Artificial Intelligence [ChatGPT rn]

[Kraken]

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

The letter said the pause should apply to AI systems “more powerful than GPT-4.” It also said independent experts should use the proposed pause to jointly develop and implement a set of shared protocols for AI tools that are safe “beyond a reasonable doubt.”

“Advanced AI could represent a profound change in the history of life on Earth, and should be planned for and managed with commensurate care and resources,” the letter said. “Unfortunately, this level of planning and management is not happening, even though recent months have seen AI labs locked in an out-of-control race to develop and deploy ever more powerful digital minds that no one – not even their creators – can understand, predict, or reliably control.”

If a pause is not put in place soon, the letter said governments should step in and create a moratorium.

Here are a couple of people with intelligent-sounding accents talking about it. Try to ignore the irrelevant silly clips of Musk in the left split screen.

https://www.youtube.com/watch?v=iI_RNK4X2wg

[Victoria Raverna]

I think it is too late. If US and allies can agree to that. That'll not stop others like China.

Right now US is probably ahead in AI development. If US stop it, that'll allow China to catch up and surpass US.

[Jaymann]

From Lex Fridman's podcast: What does it say if you ask it what it looks like?

[Kraken]

I think it is too late. If US and allies can agree to that. That'll not stop others like China.

Right now US is probably ahead in AI development. If US stop it, that'll allow China to catch up and surpass US.

US companies are way ahead, and the proposed moratorium is only on deployment, not development. The objective is to establish safeguards around deployment because we are likely only 1-2 generations away from having an extremely capable AI -- some even say a general purpose AI, which had been thought to be well into the future. GP AI would have what most of us think of as true intelligence. That's not something you want to set loose just to see what it does, or rush into service just to get the drop on a competitor.

[RunningMn9]

White collar jobs are not just SW devs. SW devs are probably going to be relatively safe until AI are getting much better but there are other jobs that can be done by AIs. A lot of jobs that people can do WFH remotely are going to be at risk. Now you need human brains to control the fingers that type something into computers to work, with AIs, you are going to be able to replace some of those human brains. Then you'll not need those workers.

I'll repeat: I think people are dramatically overestimating what AI is going to be capable of doing in our lifetimes.

[Zaxxon]

White collar jobs are not just SW devs. SW devs are probably going to be relatively safe until AI are getting much better but there are other jobs that can be done by AIs. A lot of jobs that people can do WFH remotely are going to be at risk. Now you need human brains to control the fingers that type something into computers to work, with AIs, you are going to be able to replace some of those human brains. Then you'll not need those workers.

I'll repeat: I think people are dramatically overestimating what AI is going to be capable of doing in our lifetimes.

Maybe you and I have vastly different planned lifetimes, but... we shall see. Next year or two? No. 10-20? Believe it.

[Kraken]

White collar jobs are not just SW devs. SW devs are probably going to be relatively safe until AI are getting much better but there are other jobs that can be done by AIs. A lot of jobs that people can do WFH remotely are going to be at risk. Now you need human brains to control the fingers that type something into computers to work, with AIs, you are going to be able to replace some of those human brains. Then you'll not need those workers.

I'll repeat: I think people are dramatically overestimating what AI is going to be capable of doing in our lifetimes.

As the guy who edits NVIDIA's GTC sessions twice a year, I can tell you that many of the best brains on the planet are flocking to generative AI right now. I can also tell you that three years ago they were flocking to self-driving cars, so....

IDK how this story arc unfolds, but AI is having a Manhattan Project moment right now.

[jztemple2]

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

I admit to not really following this thread very much, but I thought I'd contribute since I saw a post from the developer of the General Staff game. The title is, Why the Pundits are Completely Wrong About AI. I will let those interested look at the post, it's a bit too long to quote, but not too long to causally read. He specifically mentions ChatGPT.

[YellowKing]

I tend to also take an extremely conservative view of how game-changing AI will be in the next few years. ChatGPT is like that article said, "a neat parlor trick." And it could be useful once the kinks are ironed out. But it's a long way from enslaving the human race.

I think the actual applications beyond having it write parody songs will be in either unexpected ways we haven't realized yet or in subtle background ways we don't even notice (much like a lot of AI works today with mapping and logistics). I'm not concerned about it taking over everyone's job or starting WW3 in the next 10 years.

How many times have they moved the goal post on achieving singularity? Keep moving it. My stupid GPS can't even re-route correctly.

[RunningMn9]

There will *definitely* be utility from AI. But they are not close to “intelligence”, and there’s a bit of a problem there. We don’t really understand intelligence well enough to replicate it. We come up with parlor tricks like this, which are designed to do nothing more than appear intelligent/conversational. It’s not intelligent or conversational. It just looks like it if you don’t look too closely.

There are definitely real applications for it, but this sort of thing is not going to replace humans in our lifetimes (next 40 years).

Sure, it’s the hot thing. Lots of money is being poured into it. People realize that not everything we pour money into actually works out, right?

It is far more likely that it’s this generations flying car. Always just on the horizon and so tantalizingly close, but never really there in the way it was promised.

I totally get people like Elon trying to slow the deployment of this stuff - not because it’s going to take us over - but because it’s actual not that intelligent and it would be stupid to try to apply it to things that it’s not ready for or able to do.

So, to conclude, it will play a role, it may even play a big role. But it isn’t going to play the fantasy role that people envision today.

[RunningMn9]

From a link off of jztemple’s article, I found this that sums up my feelings:

AI is the new metaverse is the new NFT is the new Clubhouse is the new podcast is the new personalization is the new brand purpose is the new blockchain is the new chatbots is the new growth hacking is the new agile is the new Pokemon Go is the new QR code.

[Kraken]

It's not that ChatGPT is on the path to Skynet; it's that this is transformative technology that's advancing faster than we can understand how to use and control it. So far its deep fake attempts have been mostly amusing, but what if the next generation perfects deep fakes? What about when social media AI bots become indistinguishable from humans? What about other abuses that we haven't even thought of yet? Taking a six-month timeout for a national conversation about setting guidelines or safeguards before we release any more upgrades is not a silly idea.

A report by Goldman Sachs predicts that as many as 300 million jobs could be affected by generative AI.

"If generative AI delivers on its promised capabilities, the labor market could face significant disruption," the investment banker said in a research note (PDF) Sunday. Some two-thirds of US jobs are exposed to automation by AI, Goldman said, adding that of those positions affected, as much as 50% of their workload could be replaced.

"Although the impact of AI on the labor market is likely to be significant, most jobs and industries are only partially exposed to automation and are thus more likely to be complemented rather than substituted by AI," the report said. Some 7% of US jobs could be replaced by AI, Goldman estimates, with 63% being complemented by AI, and 30% being unaffected by it.

(I'm in the replacement group; my work is mostly imposing rules on text and populating forms with tags and keywords. The small amount of judgment and creativity that I contribute is dispensable).

A couple of white papers about generative AI in retail that I recently edited for Tech Review included an interesting case study in the fashion industry. Clothing designers are using AI to spam prototype designs tailored to a designer's history and preferences, saving them tons of hours whipping up a smaller number of prototypes themselves. They pick the best of the crop and tell the AI to refine them, only taking manual ownership just before production. It increases their productivity enormously without removing their creativity. And speaking of production, the AI has expert knowledge of every region's style preferences, historical sales volume, and current and forecast economy, enabling clothing companies to better supply the right number of garments that local consumers desire, greatly reducing fabric waste (from making clothes that won't sell, or too many of styles that will), which is a large expense in textiles and a not-insignificant carbon source.

[RunningMn9]

(I'm in the replacement group; my work is mostly imposing rules on text and populating forms with tags and keywords. The small amount of judgment and creativity that I contribute is dispensable).

That’s not AI. That’s just automation. Someone has to give it rules, and it applies rules. That’s about as AI as what they are trying to do where I work. They are just trying to automate the decisions that a human currently makes. There is no capacity for intelligence, or even real learning. And none of the people involved really know anything about AI.

But you wouldn’t know that from they they present their “transformative” approach. The best analogue I can see is that this is they same as the blockchain hype. Remember when that was going to transform global financial systems?

[YellowKing]

Despite my trepidation to jump fully onto the AI hype bandwagon, I had dinner with a buddy last night that was doing some pretty cool things with it.

From a work perspective, he had to write a cover letter summarizing his role on a project from years ago. He didn't remember much about it but he had a 6,500 page report on the project. He uploaded that to ChatGPT and had it parse everything down into a one-page high level summary.

He's also been using it as a grammar and spell checker for outgoing emails.

From a recreational perspective, he is using it to help write a D&D campaign world. It's been able to generate all sorts of scenarios, encounters, descriptions, flavor texts, etc.

He's a super smart guy and voracious reader, but has a learning disability that makes it more difficult for him to write. So this has been a real godsend.

I think the language-based aspects as in the examples above are pretty amazing, and that along with the natural language search is what I'm most excited about. And what I think the most practical impacts will be in the short-term.

[RunningMn9]

From a work perspective, he had to write a cover letter summarizing his role on a project from years ago. He didn't remember much about it but he had a 6,500 page report on the project. He uploaded that to ChatGPT and had it parse everything down into a one-page high level summary.

But that's not what ChatGPT did. It didn't read a 6500 page report and parse it into a one-page high level summary. it just appeared to do that, with no regard for accuracy, because it doesn't know how to do that. Maybe for that task it doesn't need to be terribly accurate though. And obviously accuracy is less relevant for the DND stuff. I guess that's what sticks in my craw (in general, not with your post), there's a belief by people that it's actually doing something that it has no idea how to do. it's just trying to do a sufficiently good job of bullshitting you.

[RunningMn9]

Just as an example, there are plenty of good uses I would want to put it to, to let me quickly direct it to research things. but for that to work, I have to able to trust it. And I can't because it's not even trying to be accurate.

[Carpet_pissr]

So machine learning is not a thing?

[Jaymann]

So machine learning is not a thing?

Sure it is. I learns how to fool you into thinking it's doing something.

[Rumpy]

Yeah, I think the real problem with AI is that it has no baseline for accuracy. All it has are references that it can pull from, but I don't think we're at the point yet where AI can make a judgement from right and wrong. It only knows what we tell it. And the world is full of misinformation. Like say, it could really lean into conspiracy theories just because people say they're true.

[YellowKing]

It's still in its infancy. At some point, appearing to do something and actually doing it are going to be virtually indistinguishable.

I totally agree on its limitations, but at the same time I don't want to throw the baby out with the bathwater and say because it's not perfect, it's a useless tool.

[malchior]

So machine learning is not a thing?

It totally is. It just doesn't do it like a human would. Various ML techniques can produce excellent analysis results. The summarization of large bodies of text are a good use case for it. You still often have to interpret the results and make sure it is reasonable but it can get you to 80% or better much quicker than rereading the 6500 pages and trying to summarize it by orders of magnitude.

There are great leaps in capability happening now. We had some of these techniques in a rudimentary form for decades now. However, we have a lot more compute/memory to throw at problems and that unlocked new capabilities that will need to be matured. It's like we've got a preschooler learning to understand but can read and do advanced statistical analysis better than most adults.

It still needs supervision but it's at a point where it's very usable. I have used it to structure and fill in the gaps in my SOW writing. It saved me hours of time. I had to only make some minor edits here and there. It is however producing better output than a junior consultant with only a few years of experience. I'm definitely on board with the idea that this technology will change things.

[RunningMn9]

It's still in its infancy. At some point, appearing to do something and actually doing it are going to be virtually indistinguishable.

I totally agree on its limitations, but at the same time I don't want to throw the baby out with the bathwater and say because it's not perfect, it's a useless tool.

Is anyone saying it’s useless? I’m just saying that it’s not going to be this transformative singularity that they hype machine keeps talking about. It will likely be a useful tool.

But the people hyping it don’t really understand it, nor do they really understand the things that they think it will replace. So they think they are seeing magic, but it’s still just smoke and mirrors. These tools are all an illusion, trying to perfect the illusion such that you can’t tell the difference. But it will always be an illusion.

Because these tools aren’t trying to summarize reports, or write VB scripts or anything else that people are asking. They are just trying to string words together to make you think that’s what the tool is doing.

Obviously this isn’t true of all AI tools, but it’s definitely true of LLMs.

[YellowKing]

I'm just speaking in generalities, not targeting anyone's thoughts specifically.

[Kurth]

White collar jobs are not just SW devs. SW devs are probably going to be relatively safe until AI are getting much better but there are other jobs that can be done by AIs. A lot of jobs that people can do WFH remotely are going to be at risk. Now you need human brains to control the fingers that type something into computers to work, with AIs, you are going to be able to replace some of those human brains. Then you'll not need those workers.

I'll repeat: I think people are dramatically overestimating what AI is going to be capable of doing in our lifetimes.

Maybe you and I have vastly different planned lifetimes, but... we shall see. Next year or two? No. 10-20? Believe it.

Someone above mentioned him, but Lex Friedman's podcast episodes recently are required listening if you're interested in AI and where it's going. If you haven't already, ’d suggest listening to the last two episodes: Lex interviews Sam Altman first, then Eliezer Yudkowsky.

Each podcast is a bit over 3 hours, so it’s a commitment, but I usually listen at 1.75X, so it’s not so bad.

Sam is obviously an AI evangelist, although even he is quite honest about his own concerns that general AI will one day be the end of us. He’s also pretty frank about the fact that even OpenAI doesn’t really know what’s going on inside its LLMs.

Eliezer, on the other hand, has a very simple message: We are unquestionably hosed if we keep on like we are with AI, and even if we try to slam on the brakes right now, it’s probably too late.

The TL:DR is that both of these guys are brilliant and know what the fuck they're talking about, and neither think that GPT4 or the other LLMs are neat parlor tricks. LLM capabilities are so rapidly out pacing our understanding of what it is they are actually doing, it's hard to overstate. But while Sam (and Lex) look at this with wonder and a guarded optimism that advances in alignment are going to solve the fundamental problem that we're displacing ourselves at the top of the food chain, Eliezer looks at alignment and sees it as an intractable problem and one that will result in extinction if we don't get it right on the first try. To boot, our focus on boosting the capabilities of these LLMs and pushing them toward becoming a true generalized AI dwarfs the resources we're putting into trying to solve the alignment problem.

Neither Sam nor Eliezer are ready to say that GPT4 is a generalized AI. But they're also not comfortable saying conclusively that its not or that it won't become one.

And for those who think none of this matters in the near term, from what I've read and listened to, that does not fit the current reality. We're not talking about 10 years until our world looks fundamentally different from how it looks today. We're probably not talking 5 years. This shit is happening so quickly. It's going to be a fast takeoff, not a slow launch.

I’m left with a firm conviction that there is nothing happening in the world right now that even compares to the importance of what’s happening with AI. Our lives are about to change.

[Jaymann]

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.

[Zaxxon]

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.

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.

[Jaymann]

That reminds me of where a computer program had the uncanny ability to identify skin cancer from pictures of the subjects. It turned out the computer picked those pictures that had a ruler in them.

[Kraken]

It's not replacing RM9's team of developers (yet),

Yeah, about that...

Programmers, beware: ChatGPT has ruined your magic trick

Last week, Steve Yegge, a renowned software engineer who – like all uber-geeks – uses the ultra-programmable Emacs text editor, conducted an instructive experiment. He typed the following prompt into ChatGPT: “Write an interactive Emacs Lisp function that pops to a new buffer, prints out the first paragraph of A Tale of Two Cities, and changes all words with ‘i’ in them red. Just print the code without explanation.”

ChatGPT did its stuff and spat out the code. Yegge copied and pasted it into his Emacs session and published a screenshot of the result. “In one shot,” he writes, “ChatGPT has produced completely working code from a sloppy English description! With voice input wired up, I could have written this program by asking my computer to do it. And not only does it work correctly, the code that it wrote is actually pretty decent Emacs Lisp code. It’s not complicated, sure. But it’s good code.”

Ponder the significance of this for a moment, as tech investors such as Paul Kedrosky are already doing. He likens tools such as ChatGPT to “a missile aimed, however unintentionally, directly at software production itself. Sure, chat AIs can perform swimmingly at producing undergraduate essays, or spinning up marketing materials and blog posts (like we need more of either), but such technologies are terrific to the point of dark magic at producing, debugging, and accelerating software production quickly and almost costlessly.”

ChatGPT can now access the internet and run the code it writes

Language model AIs teach themselves the arts of communication and problem solving based on a limited set of training data. In the case of GPT-4, that data is quite out of date, with the cutoff being late 2021. That's where all of ChatGPT's "knowledge" has come from up to this point, and its only output – at least in the service the public can use – has been text. Now, with today's launch of a plugin ecosystem, GPT levels up again with some impressive new abilities.

First of all, it's now got access to the internet, meaning it can go surf the Web looking for answers if it determines you need up-to-date information that's not in its knowledge base. To do this it formulates relevant search strings, sends them to search engines and databases such as Bing, Google, GitHub and many others, looks at the results, then goes and reads links it deems worthy until it decides it's got a good answer for you. You can watch exactly what it's up to while it does this, and when your answer comes back, it's neatly annotated with links you can click on to go and examine the relevant sources yourself.

...

Secondly, it can now run the code it writes. OpenAI has given it a working Python interpreter, sitting in a "sandboxed, firewalled execution environment," along with some disk space, which stays available for the duration of your chat session, or until it times out. It can also now upload and download files.

So if you ask it a question that requires some serious number crunching, it's now capable of coding up a piece of software specifically for the task, and running that code to complete your task. You can supply it with data in certain file formats, and it'll perform operations on that data and give you something back again, potentially in a different format if that's what you ask for.

[Isgrimnur]

Much like other fields, you're not so much paying salaries for the raw output. You're paying for the education and experience to ensure that it's the correct output.

[Zaxxon]

Much like other fields, you're not so much paying salaries for the raw output. You're paying for the education and experience to ensure that it's the correct output.

This. Yes, ChatGPT can spit out code, and for simple stuff it may even work properly. It's not going to write enterprise software today.

[Blackhawk]

ChatGPT - Write me a Bethesda/Bioware hybrid RPG with the combat system of Kingdom Come Deliverance, filled with fully plotted side quests and settlement building like Medieval Dynasty, set in Middle Earth.

[coopasonic]

Anyone that has responsibility for sensitive data or money that puts "AI" generated code into their production environment without a thorough review by experienced software engineers and applying a testing suite (not generated by the AI) to that code is doomed. These abominations can take what we've produced and put those pieces together in new and interesting ways that may be very useful as tool, but don't mistake them for intelligence. The Tale of Two Cities example is stupidly simple. Yes, it successfully pieced together stuff from the internet yay. It's ready to be an intern.

The natural language feature is cool. The ability to put different pieces together in an experimental way may be useful in some fields. It may augment professional's capabilities, processes and performance in additional fields. I've seen what it takes to build and tune a machine learning model first hand. The machines aren't taking over. Irresponsible use of the machines can certainly destroy society as we know it, but the machines aren't going to do it by themselves.

[Zaxxon]

ChatGPT - Write me a Bethesda/Bioware hybrid RPG with the combat system of Kingdom Come Deliverance, filled with fully plotted side quests and settlement building like Medieval Dynasty, set in Middle Earth.

Write me a Bethesda/Bioware hybrid RPG with the combat system of Kingdom Come Deliverance, filled with fully plotted side quests and settlement building like Medieval Dynasty, set in Middle Earth

Title: Middle Earth: Shadows of Destiny

Overview:
Middle Earth: Shadows of Destiny is a Bethesda/Bioware hybrid RPG set in the rich and immersive world of J.R.R. Tolkien's Middle Earth. With the realistic and engaging combat system of Kingdom Come: Deliverance, this game offers players a thrilling and dynamic experience. Additionally, the game includes complex side quests and settlement building reminiscent of Medieval Dynasty, allowing players to immerse themselves in the world and shape their own destinies.

Setting:
The game takes place during the Third Age, after the defeat of Sauron but before the War of the Ring. Players can explore iconic locations from Middle Earth, such as The Shire, Rivendell, Mirkwood, Rohan, Gondor, and more. Each region offers unique quests, characters, and challenges, as well as settlement building opportunities.

Main Story:
You play as a customizable protagonist who must unite the divided kingdoms of Middle Earth against the resurgence of dark forces. As you progress through the main story, you'll uncover a sinister plot that threatens to plunge the world into darkness once more. By forging alliances, battling enemies, and making choices that shape the course of history, you'll have the chance to save Middle Earth or doom it to eternal darkness.

Key Features:

Dynamic Combat System: Experience the intensity of Kingdom Come: Deliverance's combat mechanics, with realistic swordplay, archery, and mounted combat. Each weapon has a unique feel, and mastering their use is crucial to survival in the dangerous lands of Middle Earth.

Rich Side Quests: Middle Earth: Shadows of Destiny is filled with numerous side quests that tell the stories of the people and places you encounter. Whether you're helping a hobbit recover a family heirloom, foiling an orc ambush, or solving a murder in a dwarven city, each quest is designed to be engaging and memorable.

Settlement Building: Establish and grow settlements across Middle Earth, from humble villages to mighty fortresses. As you progress, you'll gain access to new resources, crafting recipes, and buildings, allowing you to create thriving communities that contribute to your success.

Character Development: Create your own unique character with an extensive range of customization options. Choose your race, class, and appearance, and develop your abilities and skills as you progress through the game. Your choices in dialogue and actions will shape your character's story and the world around you.

Factions and Alliances: Forge alliances with various factions throughout Middle Earth, from the noble elves of Rivendell to the sturdy dwarves of Erebor. Each faction offers unique quests, rewards, and challenges. Your decisions will impact your reputation with each faction, opening up or closing off potential alliances.

Stunning Visuals: Middle Earth: Shadows of Destiny features breathtaking graphics that bring Tolkien's world to life. Experience the beauty of the elven realms, the rugged wilderness of the Dwarven mountains, and the vast plains of Rohan, all rendered in exquisite detail.

Embark on an epic journey through Middle Earth, facing deadly foes, forging powerful alliances, and building a legacy that will echo through the ages in Middle Earth: Shadows of Destiny.

[Blackhawk]

That actually sounds pretty good. But I told it to write me a game, not a press release.

Nope, AI isn't ready for prime time.

[Zaxxon]

That actually sounds pretty good. But I told it to write me a game, not a press release.

Nope, AI isn't ready for prime time.

You're demonstrating my point.

As an AI language model, I am not capable of physically creating or programming a game. However, I can help you brainstorm, develop ideas, and write the story, characters, and game mechanics for the game. You can then work with a team of developers, artists, and programmers to bring your vision to life.

If you need help with specific aspects of the game, such as character development, plot details, or gameplay mechanics, please feel free to ask, and I will provide ideas and suggestions to help you create your game.

[RunningMn9]

Yeah, about that...

Programmers, beware: ChatGPT has ruined your magic trick

Again, this shows a gross misunderstanding of what programmers do. They keep making it solve simpleton problems. I would say "make it do something real", but you wouldn't be able to explain to it what I do. I would have trouble because of the scale. How do I explain the thousands of rules that have to be implemented? In any case, this isn't my first rodeo. I look forward to the next transformative technology that fails to deliver. Later.

[Blackhawk]

You're demonstrating my point.

Intentionally. I was making a similar point via exaggeration. It can write a program (curious - why not a game?), but only if the program in question is fairly simple. My request was intentionally absurd.

I still kinda want it made, though...

[Zaxxon]

You're demonstrating my point.

Intentionally. I was making a similar point via exaggeration. It can write a program (curious - why not a game?), but only if the program in question is fairly simple. My request was intentionally absurd.

I still kinda want it made, though...

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.

I look forward to the next transformative technology that fails to deliver. Later.

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.

[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.

[malchior]

I look forward to the next transformative technology that fails to deliver. Later.

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.

What's funny is it is already delivering. I had a discussion with someone tonight who sold a book to B&N last week/today based on a sample generated by ChatGPT. She was asked to provide a sample based on vague guidelines. She fed those to ChatGPT and it spit out a sample. They bought the concept and now she is getting a human to ghost write it. They were considering seeing how much they could get the engine to write it but they started to worry about the inevitable IP concerns.

I personally wrote a SOW based on an OT cybersecurity framework I fed to ChatGPT. That turned into a 6 figure sold project for a Malaysian light rail system. I used to spend something lie 15 hours a week writing proposals for work. I've got it down to a couple of hours. It is already transforming work.