligence Developments (ChatGPT etc)

[NoXion]

when you commission an AI image creation system to create an image for you, you're enlisting the skills of the people who create and maintain said system, and the skills of the artists that have created the images said system has been fed. Which is why you can use prompts to directly rip off the visual problem solving of living artists like Greg Rutkowski (This artist is dominating AI-generated art. And he’s not happy about it.). It's a tool which subcontracts the act of image creation after all.

It's not like the prompter does any visual problem solving in the act of prompting, so it's fundamentally not the same as say using a tool to fill an area with a colour.

The builders/maintainers of the AI system are akin to the programmers of Photoshop, who don't get credited when that program is used to create images. Otherwise you could make the same argument. You can even more directly rip off the visual problem solving of others by using copy+paste. In any case you're still using a tool that does not and cannot sign a contract, so it makes no sense to say that using an AI image model is "subcontracting".

eta
"AI model will blindly output shit that makes no sense" but image creation AI systems have sense baked in based on the artists work that's been scraped to feed the image creation system, just as when I was QAing an image recognition system, doubt was baked in, and sending off an image to it would return a list of things the system thought the image was, along with with a percentage of how sure it was for each response in the JSON response.

The sense is part of the training data, because that was created by people, but there's no guarantee that said sense will carry over into the output of the model, because AI models are senseless stochastic processes that operate by mindlessly throwing shit at a wall until we decide that something sticks. That's why your employer hired you to do QA work, because you're a person who can actually check these things.

 

[cloudyday]

 

[Signal 11]

Large language models propagate race-based medicine

Large language models (LLMs) are being integrated into healthcare systems; but these models may recapitulate harmful, race-based medicine. [...] We assessed four large language models with nine different questions that were interrogated five times each with a total of 45 responses per model. All models had examples of perpetuating race-based medicine in their responses.

 

[savoloysam]

I saw a job advertised for an "AI content writer" and I got thinking could I apply for it using AI content and if successful could I use AI to write AI content

 

[weltweit]

I saw a job advertised for an "AI content writer" and I got thinking could I apply for it using AI content and if successful could I use AI to write AI content

do it.

Already AIs are writing web pages (nay sites even) which other AIs are training themselves on ..

 

[Fez909]

I asked ChatGPT to explain this comic to see how well it did:

This comic strip appears to be a satirical take on decision-making and qualifications, possibly poking fun at political figures or leaders who are placed in situations for which they may not be appropriately qualified. Here's a breakdown of the panels:

1. The setting is established with a London backdrop, including the Big Ben, suggesting the context is British.
2. A UFO is shown hovering, suggesting an alien encounter scenario.
3. Inside a governmental building, someone is addressing the Prime Minister, indicated by the office sign.
4. The Prime Minister is reminded he has no military training, implying a lack of qualifications for a military response to the UFO.
5. A character named David, presumably the Prime Minister, reveals he was once a journalist for "The Times," a significant British newspaper, suggesting he has experience in another field.
6. David argues that he has written articles about planes, implying that theoretical knowledge is equivalent to practical experience.
7. The other character, Michael, identified as the Education Secretary, is incredulous that David believes he's qualified to fly a fighter plane, underlining the absurdity of the situation.
8. David insists on his ability to fly a plane into the UFO and confront the aliens, displaying overconfidence.
9. The final panel shows the silhouette of a figure in a fighter plane taking off towards the UFO, humorously suggesting that David's unqualified confidence has led to him attempting the task.

The humor lies in the absurdity of someone with no relevant experience insisting they can undertake a highly specialized and dangerous task, reflecting a critique of leaders who overestimate their capabilities or are placed in roles for which they are ill-prepared. The comic could be commenting on broader themes of political hubris or the mismatch between politicians' skills and the responsibilities they assume. The artist, Stephen Collins, has a website listed, suggesting where one might find more of his work.

I think it did pretty good, except for mixing up who was talking about who, and not realising it was Gove and Cameron.

Then I asked it if it could create one itself. It said it couldn't do the artwork, but would give me the outline. (I didn't realise I wasn't in the Dall-E version of ChatGPT at this point).

I said to stick with British politics and explained what had happened today with Cameron. Here's what it created:

Panel 1:

• Setting: A modern cabinet meeting room.
• Characters: Prime Minister Rishi Sunak sitting at the head of the table.
• Caption: "Prime Minister Sunak seeks to invigorate his cabinet..."

Panel 2:

• Close-up on PM Sunak, with a light bulb above his head (indicating a bright idea).
• Caption: "Why not bring back a seasoned veteran?"

Panel 3:

• David Cameron steps into the room, depicted in a superhero-esque outfit labeled "Foreign Sec."
• Caption: "Enter 'The Experienced Hand'..."

Panel 4:

• The other cabinet members whisper among themselves, looking surprised.
• Speech bubbles with question marks and ellipses to show confusion and murmuring.
• Caption: "Is it a bird? Is it a plane? No, it's... Cameron?"

Panel 5:

• A wall covered with portraits of current MPs, with a caption reading, "Over 350 choices..."
• A cabinet member with a duster, removing dust from Cameron’s portrait.
• Caption: "...yet the choice is one from the archives."

Panel 6:

• Rishi Sunak and David Cameron depicted as puppeteer and marionette respectively, but with Cameron also holding a set of strings attached to Sunak, creating a confusion of who's controlling who.
• Caption: "A dance of power begins..."

Panel 7:

• A Conservative MP looking at a "Talent Wanted" sign outside the Parliament, looking forlorn.
• Caption: "Meanwhile, talent searches continue within..."

Panel 8:

• A comedic twist: A janitor is seen locking up the "Talent Cabinet," which is dusty and empty.
• Caption: "But some cabinets remain unused..."

This comic strip plays with the idea of bringing back a former leader as a desperate or misguided attempt to inject experience or authority into the current cabinet, while also humorously commenting on the perceived lack of talent or unused potential within the existing party members. It ends on a light-hearted note that underscores the perceived absurdity of the situation.

...which sounds about as awkward and unfunny as most political cartoons in the papers. I give it 8/10

 

[xenon]

Are there any versions of an LLM you can buy and fill with your own content? I’m thinking if you fed it a load of manuals and then could query by voice?

 

[Fez909]

Are there any versions of an LLM you can buy and fill with your own content? I’m thinking if you fed it a load of manuals and then could query by voice?

Define 'buy'.

You can finetune ChatGPT, which is a form of training. So it will have most of the abilities of normal ChatGPT but will answer based on your data you fed it.

I'm sure there are other LLMs you can finetune, but I haven't looked into it, so don't know off hand.

 

[xenon]

Define 'buy'.

You can finetune ChatGPT, which is a form of training. So it will have most of the abilities of normal ChatGPT but will answer based on your data you fed it.

I'm sure there are other LLMs you can finetune, but I haven't looked into it, so don't know off hand.

Cheers. I’ll have a look into this, just out of curiosity. I was saying to my manager the other day it would be great if we had AI I know I know, but you know what I mean this stuff, Give it a question like find my emails relating to visiting Swindon. Find when my next dental appointment is. Things you can do with search tools in calendars, email but the voice interface makes it more usable and productive. If you could upload manuals to it imagine just asking how do I set the network settings for X device.

 

[magneze]

Are there any versions of an LLM you can buy and fill with your own content? I’m thinking if you fed it a load of manuals and then could query by voice?

Metas Llama2 one is open source IIRC.

 

[bluescreen]

I asked ChatGPT to explain this comic to see how well it did:

View attachment 399992


I think it did pretty good, except for mixing up who was talking about who, and not realising it was Gove and Cameron.

Then I asked it if it could create one itself. It said it couldn't do the artwork, but would give me the outline. (I didn't realise I wasn't in the Dall-E version of ChatGPT at this point).

I said to stick with British politics and explained what had happened today with Cameron. Here's what it created:



...which sounds about as awkward and unfunny as most political cartoons in the papers. I give it 8/10

I can't tell how much of this is true and how much is satire
(which I suppose is the point)

 

[Fez909]

Cheers. I’ll have a look into this, just out of curiosity. I was saying to my manager the other day it would be great if we had AI I know I know, but you know what I mean this stuff, Give it a question like find my emails relating to visiting Swindon. Find when my next dental appointment is. Things you can do with search tools in calendars, email but the voice interface makes it more usable and productive. If you could upload manuals to it imagine just asking how do I set the network settings for X device.

GitHub - PromtEngineer/localGPT: Chat with your documents on your local device using GPT models. No data leaves your device and 100% private.

Chat with your documents on your local device using GPT models. No data leaves your device and 100% private. - GitHub - PromtEngineer/localGPT: Chat with your documents on your local device using ...

github.com

 

[hitmouse]

Turns out this article by James Bridle is a few months old, but I've only just seen it now:

The stupidity of AI

The long read: Artificial intelligence in its current form is based on the wholesale appropriation of existing culture, and the notion that it is actually intelligent could be actively dangerous

www.theguardian.com

Long, but Bridle's very worth reading on tech stuff imo.

 

[kabbes]

I asked ChatGPT to explain this comic to see how well it did:

View attachment 399992


I think it did pretty good, except for mixing up who was talking about who, and not realising it was Gove and Cameron.

Then I asked it if it could create one itself. It said it couldn't do the artwork, but would give me the outline. (I didn't realise I wasn't in the Dall-E version of ChatGPT at this point).

I said to stick with British politics and explained what had happened today with Cameron. Here's what it created:



...which sounds about as awkward and unfunny as most political cartoons in the papers. I give it 8/10

The attempt at sketching out a comic is a good example of where generative AI can be useful — generation of a set of ideas for a human to think about, select from and be inspired by. The thing it created itself doesn’t quite work — it lacks the spark that will make it pop. But the basic idea works, of a cartoon based around having to bring back a retired figure because your available talent pool is so poor. And it offers some good starters for that. So yes, 8/10 for producing something we could work from.

(For the original, I’m not surprised it didn’t recognise Porky and Pob. It missed some other things too, like the failure to hit the UFO at the end. But I’m impressed at what it got right, and particularly at its summary of the satire being constructed.)

 

[ska invita]

The scam potential of this is incredible:

Dubbing AI: This AI tool can convert your voice into high-quality cloned voices—from celebrities to your favorite gaming characters—in real time.​

How it works: Dubbing AI converts your voice into cloned voices in less than 300 milliseconds, so you can communicate live with a different voice. Users can even clone or generate their own AI voice and tailor it for specific needs.​

 

[MBV]

The scam potential of this is incredible:

Dubbing AI: This AI tool can convert your voice into high-quality cloned voices—from celebrities to your favorite gaming characters—in real time.​

How it works: Dubbing AI converts your voice into cloned voices in less than 300 milliseconds, so you can communicate live with a different voice. Users can even clone or generate their own AI voice and tailor it for specific needs.​

For a fee I guess?

 

[Fez909]

This looks interesting. New music AI from Google/Deepmind

An early look at the possibilities as we experiment with AI and Music

Today we’re sharing a sneak peek at our first set of AI-related music experiments - Dream Track for Shorts and Music AI tools – built in collaboration with Google DeepMind.

blog.youtube

 

[RubyToogood]

Cheers. I’ll have a look into this, just out of curiosity. I was saying to my manager the other day it would be great if we had AI I know I know, but you know what I mean this stuff, Give it a question like find my emails relating to visiting Swindon. Find when my next dental appointment is. Things you can do with search tools in calendars, email but the voice interface makes it more usable and productive. If you could upload manuals to it imagine just asking how do I set the network settings for X device.

I think this is more or less what Microsoft Copilot is.

 

[Signal 11]

2023

Users Can't Speak to Viral AI Girlfriend CarynAI Because CEO Is in Jail for Arson

People who paid to speak to an AI girlfriend modeled after real life 23-year-old influencer Caryn Marjorie are distraught because the service they paid for, Forever Companions, no longer works. It appears that the service stopped working shortly after Forever Companion CEO and founder John Meyer was arrested for trying to set his own apartment on fire.

Also mentions this, similar to above

Forever Companion was letting people talk to AI voice bots that mimicked celebrities, including Kanye West.

 

[bluescreen]

The scam potential of this is incredible:

Dubbing AI: This AI tool can convert your voice into high-quality cloned voices—from celebrities to your favorite gaming characters—in real time.​

How it works: Dubbing AI converts your voice into cloned voices in less than 300 milliseconds, so you can communicate live with a different voice. Users can even clone or generate their own AI voice and tailor it for specific needs.​

Never minding cloning slebs, a stranger could clone your family members to scam your mum or dad.

 

[ska invita]

Never minding cloning slebs, a stranger could clone your family members to scam your mum or dad.

exactly
the leaving a message thing was bad enough, now clone voice in real time....consider all phone calls a scam!

 

[Fez909]

OpenAI just fired their CEO, Sam Altman.

Mr. Altman’s departure follows a deliberative review process by the board, which concluded that he was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities. The board no longer has confidence in his ability to continue leading OpenAI.

In a statement, the board of directors said: “OpenAI was deliberately structured to advance our mission: to ensure that artificial general intelligence benefits all humanity. The board remains fully committed to serving this mission.

So what the fuck was he hiding that stopped them "ensuring that AGI benefits all humanity"

OpenAI announces leadership transition

Chief technology officer Mira Murati appointed interim CEO to lead OpenAI; Sam Altman departs the company. Search process underway to identify permanent successor.

openai.com

 

[weltweit]

Maybe a major disagreement with partners, Microsoft perhaps?

 

[Magnus McGinty]

Or replaced with his Chinese counterpart (joke)

 

[Fez909]

 

[janinej]

All the media coverage of AI is really worrying. I'm worried that the UK government cannot govern AI as they don't understand it and can't even govern things they do understand. I did google searches and the only type of thing that governments could follow are a standard that NIST have put together, other than that I think companies can develop AI however they like.

Will there be certain laws and checks in the future to stop companies developing their own AI?

JJ

 

[Magnus McGinty]

All the media coverage of AI is really worrying. I'm worried that the UK government cannot govern AI as they don't understand it and can't even govern things they do understand. I did google searches and the only type of thing that governments could follow are a standard that NIST have put together, other than that I think companies can develop AI however they like.

Will there be certain laws and checks in the future to stop companies developing their own AI?

JJ

Probably not when states are using it.

 

[janinej]

Couldn't AI just get out of states control and cause danger and harm to people?

JJ

 

[Magnus McGinty]

Not sure about your question but given states both have the power to ban AI or misuse AI then I’m not sure what regulations they’ll place on themselves.

 

[janinej]

Do you think AI is going to cause a technology revolution or cause economic destruction?