Artificial Intelligence [ChatGPT rn]

[Kraken]

To take this to a far-ish future example that may be absurd (but which none of us here knows is absurd), we're not going to care altogether too much whether it has a holistic understanding of the cure for cancer it eventually develops, or whether it came to a cure by simply analyzing successive iterations of work against a series of inputs. Just that it gets there.

Exactly. We're seeing this already in radiology. Multiple studies have shown that ML classifiers (not LLMs but CNNs mostly) were more accurate than human radiologists by 25-30%. It probably wouldn't replace a radiologist but it could reduce the need to hire as many radiologists in each system.

Have you heard about DeepMind?

DeepMind has predicted the structure of almost every protein so far catalogued by science, cracking one of the grand challenges of biology in just 18 months thanks to an artificial intelligence called AlphaFold. Researchers say that the work has already led to advances in combating malaria, antibiotic resistance and plastic waste, and could speed up the discovery of new drugs.

Determining the crumpled shapes of proteins based on their sequences of constituent amino acids has been a persistent problem for decades in biology. Some of these amino acids are attracted to others, some are repelled by water, and the chains form intricate shapes that are hard to accurately determine.
...
Demis Hassabis, CEO of DeepMind, says that the database makes finding a protein structure – which previously often took years – “almost as easy as doing a Google search”. DeepMind is owned by Alphabet, Google’s parent company.

The archive has already been used by scientists to advance research in a number of areas. Matt Higgins at the University of Oxford and his colleagues were researching a protein that they believed was key to interrupting the lifecycle of the malaria parasite, but were struggling to map its structure.

“One of the experimental methods that we use is X-ray crystallography,” says Higgins. “We cause the proteins to form into lattices, fire X-rays at them and get information from those X-ray diffraction patterns to see what the molecule looks like. But we were never able, despite many years of work, to see in sufficient detail what this molecule looks like.”

But when AlphaFold was released, it gave a clear prediction of the structure of the protein that matched the information the researchers had been able to glean. They have now been able to design new proteins that they hope could serve as an effective malarial vaccine.

Birney says that using X-ray crystallography to map the structure of a protein is expensive and time-consuming. “That means that experimentalists have to make choices about what they do, and AlphaFold hasn’t had to make choices,” he says. “I think we can be confident that there are new experiments and new insights coming through due to AlphaFold, which will impact ‘how does this particular parasite work’ or ‘why does this particular disease happen in humans’, for example.”

[Kurth]

I listened to both podcasts, and I'm glad I did Eliezer Yudkowsky first, as it is probably the most depressing session I have heard in years.

Same. I agree that while it's clearly not AGI, it's already immensely disruptive in many different areas. It's not replacing RM9's team of developers (yet), but focusing on whether it 'understands' its output is a mistake. It's doing useful things today, and the pace of development is scorching.

I think the notion that "it's clearly not AGI" is an interesting one, especially in light of the back and forth between Eliezer and Lex and between Sam and Lex. All seemed to agree that while we might be able to defend the position that GPT4 is not AGI right now, they all seemed genuinely challenged to come up with a bright line for when AI crossed over to the realm of AGI. This bit by Eliezer regarding how AI development is like "boiling a frog" struck me:

It probably isn't happening right now. We are boiling the frog. We are seeing increasing signs bit by bit, but not like spontaneous signs. Because people are trying to train the systems to do that using imitative learning. And the imitative learning is like spilling over and having side effects. And the most photogenic examples are being posted to Twitter. Rather than being examined in any systematic way. So when you are boiling a frog like that, what you're going to get like . . . a thousand people looking at this. And the one person out of a thousand who is most credulous about the signs is going to be like, "that thing is sentient." While 999 out of a thousand people think, almost surely correctly, though we don't actually know, that he's mistaken. And so the like first people to say like, "sentience," look like idiots. And humanity learns the lesson that when something claims to be sentient and claims to care, it's fake because it is fake. Because we have been training them using imitative learning rather than, and this is not spontaneous. You're going to have a whole group of people who can just like never be persuaded of that. Because to them, like being wise, being cynical, being skeptical is to be like, oh, well, machines can never do that. You're just credulous. It's just imitating. It's just fooling you. And like they would say that right up until the end of the world. And possibly even be right, because, you know, they are being trained on an imitative paradigm. And you don't necessarily need any of these actual qualities in order to kill everyone, so.

To take this to a far-ish future example that may be absurd (but which none of us here knows is absurd), we're not going to care altogether too much whether it has a holistic understanding of the cure for cancer it eventually develops, or whether it came to a cure by simply analyzing successive iterations of work against a series of inputs. Just that it gets there.

I continue to be puzzled by the notion that thoughts of AI curing cancer are thoughts about a "far-ish future . . . that may be absurd." Why? Given the rate that current AI capabilities are increasing, why would we think that?

[Zaxxon]

I'm couching my language because we don't know.. I'm with you that it could happen much more quickly. But RM9 isn't crazy that it could... not.

[malchior]

To take this to a far-ish future example that may be absurd (but which none of us here knows is absurd), we're not going to care altogether too much whether it has a holistic understanding of the cure for cancer it eventually develops, or whether it came to a cure by simply analyzing successive iterations of work against a series of inputs. Just that it gets there.

Exactly. We're seeing this already in radiology. Multiple studies have shown that ML classifiers (not LLMs but CNNs mostly) were more accurate than human radiologists by 25-30%. It probably wouldn't replace a radiologist but it could reduce the need to hire as many radiologists in each system.

Have you heard about DeepMind?

Yep. Another great example.

I'm couching my language because we don't know.. I'm with you that it could happen much more quickly. But RM9 isn't crazy that it could... not.

Sure but that outcome is pretty low probability. Maybe I need to ask the AI?

Seriously though there are use cases where it'll absolutely change everything. There is no question to that now. It's already happened and people just don't see it yet. Biotech and pharmaceuticals are good candidates for big time disruption.

Will it replace large swaths of white collar labor? Probably not but it's a possibility and if history tells us anything we aren't ready for that. That scenario wouldn't be solved in a fair or equitable way. So we should be wary of it. Mostly because the non-AIs are the biggest problem.

[Zaxxon]

[Smoove_B]

2029
There is a growing discussion about the legal rights of computers and what constitutes being human

[Zaxxon]

It's also worth noting that these one-prompt examples are neat, but the real usefulness comes from repeated, targeted sessions involving dozens of prompts. Is ChatGPT going to write your game? No. But it would go a lot further than what's posted here.

Open-world RPG it is not, but this is cool nonetheless.

[Kurth]

To take this to a far-ish future example that may be absurd (but which none of us here knows is absurd), we're not going to care altogether too much whether it has a holistic understanding of the cure for cancer it eventually develops, or whether it came to a cure by simply analyzing successive iterations of work against a series of inputs. Just that it gets there.

Exactly. We're seeing this already in radiology. Multiple studies have shown that ML classifiers (not LLMs but CNNs mostly) were more accurate than human radiologists by 25-30%. It probably wouldn't replace a radiologist but it could reduce the need to hire as many radiologists in each system.

Have you heard about DeepMind?

Yep. Another great example.

I'm couching my language because we don't know.. I'm with you that it could happen much more quickly. But RM9 isn't crazy that it could... not.

Sure but that outcome is pretty low probability. Maybe I need to ask the AI?

Seriously though there are use cases where it'll absolutely change everything. There is no question to that now. It's already happened and people just don't see it yet. Biotech and pharmaceuticals are good candidates for big time disruption.

Will it replace large swaths of white collar labor? Probably not but it's a possibility and if history tells us anything we aren't ready for that. That scenario wouldn't be solved in a fair or equitable way. So we should be wary of it. Mostly because the non-AIs are the biggest problem.

I was talking with someone the other day about AI and disruption and the fact that AI is going to make a veritable shit ton of jobs obsolete sooner rather than later. I think all of that is true. Every great technological shift results in labor displacement.

But compared to the other issues we are facing with AI, rising unemployment and labor disruption due to AI and automation should really be the least of our concerns.

[malchior]

But compared to the other issues we are facing with AI, rising unemployment and labor disruption due to AI and automation should really be the least of our concerns.

I don't know - this seems like this could lead to immediate problems. It depends on scale and all that but this is towards the top of the list IMO. This is probably a longer discussion but this could end up in magnifying wealth distribution issues, stressing financial systems, impacting the dire political situation, etc.

Edit: It could help too. For example, there is a known use case in cybersecurity. There is a real need for an army of cybersecurity people who simply don't exist right now. This is something I work on part-time because there are real opportunities and good matches with current AI tech for that. When incidents occur there is a period of analysis required. If we can "automate" some of that and present options to a human operator it could have lots of benefits but we're early times there right now.

[RunningMn9]

You have a remarkable clarity to know what a technology will be capable of 10-20 years down the road, when looking back half that far would confirm that we have no earthly idea what will happen over the next decade-plus in this space. You keep coming back to the fact that it's not replacing teams of programers today as though that's news to anyone paying more than peripheral attention.

I just know when I smell bullshit. This smells like bullshit.

The giveaway is the breathless evangelism and fighting over which superlatives / buzzwords to throw at it. The same people time and again. Certainly THIS time they might be right. We'll see.

[Kraken]

We are hurtling toward a glitchy, spammy, scammy, AI-powered internet

for a type of attack called indirect prompt injection, all you need to do is hide a prompt in a cleverly crafted message on a website or in an email, in white text that (against a white background) is not visible to the human eye. Once you’ve done that, you can order the AI model to do what you want.
...
First, an attacker hides a malicious prompt in a message in an email that an AI-powered virtual assistant opens. The attacker’s prompt asks the virtual assistant to send the attacker the victim’s contact list or emails, or to spread the attack to every person in the recipient’s contact list. Unlike the spam and scam emails of today, where people have to be tricked into clicking on links, these new kinds of attacks will be invisible to the human eye and automated.

This is a recipe for disaster if the virtual assistant has access to sensitive information, such as banking or health data. The ability to change how the AI-powered virtual assistant behaves means people could be tricked into approving transactions that look close enough to the real thing, but are actually planted by an attacker.

Personal aside: I mentioned earlier that I recently edited a white paper on generative AI in consumer products. The article above links to it. It's worth the modest price tag to read volume 1.

[Kurth]

You have a remarkable clarity to know what a technology will be capable of 10-20 years down the road, when looking back half that far would confirm that we have no earthly idea what will happen over the next decade-plus in this space. You keep coming back to the fact that it's not replacing teams of programers today as though that's news to anyone paying more than peripheral attention.

I just know when I smell bullshit. This smells like bullshit.

The giveaway is the breathless evangelism and fighting over which superlatives / buzzwords to throw at it. The same people time and again. Certainly THIS time they might be right. We'll see.

I could not be pulling any harder for you to be right on this one. But I'm 100% convinced that in the next 18 months, you will look back on this and determine that your bullshit detector needs adjustment.

[Zaxxon]

After watching the Flappy Bird demo yesterday, I decided to see whether I could pick a random project and have similar success. With ChatGPT4's help, I wrote a simple web app last night. It takes a user-provided location, queries OpenWeatherMap to get local weather details, then sends the location and current conditions as a prompt to OpenAI/GPT-3, asking for poetry describing the current situation. Then it displays the poem along with the actual weather conditions.

I've not written JavaScript in decades, and have never used Node.js. This process took me around an hour from first prompt to working prototype. From accurately producing workable code, to explaining what it was doing, to walking through installation of the required components, to debugging issues I hit (both with my code and with the API keys--example: OpenWeatherMap will tell you that your key is active, but it'll take 10-30 min to actually go active), to altering the code to fit changing requirements I fed it, it did a remarkable job. One of the most-impressive parts to me was that it maintained a grasp of context throughout the chat, despite the program being made of a half-dozen parts (client-side and server-side JavaScript files, a node.js installation and config file, html and CSS files). Would RM9 belly-laugh at the code? Yeah. Me from 20 years ago would, too. It's by no means a product I'd put on the open internet as it stands. But it's a start that would have taken me days to weeks working at in my free time to come up with on my own, given my heavy rust and lack of experience with the tech choices used. One hour!

This is already indisputably life-changing tech for anyone motivated to use it. If I was a young kid who wanted to build something, this is like Khan Academy on steroids. I'll stick to the 10-20 year guess to replacing teams of programmers, but we're today days away from it being crazy useful. Overhyped? By some, sure. But this isn't a thing that existed before, and now it does. Look beyond the fact that it has blind spots and try to focus on what it is doing.

[YellowKing]

My buddy who uses it was like, "OUR WORLD IS ABOUT TO CHANGE" and I think that's the kind of stuff Rmn9 cringes at. I mean yay it helped you somewhat write a D&D module, let's not get carried away here. But certainly it has potential, certainly it will get better, and it will likely change life in ways we can't predict yet.

I'm conservatively optimistic. I don't buy into the doom and gloom scenarios. However, I don't think it's nonsense.

[LawBeefaroni]

I remember when GUI/WYSIWYG HTML editors meant that everyone and their grandmother would be a web developer.

Can, will, and should are all very different.


I fall on the RM9 side of things - that the existential threat is overblown. However, I do think the current generation of rudimentary AI will still cause large economic disruptions in many fields.

[noxiousdog]

Here's my test for today:
Name the Houston Astro that fits the clues "wildcat, 7th heaven, fresh on the scene"

Bard guessed wrong by saying Jeremy Pena went to university of Texas Arlington because their mascot is a maverick.

1) he went to the university of Maine
2) bard also said a maverick is a type of wildcat

Chatgpt thought it was Chas McCormick since he wears #7 and played for the Houston cougars

1) #7 is retired by the team due to Craig Biggio
2) he didn't play for the cougars. He played for millersville university


I'm skeptical that these fundamental errors can be fixed with enough precision to allow ai to code advanced business logic in a way that can be trusted any time soon.

[Isgrimnur]

It will function as a force multiplier for programmers. It will dilute the value of the position, lower the growth rate of the career field, and short-sighted managers will, once again, use it as a measure to cut costs and overwork their staff.

[Smoove_B]

I was going to put this the general gaming area, but since y'all have already talked about it here, I guess it makes more sense. An article from last week about a guy that used ChatGPT to play D&D with his kids. The article is a transcription of their session. What's interesting to me is that they didn't use the AI to do die rolling - instead they told the AI what their rolled results were and let the AI create the narrative. Some of it is very "clinical" - like the AI is reading right from the DM manual, but other stuff is surprisingly creative. Here's a snippet what the AI spit out after being told what a thief character rolled for various skill attempts:

Here’s the outcome of Tenzin’s attempted theft:

Tenzin, with a sense of daring and excitement, carefully approaches the Giant Riding Lizard’s enclosure. He remains hidden in the shadows, moving silently and gracefully (Stealth check: 18). The stable owner and other bystanders remain unaware of his intentions.

As Tenzin reaches the Giant Riding Lizard, he attempts to gain its trust and calm the creature (Animal Handling check: 8). Unfortunately, his efforts fall short, and the lizard becomes agitated, hissing and snapping at him. This draws the attention of the stable owner and nearby bystanders.

Realizing that his plan has been compromised, Tenzin quickly shifts gears and uses his Sleight of Hand skills to pick the lock on a nearby Riding Elk’s enclosure (Sleight of Hand check: 21). He succeeds in unlocking the gate, and the startled elk bursts out, creating chaos in the stable.

Seizing the opportunity, Tenzin uses the confusion to slip away from the scene and regroup with the party. While he didn’t manage to steal the Giant Riding Lizard, he did escape without being apprehended.

The party now faces a decision: continue shopping at the market, where Tenzin’s actions may have drawn unwanted attention, or depart for the Wildthorn Forest and put some distance between themselves and the city.

[Blackhawk]

It let him pick locks with Sleight of Hand...

[Smoove_B]

Creative!

[jztemple2]

I didn't see this posted (I think), thought it would be interesting:

ChatGPT is 'so wildly incorrect' that an Australian whistleblower is suing it for defamation

We all know ChatGPT gets stuff wrong. As in very wrong, and all the time. While that can be amusing, it's less funny if ChatGPT is mistakenly identifies you as a criminal. And it's less funny still if you were in fact the person who originally uncovered the crime in question.

Indeed, you might find it so unfunny, you decide to sue for defamation. Which is exactly what Brian Hood, a Melbourne Australia-based politician is doing.

[RunningMn9]

My buddy who uses it was like, "OUR WORLD IS ABOUT TO CHANGE" and I think that's the kind of stuff Rmn9 cringes at. I mean yay it helped you somewhat write a D&D module, let's not get carried away here. But certainly it has potential, certainly it will get better, and it will likely change life in ways we can't predict yet.

I'm conservatively optimistic. I don't buy into the doom and gloom scenarios. However, I don't think it's nonsense.

I mean, if they move past the “it’s wrong a lot” phase, I certainly think there’s a chance that this is akin to the internet or the arrival of the PC. But those things were pretty dramatic productivity gains. But has my live fundamentally changed? No. I still have to get up and drive to work. I still have to shop (like, I don’t always have to drive now, but that’s not a fundamental change to my existence).

But I think this will be less transformative than powered flight, or a world war.

[Isgrimnur]

I didn't see this posted (I think), thought it would be interesting:

ChatGPT is 'so wildly incorrect' that an Australian whistleblower is suing it for defamation

We all know ChatGPT gets stuff wrong. As in very wrong, and all the time. While that can be amusing, it's less funny if ChatGPT is mistakenly identifies you as a criminal. And it's less funny still if you were in fact the person who originally uncovered the crime in question.

Indeed, you might find it so unfunny, you decide to sue for defamation. Which is exactly what Brian Hood, a Melbourne Australia-based politician is doing.

One wonders if Australian defamation law also requires actual malice for defamation of a public figure as it does in the US.

[Kraken]

The Manhattan Project theory of generative AI

Is it a valid comparison?

Nukes could literally wipe out most of humanity in minutes, but relatively few people can get their hands on one. With generative AI, on the other hand, pretty much everyone will be able to use it, but it cannot wipe out most of humanity at a stroke.

True, but too clever.

Still, the Manhattan Project analogy does feel right to me in one respect: There is a world before mass access to generative AI and a world after it, and they are not the same.

I agree with that much.

[Max Peck]

Do we need a moratorium on further AI deployments? Elon Musk and a bunch of other people who ought to know think so.

I don't know about the "others" but I'll bet all my dogecoin that Musk actually just wants everyone else to take a breather so that he can catch up.

Elon Musk reportedly purchases thousands of GPUs for generative AI project at Twitter

Despite recently calling for a six-month pause in the development of powerful AI models, Twitter CEO Elon Musk recently purchased roughly 10,000 GPUs for a generative AI project within Twitter, reports Business Insider, citing people familiar with the company. The exact nature of the project, however, is still a mystery.

GPUs, or graphics processing units, are purpose-built chips originally designed for computer graphics, but their massively parallel designs make them ideal for doing generative AI processing as well. Training (creating) a new AI model usually requires a large amount of computing power, including many GPUs, which means that Musk's acquisition could represent a significant commitment toward developing a deep-learning AI model within Twitter.

In late February, The Information broke news that Musk had approached AI researchers to form a new AI lab to compete with OpenAI's ChatGPT, including former DeepMind researchers Igor Babuschkin and Manuel Kroiss. Earlier, Musk had publicly complained about bias in OpenAI's products, saying they were too "woke." Musk co-founded OpenAI in 2015 but left the company after an internal disagreement in 2018.

As for the Twitter-based AI project, Business Insider reports that it's a large language model (LLM), the type of generative AI tech that powers ChatGPT. The firm could potentially utilize its massive library of user tweets to help train the model for natural language output.

[coopasonic]

Dear god, a training library consisting only of tweets. We're doomed.

[jztemple2]

Dear god, a training library consisting only of tweets. We're doomed.

You just now figured that out?

I've always figured it would be a close race between nuclear war, global warming, rioting in the streets and Colossus: The Forbin Project

[Unagi]

Dear god, a training library consisting only of tweets. We're doomed.

You just now figured that out?

I've always figured it would be a close race between nuclear war, global warming, rioting in the streets and Colossus: The Forbin Project

Whoever would have dreamed of a 4-way tie?

[Max Peck]

Giant Space Rock is still in the running. Especially if an AI is involved in the effort to deflect it.

[YellowKing]

Turning aside from how smart AI actually is for a moment, a good example of what I'm looking forward to over the next few years is Spotify's personal DJ feature. For those who haven't tried it, it's a virtual AI-driven DJ that knows your name and spins batches of tracks based on your listening habits along with new stuff that fits in the same lane.

The cool part about it is that it gives the illusion of talking directly to you - it calls you by name, it can speak the names of artists and tracks it's playing, and it gives reasons for its selections (for instance, it told me it was going to play some stuff I may have listened to when I was younger, then proceeded to launch into a 90s indie rock batch).

What it's doing is relatively simple under the covers - it's essentially just an AI natural language overlay on Spotify's already pretty clever playlist algorithms. But I can imagine a time in the near future where maybe I could hold full conversations with this thing.

Spotify: "Here's some Kyuss"
Me: "Didn't someone from that band go on to form another band?"
Spotify: "Yeah, you're thinking of Josh Homme who formed Queens of the Stone Age. Do you want me to play some of that next?"
Me: "Yeah that would be awesome, but don't play anything off their latest album."

etc.

In short, I'm looking forward to the point where the language feels natural enough that I don't feel like I'm changing my speech in order to make a machine understand. When I ask my Alexa something, I'm keenly aware that I'm asking my Alexa something. I'm choosing specific phrasing, and essentially acting like my Echo is a hard of hearing foreign exchange student. Right now the Spotify virtual DJ gives a glimpse of that, albeit it's one-sided and non-interactive. But it feels natural talking to me, and when I can interact with that natural voice, look out - things are going to be super cool.

[hepcat]

And so it begins.

One of the practice area leads in our company now believes he can use AI to organize data from a collection of spreadsheets issued by an energy industry analysis service, and then create dashboards and ad hoc reports off that. So he sends me a list of companies that advertise as being "AI driven data analysis". The first meeting was yesterday after I set something up with one of their sales rep.

20 minutes into the meeting, the rep had to admit that they don't actually have AI integrated into the system and they're essentially a dumbed down version of Power BI (I added that last thing, but that's what she was basically admitting to).

It's like the 80s again, where a ton of companies started adding ".com" to their business name and then seeing their stock go up. Even if they weren't even a goddamn IT company.

Sigh....

[Max Peck]

Generative AI comes to Amazon Web Services

On Thursday, Amazon released a new suite of AI technologies, including foundational large language models (LLMs) called Titan and a cloud computing service called Bedrock, reports Reuters. The move comes as competitors Microsoft and Google integrate AI chatbots into their search engines and cloud operations.

Also on Thursday, Amazon announced a preview of Amazon CodeWhisperer, an AI-powered coding assistant similar to GitHub Copilot and Replit Ghostwriter. It's free for individual use and available for evaluation today.

[Max Peck]

“A really big deal”—Dolly is a free, open source, ChatGPT-style AI model

On Wednesday, Databricks released Dolly 2.0, reportedly the first open source, instruction-following large language model (LLM) for commercial use that's been fine-tuned on a human-generated data set. It could serve as a compelling starting point for homebrew ChatGPT competitors.

Databricks is an American enterprise software company founded in 2013 by the creators of Apache Spark. They provide a web-based platform for working with Spark for big data and machine learning. By releasing Dolly, Databricks hopes to allow organizations to create and customize LLMs "without paying for API access or sharing data with third parties," according to the Dolly launch blog post.

[Zaxxon]

Nice interview of Max Tegmark on Lex Fridman's show yesterday. Long, but can be easily viewed at 1.25x. Covers the open letter from earlier this month and many other AI and AI-adjacent topics.

[Kurth]

Turning aside from how smart AI actually is for a moment, a good example of what I'm looking forward to over the next few years is Spotify's personal DJ feature. For those who haven't tried it, it's a virtual AI-driven DJ that knows your name and spins batches of tracks based on your listening habits along with new stuff that fits in the same lane.

The cool part about it is that it gives the illusion of talking directly to you - it calls you by name, it can speak the names of artists and tracks it's playing, and it gives reasons for its selections (for instance, it told me it was going to play some stuff I may have listened to when I was younger, then proceeded to launch into a 90s indie rock batch).

What it's doing is relatively simple under the covers - it's essentially just an AI natural language overlay on Spotify's already pretty clever playlist algorithms. But I can imagine a time in the near future where maybe I could hold full conversations with this thing.

Spotify: "Here's some Kyuss"
Me: "Didn't someone from that band go on to form another band?"
Spotify: "Yeah, you're thinking of Josh Homme who formed Queens of the Stone Age. Do you want me to play some of that next?"
Me: "Yeah that would be awesome, but don't play anything off their latest album."

etc.

In short, I'm looking forward to the point where the language feels natural enough that I don't feel like I'm changing my speech in order to make a machine understand. When I ask my Alexa something, I'm keenly aware that I'm asking my Alexa something. I'm choosing specific phrasing, and essentially acting like my Echo is a hard of hearing foreign exchange student. Right now the Spotify virtual DJ gives a glimpse of that, albeit it's one-sided and non-interactive. But it feels natural talking to me, and when I can interact with that natural voice, look out - things are going to be super cool.

It’s cute you think you’re still going to be listening to music made by actual musicians and not songs created by AI in the style of the bands you used to enjoy.

Also, Lex’s latest (posted by Zaxxon above) with Max Tegmark is excellent. Max and Eliezer appear to be very much on the same page. I just don’t see the “pause” happening. Too much money involved. I also think Max’s take on “Moloch” driving this is a little off. He asks at one point, “Why are we doing this? We need to stop and ask ourselves, why are we racing forward to create AGI?” Max puts the blame on “Moloch,” his interpretation of the selfish, competitive, zero-sum game impulses that tend to push humans to bad outcomes. It seems to me it’s simpler than that: Humans just aren’t smart enough to do all the things we like to imagine our best selves doing (curing cancer, addressing climate issues, exploring the galaxy, etc., etc.). Or, more accurately, we’re not patient enough to wait for our just better than primate brains to figure out these problems. So we’re creating an intelligence so superiors to ours so we can use it to solve these problems for us. It’s just that in our hubris, we believe we’ll be controlling it and not the other way around.

[Kurth]

Speaking of AI and music, the NYT ran a piece this morning on the new Drake/Weekend song, “Heart on My Sleeve.”

Spoiler: It’s not a Drake/Weekend song. It’s entirely generated by AI. But that didn’t stop it from spreading quickly and going viral on streaming sites, where most fans couldn’t tell the difference. Crazy!

Here’s another great example of what AI is doing to music:



This is today. Not years or months or weeks away.

[jztemple2]

On the other hand, maybe Taylor Swift might have something to worry about... We Asked AI to Write a Taylor Swift Song About Dating an F1 Driver

Celebrity relationships are pretty common for Formula 1 drivers, seeing as they themselves are global superstar athletes. But the latest rumors coming from Spanish media about Fernando Alonso’s dating habits have been supremely delectable. Purportedly, Alonso and Taylor Swift (yes, that one) have become something of an item.

But beyond some normal internet chatter, there hasn’t been much substantiation to the rumor. Personally, though, I could see it happening. Alonso’s eyebrows might have a special superpower that could make this possible. With the man’s 2023 season looking rather strong so far, success begets success. One of the prevailing (and definitely most likely false) rumors from this little shot of gossip is that Swift wrong the song "Style" about Alonso. Though I’m not sure that Alonso has “that James Dean daydream look in [his] eye.” Maybe for some avid Aston Martin fans.

Alonso can also come off as an odd choice given the fact that pre-rumors, the one man on the grid one would deem Most Likely to Date Taylor Swift would probably be George Russell, given the singer's noted affinity for "London Boys." Although as Staff Writer James Gilboy wisely points out: "Honestly, I think Alonso is a good pick. He's shown he puts everything into relationships long after they obviously aren't working out."

Either way, this got Reviews Editor and self-proclaimed Swiftie Chris Tsui thinking: what would a Taylor Swift song about dating an F1 driver look like? You have our deepest apologies for reporting this, reader, but this is what ChatGPT thinks it would look like:

Those interested may got to the article to see the lyrics, but warning, you can never un-see them

[Victoria Raverna]

Now there is the Auto-GPT. Not sure how it really works but you give it some goals and it'll do research using internet to achieve it.

You can install Auto-GPT on your PC but it'll still need to connect to OpenAI and Pinecone for it to works. It is using Chat-GPT at OpenAI to work. It uses Pinecone as database.

[hepcat]

[jztemple2]

I posted this over on the video games forum under the appropriate game thread, but I thought folks over here might find it of interest:

Really interesting article, GalCiv IV: Supernova Dev Journal #13 - AlienGPT

GPT stands for “Generative Pre-trained Transformer” and since the release of OpenAI’s ChatGPT the world of generative AI has really taken off. For the past several months, we have been working to build GPT technology into our Tachyon engine, Stardock’s cloud-based metagaming technology that has previously been seen in Ashes of the Singularity, Offworld Trading Company and Star Control: Origins. Thanks to OpenAI’s ChatGPT API and Tachyon, we are proud to demonstrate what this technology can do for games.

AI generated civilizations
In the past, Stardock would release new civilizations for $4.99 for 5 as a DLC. Creating a new civilization was a lot of work. Lots of writing and lots of new art and game balance. With Galactic Civilizations IV: Supernova, it’s just a matter of entering a description of your civilization and pressing a button:

Step 1: Enter a description for your civilization.



Step 2: Press the button.



There is no step 3.