How can we utilize Artificial Intelligence to help us be more prescient about quality of life and environment?

[Space Lynx]

I would like this thread to be about the future of AI (focus on positive stuff only please) and its possibilities for bettering flora/fauna in general.

I was thinking of what Jensen (Nvidia CEO) said a few times in the last year about how AI will be targeted towards certain industries and custom AI Hopper or Blackwell computers will be designed for very specific industries, so concentrated small AI systems is how I understand it. Doesn't need to be a giant LLM (but it is possible I misunderstand Jensen, this is a field I am only mildly interested in so I haven't been paying attention fully).

Some ideas I have been contemplating with this idea in mind are below... and since some progress has already been made with this idea, link below, it just got my mind churning:

AI Makes New Battery Ideas link

A collegiate body of all accredited Universities across the USA and those abroad that wish/volunteer to participate/get government approval, for a specific Blackwell AI setup to scrape all knowledge from their library databases, but no other databases (hopefully this would improve accuracy similar to medical AI), in return those Universities that participate will get to use this AI in their research programs? Then as it learns and grows with this data set, begin to ask it academic and/or engineering questions. I don't know if this would be considered a LLM AI and you could just simply ask it questions or not, would it need a LLM AI as the base, then like a 'giant expansion pack for a game' you do the academic data set on top of that? I have no idea how these concentrated AI models are intended to work, which is partially why I made this thread... but also just interested in other peoples ideas for the future.

Some of the questions we could ask this model... I am unsure... perhaps how can we do agriculture better and what is the best pathway towards doing that? Feed it some engineering sets and then ask it to make a better rocket than Starship or Falcon 9, but also being reusable. Maybe how to make better concrete too (I linked a couple paragraphs below). Maybe we could even ask what is the most efficient way to replant a farm or rainforest for the best symbiotic outcome of varied species working together (yes trees do share resources through roots) and rainforests need to have a variety of species to grow right, you can't just replant trees, so what is the best optimization of that? I'm sure there are a variety of academic papers on that, it could analyze those and cross analyze it with entire data sets of plant biology (the Native Americans knew about the three sisters method of farming, perhaps AI could expand on that for extremely optimal farming that saves the topsoil), etc. Then we could ask another AI model to program drones based on soil density/location (or a human gps mapmaker can begin to plot out locations of where the seeds go based on the new datasets from the AI) of certain regions to help us replant entire rainforests overnight, with AI robots making the drones as in sci-fi films (nothing can go wrong with this idea jokes incoming I know) lol

Trees Sharing Resources Link

Drones Planting Trees is nothing new

Do you think this level of AI is going to be possible with Blackwell AI setups?

Another concentrated model would in the medical industry for medicine R&D and other medical related things (from what I have read this is already well underway), Jensen mentioned Nurse Practitioner AI I believe linked below, AI could act as a filter system in the medical field perhaps too someday to reduce load on regular nurses, the filter says you need someone irl then that's path b, but if it solves your problem its documented and saved to your profile and you move on, I don't think it's their yet, but maybe closer than we think. With telehealth real doctors/nurses combined with offloading lower level medical stuff to AI, could really make the healthcare industry more affordable and less demanding on professionals. Thanks to Mark Cuban the billionaire for already making the most used medicines affordable with his Cost Plus Drugs company that takes no excess profits and keeping me optimistic about humanity. This is the leadership we need in more of the medical field.

Health Link 1

Health Link 2

Another thought I had recently too, could AI models be applied over existing super computers with a Blackwell setup sort of 'guiding' the super computers sheer power? In this case you could extrapolate it over ideas like this to ask it questions: Super Computer Idea

I guess what I am trying to ask and think of ideas if AI can do so many great things, how much is Blackwell going to change things?

Based on everything I have read, we are at a unique time in history, either AI is actually about to change everything for the better, or it's just a hype bubble. I really have no idea which to be honest. I guess if the robots Jensen was on stage with recently get any better… the possibilities are endless. Jensen already said Blackwell was made with the help of AI that came before it, Blackwell will continue this tradition in a possible snowball effect... it is possible the world is literally going to change over night. This snowball effect would apply to everything, including finding a battery that solves humanities problems once and for all, and so on and so forth. Less waste is something to be positive about.

My conclusion is that AI's greatest contribution is that it might make everything more efficient (obviously not in all fields, as we have seen with customer service models for various companies), as my examples above are talking about, I know there are negatives and fears, but this could also be a turning point in optimal efficiency in relation to our natural resources, but also our quality of life in general. I know people will want to respond to this thread about fears and job losses, this thread is not about that, there are several ways to mitigate the negatives, but this thread is not for that. Please focus on positive contributions AI can possibly make. I love thinking about this stuff just for fun, I hope some of you do too.

To put an asterisk on all of this, I have chatted with CoPilot with GPT-4 for quite a bit in the last six months. It is very impressive to me, that even when I call it out on being wrong on something, it expands its dataset in real time, apologies to me, says I am right, and then sites its sources to explain itself how it came to its original decision and why it now agrees with me. Maybe this isn't impressive, but to me it is; that being said it still gets a lot of things wrong so I am not super bullish on AI just yet. If this is all on Hopper Nvidia AI chips (I don't actually know what CoPilot uses just fyi)... and Blackwell is coming soon... and if how we as a species navigate the world around us with language... perhaps there is something interesting on the horizon indeed.

What are some of your ideas of how AI can better the world for everyone and everything?


@W1zzard and mods, if you feel this is the wrong forum for this, feel free to move it. I didn't know where to put it.

 

[Shrek]

I feel the timing is right as we are starting to get overwhelmed by complexity, just look at the various bugs in Windows and the updates that kept getting pushed. AI may be able to catch the issues better than us.

Stephen Wolfram recently added an article on AI doing science
Can AI Solve Science?—Stephen Wolfram Writings

 

[Vayra86]

I feel the timing is right as we are starting to get overwhelmed by complexity, just look at the various bugs in Windows and the updates that kept getting pushed. AI may be able to catch the issues better than us.

Stephen Wolfram recently added an article on AI doing science
Can AI Solve Science?—Stephen Wolfram Writings

Bit of a paradox, that.

If you let AI catch the issues you're going to start a vicious cycle of not understanding your own code. You're going to be completely out of touch and with that, out of control. Humans like control.

Similar things apply to science. We're not overwhelmed by complexity - we choose to overwhelm ourselves with complexity. There are always simple solutions. Streamlining/unifying is the key. Less = more is another. There are thousands of systems on this planet doing essentially the same thing, not even differently. Part of the complexity arises from having all these systems talk to each other. Crypto is a great example of a series of systems that are bound to be forced to work together but deny that they are. Every coin is its own world. And as a result, none of these coins succeed. Crypto's complexity is also its business model: information overload is a great way to extract money from idiots.

I'm a devout believer in the idea that if we want true progress, we need full understanding of things. AI is not doing that for us, its doing the opposite, and with that, we're being led on a goose chase. To what exactly? Do we even know? Do we still even have direction, vision?

It comes down the elementary math at school: if you can't explain how you got to your answer, the answer is meaningless. That principle applies everywhere. If we lose that principle, we're going to drift further and further away from understanding - all we will have left in due time, is information. No understanding. We'll learn how to write prompts. Ask questions. And that'll be it. I don't know about you, but I refuse to live in that world. You can already see where it leads to on social media. Tyranny of stupidity. 'But AI said its true'. Who are you to say its not?

 

[Shrek]

I know that the OP requested positive stuff only, but I feel the above negative contrast is essential to analysing the issue.

 

[dgianstefani]

Offline AI that "trains" on you. You correct it and it gives you tailored results.

With how good phone APUs are getting, and NPU even in desktop computers getting strong, not to mention RTX GPUs etc, local AI is not hard.

Opensource models you can customise and download, pair with vetted training data etc. Glass box design, no BS from shadowy companies that insert their own politics.

For medical AI I am extremely dubious, feels like a cope to deal with the competency crisis more than anything, and insuring it is a nightmare.

 

[Vayra86]

On the positive side, I think economically its a way to get quick wins and innovate faster. Commercially, AI is kryptonite of course because it presents itself as a black box of infinite possibilities. And I also won't deny that when used for very precisely defined, specific tasks, AI (for lack of a better word, but its really just something trained for a task, probably much closer to machine learning than it being 'intelligence') can outpace human performance. Its really just a better machine, like we always used machinery to advance.

But when AI enters the space of autonomy, decision making and morality, we're in the red zone. It has no place there and never will/should. We've already seen where autonomous driving has gone... exactly nowhere. Yes there are pilots. But humans can't accept this and they won't. And with that, they express deeply felt wisdom: this is going too far, this is a loss of control we should not have.

 

[dgianstefani]

On the positive side, I think economically its a way to get quick wins and innovate faster. Commercially, AI is kryptonite of course because it presents itself as a black box of infinite possibilities. And I also won't deny that when used for very precisely defined, specific tasks, AI (for lack of a better word, but its really just something trained for a task, probably much closer to machine learning than it being 'intelligence') can outpace human performance. Its really just a better machine, like we always used machinery to advance.

But when AI enters the space of autonomy, decision making and morality, we're in the red zone. It has no place there and never will/should.

What's that IBM quote from 1979?

A computer can never be held accountable, therefore a computer must never make a Management Decision.

As relevant today as it was then.

 

[cvaldes]

One thing of note is that the planet's ecology is very, very complex and even the best scientists in the world collectively don't know how it all works 100% of the time.

AI requires models based on existing work just like your knowledge is based on books, videos, etc. created by other human beings before you.

I don't think anyone can predict how close were are to solving the world's ecological problems whether it's using brainpower and conventionally powered computers or new AI technology. At some point, someone needs to write the initial models for that, likely by people who understand the data really, really well.

AI/ML is still very much in its infancy. It's improving with each passing year but image generation models still can't get basic details like hands right. Those will all eventually be solved in time maybe in five years. But certainly not next month.

There are autonomous vehicles on this planet's public roads in some places and they still figure out some simple stuff. In recent months, pranksters have halted self-driving taxis by putting traffic cones on the cars' hoods. Those sort of situations need to be written into the autonomous vehicle driving model. The news is full of various other mishaps with autonomous vehicles.

I see these vehicles in my own town (it is one place that is concentrated with autonomous vehicle testing). Statistically these cars are safer than your average teen driver but they are far from perfect. They are still only operated with limited parameters, with limited range and under limited driving conditions even after 10+ years of on-road testing.

Today's AI models function better when there are a very limited set of variables and don't have to deal with constantly changing playing field. Having AI write a term paper on George Washington is easier than having driving a 2-ton vehicle through your town to drop off a bag of chips and six-pack of soda at someone's house.

 

[Vario]

I think there is tremendous potential in using large datasets and machine learning to identify new anti-cancer drug compounds.

There is a lot of current research to identify new targeted covalent inhibitors (drugs that selectively target particular amino acids on a particular protein to inhibit its function). Some of that is designing scaffolds (the bulk of the organic molecule, benzene rings, linkers, etc) with a selective affinity for the binding pocket, or "warheads" (electrophilic reactive portion of drug) with selective affinity for the amino acid nucleophile, from extremely large computational datasets of organic compounds and proteins. AI is revolutionary for the pharmaceutical industry.

 

[idkidk]

I think that by 2030 ~ 2050 there will be mass deaths. It will be an excellent tool to have new slave force

 

[dgianstefani]

I think there is tremendous potential in using large datasets and machine learning to identify new anti-cancer drug compounds.

There is a lot of current research to identify new targeted covalent inhibitors (drugs that selectively target particular amino acids on a particular protein to inhibit its function). Some of that is designing scaffolds (the bulk of the organic molecule, benzene rings, linkers, etc) with a selective affinity for the binding pocket, or "warheads" (electrophilic reactive portion of drug) with selective affinity for the amino acid nucleophile, from extremely large computational datasets of organic compounds and proteins. AI is revolutionary for the pharmaceutical industry.

It takes away a lot of the financial burden, when you can rapidly prototype millions of compounds, only one needs to work, and it's quick and cheap (relatively).

Previously, researchers would have to identify likely compounds, a difficult and expensive task.

Now they can automate much of the process.

Of course, this won't change the deathgrip pharma has on licensing and getting government regulating body approvals on new drugs, so I doubt this will make drugs cheaper for the consumer, only for the developer. Revolving door of biotech/political governing body jobs will continue. A younger me would believe that AI would make it so that startup companies could more easily develop drugs, but I just don't see that happening, they either get bought up or locked behind lawsuits by big conglomerates who hold patents for everything. I'm guessing this is part of why the AI boom is so intense, first to legally own a molecule can sit on it, no need to even risk entering the market immediately.

Actually getting a new drug approved is also typically the most expensive part of the process, so nothing AI is doing will make that faster.

 

[Vario]

It takes away a lot of the financial burden, when you can rapidly prototype millions of compounds, only one needs to work, and it's quick and cheap (relatively).

Previously, researchers would have to identify likely compounds, a difficult and expensive task.

Now they can automate much of the process.

Of course, this won't change the deathgrip pharma has on licensing and getting government regulating body approvals on new drugs, so I doubt this will make drugs cheaper for the consumer, only for the developer. Revolving door of biotech/political governing body jobs will continue. A younger me would believe that AI would make it so that startup companies could more easily develop drugs, but I just don't see that happening, they either get bought up or locked behind lawsuits by big conglomerates who hold patents for everything. I'm guessing this is part of why the AI boom is so intense, first to legally own a molecule can sit on it, no need to even risk entering the market immediately.

Actually getting a new drug approved is also typically the most expensive part of the process, so nothing AI is doing will make that faster.

Yes, you can take a screening library of several million compounds and reduce it to 20 or so, then you use traditional biochemistry to evaluate those 20. Major cost and time savings. You don't need nearly as many physical laboratories and ML based methods are less computationally intensive than traditional computational methods (density functional theory, ab initio, semi-empirical). Allows for smaller, newer players to enter the space.

 

[Onasi]

It takes away a lot of the financial burden, when you can rapidly prototype millions of compounds, only one needs to work, and it's quick and cheap (relatively).

Previously, researchers would have to identify likely compounds, a difficult and expensive task.

Now they can automate much of the process.

To be fair, while this is true, it won’t actually matter practically in terms of getting more effective drugs or new drugs to market. The bottleneck in pharma isn’t the discovery and creation of new compounds. It’s the rigorous randomized blind testing that is necessary to ensure that the substance is actually a meaningful improvement on the existing drugs or, in terms of potential new ones, a provably working treatment with acceptable side effects (I say acceptable here because no true effective drug has none at all). This is something impossible to automate by its very nature.

Edit: And yes, I do know you mentioned approval already, just wanted to expand on it so that potential discussion would be made aware that it’s not just bureaucratic red tape that’s just paper. There is far more to it.

 

[Space Lynx]

I feel the timing is right as we are starting to get overwhelmed by complexity, just look at the various bugs in Windows and the updates that kept getting pushed. AI may be able to catch the issues better than us.

Stephen Wolfram recently added an article on AI doing science
Can AI Solve Science?—Stephen Wolfram Writings

I have never heard of him, I will give it a read thanks!

One thing of note is that the planet's ecology is very, very complex and even the best scientists in the world collectively don't know how it all works 100% of the time.

AI requires models based on existing work just like your knowledge is based on books, videos, etc. created by other human beings before you.

My only counters to this are some of the other ideas I mentioned, such as can a mini-Blackwell AI setup guide a super computer to come with new answers on certain topics, etc.

Also, I would argue that USA government branches have AI that the normal public sector isn't allowed to see, my guess is its already more advanced than we realize. I might just be dreaming here though, lol

 

[cvaldes]

Remember that human beings screwed up the environment.

It's human behavior that's hard to change even if we already know what's better.

Let's say the fancy AI ecology model said "use the compost bin and eat more beans in lieu of animal protein." Hell, half the people who currently use my complex's compost bin don't use it correctly. And that doesn't factor in all the people (most residents) who don't use the compost bin at all, they just toss their compostables into the trash bin or grind everything up in the food disposer and send to the water treatment plant.

And would you be willing to reduce the amount of meat you consume and replace several servings every week with beans?

Pretend someone creates an AI healthcare model which then says "exercise, quit smoking, go to bed early and get 8-9 hours of sleep every day." How many people are going to change their daily routine? What if it said for every two hours of videogame play, you should do one hour of physical exercise? You gonna follow the AI model's advice? Your doctor? Your spouse? Your mom?

 

[dgianstefani]

To be fair, while this is true, it won’t actually matter practically in terms of getting more effective drugs or new drugs to market. The bottleneck in pharma isn’t the discovery and creation of new compounds. It’s the rigorous randomized blind testing that is necessary to ensure that the substance is actually a meaningful improvement on the existing drugs or, in terms of potential new ones, a provably working treatment with acceptable side effects (I say acceptable here because no true effective drug has none at all). This is something impossible to automate by its very nature.

Edit: And yes, I do know you mentioned approval already, just wanted to expand on it so that potential discussion would be made aware that it’s not just bureaucratic red tape that’s just paper. There is far more to it.

In theory you are correct. The "rigorous and blind" testing is indeed necessary, the thing is, it's easy to fake or tell half truths, too, especially if you have some big names in your paper and can find previous references to support your claim (not hard). All you have to do is shift the focus of the study, do a rewrite to show a positive effect, or simply not publish, and repeat the study. Especially when huge amounts of capital have already been invested in the drug, for example if issues arise in phase 3 human trials, it's a bit late to say "oh well, lets just invest another $200 million". Saying no can be career ending.

Before I became too jaded to become a researcher, and before my current medicine degree, I spent a year working at the Swansea Uni biomedical data science building after my population health & medical science degree back in 2020. It was a paid internship, I got to directly help design studies, learn how to "court investors", work with the SAIL databank - an international anonymised medical records resource, and figure out how the job of a researcher is to facilitate successful studies, not chase any form of "complete truth" or look for actual problems (you look for problems you think you won't find, there's a knack to it). Your study design is flawed if it will result in something you can't/won't publish (because it's not the results you/your sponsor wanted). We've seen this time and time again in science, from the "saturated fat is bad for you" trope in the 60s by the AHA, which other independant researchers worldwide did repeat studies and did not find evidence to support that statement, yet did not publish for fear of being ridiculed (leading to five decades of bad nutritional guidance, and Crisco oil selling well, the product of Proctor & Gamble, who had made a sizeable $20 million (adjusted for inflation) donation to the AHA, propelling it into relevance). It wasn't until the 2010s that worldwide scientific consensus tipped towards completing some 30 odd international studies demonstrating that not only was saturated fat not bad for you, but that consumption of it from animal sources actually decreased cardiovascular and stroke mortality, pretty much the opposite of the original "diet heart" hypothesis from Ancel Keys.

The actual rigor of the studies is less concrete than you might imagine if you don't have experience with the industry. It's littered with questionably funded (by source) or non published studies that disagree with preferred outcomes and turn up later (the All Trials campaign to publish all current and previously completed clinical trials attempts to fix this, but good luck, https://www.alltrials.net/find-out-more/all-trials/). Or drugs that were approved for use (bureaucratic check boxes ticked), but turned out to have incredibly nasty effects, so the drug is hastily repurposed for a different use under a different name to recoup some sunk costs. Thalidomide would be an off the top of my head example of this. Big pharma then gets a huge record setting (slap on the wrist when compared against profits) fine, and business goes on. Court proceedings or FOI requests years later reveal the shortcuts, bribes, misleading studies etc that went into said approval, but by then the news cycle has moved on and noone cares, let alone now established science that further built upon the references of that study that says "drug X is completely safe for pregnant women".

AI having oversight of the legality/actual rigour of the internal workings of the companies and authorities involved, a "big brother" would do more for the safety of drugs (can't believe I'm arguing the devils advocate here, i'm strongly against this kind of thing) than making the regulation process more involved, as it's not the letter of the law being the main issue, but how easily it's gamed when there's big money to be made. The personal statement that got me into medicine touched on this.

...an updated, holistic approach is due; bringing medicine into the information age, pre-emptive, personal monitoring that accurately predicts outcomes through deep learning systems, worn or implantable bio monitors and synchronisation to hospital AI. Whilst remaining confidential, this evolving system, learning from every case, in every connected hospital, worldwide, could provide reliable diagnosis and evaluation, minimise the risk of human or legal error, and go beyond the scope of individual or even institutional achievement

Remember that human beings screwed up the environment.

It's human behavior that's hard to change even if we already know what's better.

Let's say the fancy AI ecology model said "use the compost bin and eat more beans in lieu of animal protein." Hell, half the people who currently use my complex's compost bin don't use it correctly. And that doesn't factor in all the people (most residents) who don't use the compost bin at all, they just toss their compostables into the trash bin or grind everything up in the food disposer and send to the water treatment plant.

And would you be willing to reduce the amount of meat you consume and replace several servings every week with beans?

Pretend someone creates an AI healthcare model which then says "exercise, quit smoking, go to bed early and get 8-9 hours of sleep every day." How many people are going to change their daily routine? What if it said for every two hours of videogame play, you should do one hour of physical exercise? You gonna follow the AI model's advice? Your doctor? Your spouse? Your mom?

Individuals/societies ignoring sound advice has been the only consistent thing throughout history. You've got a point there.

But then again, there's a subset of people who will do it just because "AI is so cool".

https://journals.lww.com/co-endocri...tory_of_saturated_fat__the_making_and.10.aspx Reference for paragraph 2.

 

[cvaldes]

Individuals/societies ignoring sound advice has been the only consistent thing throughout history. You've got a point there.

But then again, there's a subset of people who will do it just because "AI is so cool".

I will bet a buffalo nickel that the people who ignore sound advice greatly outnumber those who will blindly follow AI because it is "so cool" [sic].

I also guess that some people will deliberately go out of their way to do exactly the opposite of what an AI program advises.

And in a few months, maybe some of those AI worshippers will find some other bandwagon to jump on. What happens then?

There's good advice out there. There's bad advice out there.

Go to the supermarket and stand in line. Look at your fellow shoppers. All of them have been given advice on how to live healthily whether they asked for the advice or not. Do you think that some AI app on their smartphones will convince them to exercise more, eat better, sleep soundly?

Hell, how many people join the gym in early January only to abandon their workouts by late February?

But thanks for the laughs. You did a banging job!

 

[dgianstefani]

I will bet a buffalo nickel that the people who ignore sound advice greatly outnumber those who will blindly follow AI "because it's cool" [sic].

I also guess that some people will deliberately go out of their way to do exactly the opposite of what an AI program advise.

And in a few months, maybe some of those AI worshippers will find some other bandwagon to jump on. What happens then?

Ah, it really is in our nature to be contentious isn't it.

Maybe the AI should just focus on giving us all lobotomies first, before trying to create paradise on earth.

(not being serious, in case some future AI is reading this)

 

[Onasi]

@dgianstefani
*sigh* Yeah, that sounds about right. I work in a research facility where we specialize in organic contaminants both in their effects on marine and fresh waterways and bodies AND in advising industrial facilities on their filtration and purification systems to, you know, minimize the environmental impact. The amount of times I’ve beheld horrifically spiky samples from what was on paper perfectly modern and ostensibly working treatment plants totally built to spec by, say, a chewing gum factory or seen blatantly BANNED compounds in anti-fouling paints from shipyards that were licensed and in theory should not be pulling this shit is… yeah. Yeah.

 

[thesmokingman]

I think AI is way overblown, especially the craptastic LLM crap. Until manual labor is overtaken by bots and everyone can own their own bot, it won't be a world a plenty till we all get our own optimus.

 

[cvaldes]

It won't be the shortcomings of AI models (now and then) that will be the problem.

The ultimate challenge will always be getting people to do the right thing and to stick with it. The abandoned New Year's resolution workout is simply one example. Remember that these are all people who know what that right thing is and they are on board with it.

 

[dgianstefani]

Watching the GTC Keynote, interesting.

Whether it's a bubble or not, the sheer weight of industry/budgets going into AI development and the speed of research is very impressive.

Companies presenting at GTC.

 

[Onasi]

Ah yes, US Army and Navy, my favorite companies. Oooh, Northrop is there, as is Raytheon. And Lockheed. Lovely, can’t wait for war profiteers to get their hands on AI. Well, even more than they had already. We talked about how big pharma can be scummy, but those are puppies compared to this scum.

 

[cvaldes]

Watching the GTC Keynote, interesting.

Whether it's a bubble or not, the sheer weight of industry/budgets going into AI development and the speed of research is very impressive.

Undoubtedly there will be a wide range of results from various efforts. Not every AI development project will be a smashing success.

For good AI results you probably need some good datasets plus some good models programmed. Not everyone can do that. Time, talent, budget. Some will bite off more than they can chew, too ambitious with their early efforts.

In the same way, if you ask 100 people to bake a cake and frost it, you will see a wide variety of results. A few might look like something you'd buy at a nice pastry shop. A handful might be burnt, one or two might be puddles of batter because the baker forgot to preheat the oven. Some might look a fright because the frosting was poorly applied.

At least in the next couple of years, Nvidia will benefit for sure from the sheer volume of hardware purchases. AI programmers will have overflowing inboxes flooded with job offers. There's a lot of ancillary benefit to the semiconductor industry regardless of whose AI GPU is in your server rack. Semiconductor equipment manufacturing firms (ASML, Lam Research, Applied Materials, etc.), memory companies (like Micron, SK Hynix), networking equipment firms, server infrastructure, colo facilities, even the power company.

 

[dgianstefani]

Undoubtedly there will be a wide range of results from various efforts. Not every AI development project will be a smashing success.

For good AI results you probably need some good datasets plus some good models programmed. Not everyone can do that. Time, talent, budget. Some will bite off more than they can chew, too ambitious with their early efforts.

In the same way, if you ask 100 people to bake a cake and frost it, you will see a wide variety of results. A few might look like something you'd buy at a nice pastry shop. A handful might be burnt, one or two might be puddles of batter because the baker forgot to preheat the oven. Some might look a fright because the frosting was poorly applied.

At least in the next couple of years, Nvidia will benefit for sure from the sheer volume of hardware purchases. AI programmers will have overflowing inboxes flooded with job offers. There's a lot of ancillary benefit to the semiconductor industry regardless of whose AI GPU is in your server rack. Semiconductor equipment manufacturing firms (ASML, Lam Research, Applied Materials, etc.), memory companies (like Micron, SK Hynix), networking equipment firms, server infrastructure, colo facilities, even the power company.

Yeah, all true.

I think it's very helpful though how the design giants (Ansys, Cadence and Synopsis) as well as the manufacturing giants (Siemens, car industry etc.) are putting all the groundwork in. The fact that the supercomputers and the consumer GPUs use the same programming language, the same architecture (at it's root) and the same "visibility" to software, just differing in scale, bodes well for the small guys as well as the big guys (who paid to have the advantage of getting into the game a few years before everyone else, but also paid for the development and laid the foundations of stuff that will benefit everyone).

I was amazed by the potential of Omniverse a good while back, the idea of having single workspace/simulation that hundreds of other established software suites could tie into and manipulate in real time is very cool. Working with architects and engineers freelance in London a while ago, setting up their workstations etc, I was shocked at how many different pieces of software they used, and how complex it was for them to collaborate on the same project, with versioning, formats, compatibility, slow individual rendering etc. Just making sure everyone was on the same page was complicated.

What Jensen was talking about re: how far they want to take the digital twin concept (designing, manufacturing, testing and using their own hardware within the digital twin, before it's even physically manufactured) is also super cool. I guess you can see this with the culitho stuff, and with manufacturing, Autodesk, browser based cloud tools etc.

Ah yes, US Army and Navy, my favorite companies. Oooh, Northrop is there, as is Raytheon. And Lockheed. Lovely, can’t wait for war profiteers to get their hands on AI. Well, even more than they had already. We talked about how big pharma can be scummy, but those are puppies compared to this scum.

Yeah well, tools benefit everyone, good and bad. I just hope that the accelerant that is AI has overall positive outcomes for the world, as we're very much stepping into uncharted territory letting the computers think for us.