US lawmaker proposes a public database of all AI training material used by AI models.

[Kisai]

12 hours ago, QuantumSingularity said:

I see generative AI the same way i see battery EVs, which is basically like the early laser disk of our time - came with a lot of hype, supposed to be the future, but something better will actually replace it before it gets adopted by the masses.

With laser disk it was CD and then DVD. With the battery EVs it is going to be the HFC cars.

HFC is extremely doubtful since most of the Hydrogen is coming from natural gas. We'll probably see HFC for some high-range vehicles, but EV's are just in their teething stage where we have yet to design safe enough batteries. (Good gawd, like 40 EV's catch fire a day in PRC.) Where as in the West, people are brainwashed into a lot of silly ideas about EV's that they will continue to use ICE until they're not available.

 

The writing is on the wall for Gasoline/Petrol though. Large cities gas stations are being redeveloped and will soon end up as EV-only zones. They're not being replaced with superchargers either. Most new buildings are being designed to have destination changing. No more massive parking lots either.

 

 

12 hours ago, QuantumSingularity said:

 

In the case of generative AI it will be True AI which as we all know, once the genie is out of the bottle, there is no way back. Yes, everyone is constantly talking about AI right now, but in its current state, it's essentially just an advanced automation for wider range of processes, which were previously unable to be automated. And as such it should be governed. I am ok with that, but i want the training models to be disclosed not only with the government, but with the public as well. Everyone should have access to that information, just like all the ingredients are listed on all of the food items, instead of simply "it's a drink that gives you energy".

 

No AGI will be seen in 5 years, or 25. The current stuff, Generative AI, is just autocomplete. It doesn't understand anything, and unfortunately the people running or investing in AI fail to understand this, or maybe they perfectly understand this and are all showing off their best paper AI tigers.

 

There are cases where, yes, generative AI might result in some job losses along the lines of the "nobody actually wants to do this", but at present we keep seeing a lot of cherry picked paper tigers.

 

Something "wow" inducing comes out, and then a few months later another company has one-upped that one. Yet, can any of these AI's actually write, draw or sing from experience? No, because it has none. 

 

It's like the old addage of "book smarts" and "street smarts", Generative AI is "book smarts" without a single drop of "street smarts". INT 30, WIS 0.

 

 

 

[StDragon]

On 4/21/2024 at 3:03 PM, Kisai said:

The writing is on the wall for Gasoline/Petrol though.

Gasoline will never go away due to how products are fractioned from crude oil in the distillation process. In fact, the gasoline engine was developed specifically to "consume" what was once waste  dumped into rivers and oceans in the 1800s.

"Edwin Drake dug the first crude oil well in Pennsylvania in 1859 and distilled the oil to produce kerosene for lighting. Although other petroleum products, including gasoline, were also produced in the distillation process, Drake had no use for the gasoline and other products, so he discarded them. It wasn't until 1892, with the invention of the automobile, that gasoline was recognized as a valuable fuel. By 1920, 9 million vehicles powered by gasoline were on the road, and service stations selling gasoline were opening around the country. Today, gasoline is the fuel for nearly all light-duty vehicles in the United States." -eia.gov

So yes, due to the way crude oil is refined, you must do something with the waste (aka, gasoline). That's not to say you can't bias your refining and processing to reduce the amount of gasoline produced, but there will be a substantial amount to deal with. And, it will take a lot of investing with the existing refining infrastructure to do this.

Everyone will be driving ICE for the next 100+ years. It might make more sense for something like AI to assist in bioengineering life to consume and sequester CO2 instead rather than deal with the gasoline problem.

 

On 4/21/2024 at 3:03 PM, Kisai said:

No AGI will be seen in 5 years, or 25. The current stuff, Generative AI, is just autocomplete. It doesn't understand anything, and unfortunately the people running or investing in AI fail to understand this, or maybe they perfectly understand this and are all showing off their best paper AI tigers.

It's more than just "autocomplete". There's utility in the structure with the results it provides. In 5 years or less, there will be massive layoffs as people are replaced with AI. It will be global, and there will probably be treaties to regulate due to the economic upheaval it will have.

AI is no joke.

[igormp]

Hmmm, I find this complicated due to many reasons:

 

1. It kinda makes sense for public stuff, for internal-usage stuff it doesn't really make sense anyway.

2. It'll be hard to properly check what has been used or not. If someone says they're not using some specific dataset, or include a link to a dataset they're actually not using, you can't easily validate this.

3. As others have already said, just the metadata for the training stuff is useful to check which copyrighted content it was trained upon, but this stuff will eventually bitrot so there's not much more usefulness to it other than that.

4. Providing the dataset itself seems troublesome. As an example: what if they paid for content to train the model upon (with agreement from the data source), and then they'll have to release this to the public, then that's basically copyright infringement since 3rd parties haven't paid for the original data source. The question of how it should be provided, if it should be the "raw" source or the actual preprocessed data is also confusing.

5. What about fine-tune of models? How would the law apply to those derivatives?

 

I also agree with others that copyright should be enforced on the generated content, not on the training data, given that this is how the current laws apply to humans. You're free to look and take inspiration from whatever you want, but if you create something that's way too similar to someone's else work then you're liable to copyright stuff, the same should apply if someone makes use of some generative AI to produce work that's way too similar to some existing work and profit out of it, then the person should be liable for the infringement, not the tool that was used to create it.

 

On 4/12/2024 at 4:38 AM, 05032-Mendicant-Bias said:

I go further, and demand that the weights and the models themselves have to be public.

Personally I believe that this should be similar to any other copyrighted work, the company should be able to keep it "private" for some years, but then could be forced to disclose it after some years. However, that'd be pretty useless thinking of now, since by then you'd have a better model somewhere else anyway and those weights wouldn't be that interesting.

Models are also often fine-tuned further along the line, should all versions be released?

On 4/12/2024 at 4:38 AM, 05032-Mendicant-Bias said:

What large companies are allowed to keep secret, is the "secret sauce" of how they turn training data into models. That is the actual original work they are doing, and they are allowed to have that as trade secret.

I disagree on that. If I create or fine tune a model (its licensing allowing, ofc) for a specific use-case, like medical stuff or some other area, and assuming that I'm a small company, I'd like to keep the model itself private so larger companies don't just build a competing system on top of it and end up killing me. This has happened a lot with other open source software and tons of projects changed their licenses due to that.

[Kisai]

4 hours ago, StDragon said:

Everyone will be driving ICE for the next 100+ years. It might make more sense for something like AI to assist in bioengineering life to consume and sequester CO2 instead rather than deal with the gasoline problem.

 

We'll very likely see no more ICE passenger vehicles. It'll be things like muscle-car type designs (think race cars) and vehicles that don't make sense as EV's (eg small boats with outboard motors) that will continue to use gasoline. Diesel will likely just see a lot of bio-diesel conversion from waste oil products. Heavy machinery will likely continue to be diesel.

 

But if you live in a major city, the actual gas stations are going bye-bye. The large truck-stops will stick around on the periphery, but there won't be any more "drive 5 minutes to find a shell station", it'll end up being "Park your gas fueled cars here and take the subway/light rail"

 

https://www.theglobeandmail.com/business/article-gas-stations-electric-vehicles/

https://www.cstoredive.com/news/shell-sell-1000-c-stores/710299/

https://calgaryherald.com/business/energy/parkland-convenience-store-gas-stations-sale

 

Oil companies have been trying to get rid of gas stations because they see the writing on the wall.

 

4 hours ago, StDragon said:

 

It's more than just "autocomplete". There's utility in the structure with the results it provides. In 5 years or less, there will be massive layoffs as people are replaced with AI. It will be global, and there will probably be treaties to regulate due to the economic upheaval it will have.

AI is no joke.

I disagree. What we keep being shown are cherry picked results from cherry-picked trained models.

 

Two companies or people with the same dataset and same software, and same hardware, can end up with a completely different model just from the change in the GPU/NPU driver version.

 

A lot of people are just downloading models they find for free on the internet, damn the consequences. Anyone seriously thinking that the output from something like ChatGPT, Midjourney, or Stable Diffusion is "Good" and not heavily cherry picked first. The output from these things only impress people with no creative skill.

 

Hence we get al these snake oil AI's being pushed by companies, and sometimes they literately ARE fake

https://theticker.org/14111/business/amazons-ai-technology-mostly-relies-on-the-work-of-humans/#:~:text=However%2C in a recent report,could not determine a purchase.

https://www.theverge.com/2024/4/17/24133029/amazon-just-walk-out-cashierless-ai-india

 

All they've done is pushed the work from workers in person to workers over seas being paid 1/20th of the wages. What did the AI actually do here?

 

These present models, particularly the "LLM Prompt to output" understand nothing. These will not be replacing people, these will just contribute to the enshittening of all content types.

 

[StDragon]

6 hours ago, Kisai said:

We'll very likely see no more ICE passenger vehicles. It'll be things like muscle-car type designs (think race cars) and vehicles that don't make sense as EV's (eg small boats with outboard motors) that will continue to use gasoline. Diesel will likely just see a lot of bio-diesel conversion from waste oil products. Heavy machinery will likely continue to be diesel.

 

But if you live in a major city, the actual gas stations are going bye-bye. The large truck-stops will stick around on the periphery, but there won't be any more "drive 5 minutes to find a shell station", it'll end up being "Park your gas fueled cars here and take the subway/light rail"

 

The major issue is lack of electric load carry capacity; specifically the HV transmission lines and sub-stations at the local local level. You're not going to offload literal pipelines of hydrocarbon BTUs worth of energy to an already overburden grid. If anything, you'll have fleets of trucks running on natgas due to energy density and ease of distribution. It's basically hydrogen, locked up with carbon so you don't have to worry about transportation and energy loss from cryogenics of liquid hydrogen alone.

Speaking of, the industry is predicting a 20% to 25% increase in US electricity by 2030 just from datacenter usage alone for AI !!!

The Houston Chron reported a few days ago this:

ERCOT reported earlier this month that peak power loads on its system would rise 6% by 2030 to 94.3 gigawatts — with the caveat there was an additional 62 gigawatts of additional load asking to connect to the grid.

“Some of (the 62 gigawatts) will come, some of it won’t,” he said. “But even if it’s just one third of that, in five to six years time that’s shocking.”

Dominion Energy supplies power to Northern Virginia, which has the nation’s largest concentration of data centers and is known as the “crossroads of the internet,” and is predicting power demand across its entire territory will grow at a rate of 7.4% per year over the next decade.

 

 

6 hours ago, Kisai said:

All they've done is pushed the work from workers in person to workers over seas being paid 1/20th of the wages. What did the AI actually do here?

 

These present models, particularly the "LLM Prompt to output" understand nothing. These will not be replacing people, these will just contribute to the enshittening of all content types.

 

There won't be one monolithic AI, at least not in physical form. They will be connected with edge-AI performing NPU hardware. In many ways, this will mimic biological brains where regions have specialized functions, only these will be networked together. Density and locality will come later if/when needed. 

As for the "enshittening", absolutely! Refinement comes later. What matters is creating deflationary forces in the economy by cheapening the labor pool. Obviously this is a bad thing, but I won't digress there. But I think you and I are on the same page with regards to that.


 

[Techstorm970]

On 4/12/2024 at 4:39 AM, manikyath said:

it really feels like the law was made to prevent AI models from snooping on copyrighted works and "profiting" from them without paying the copyright owner their due share... which is really a good thing.

BUT they then didnt stop and consider just how such a thing would be achieved and what the result would look like... which is a really bad thing.

The entire story of how US Congress writes laws.

[Donut417]

On 4/24/2024 at 1:05 PM, Kisai said:

Oil companies have been trying to get rid of gas stations because they see the writing on the wall.

Well last I seen inflation has taken hold and there are in some cases 2 years worth of inventory sitting in car lots waiting to be sold. People are not buying. I just paid off my Nissan Rouge last month. I wont be investing in a new car / EV any time soon. Furthermore a lot of people have no way to charge them. Lots of older homes are not set up for charging an EV. OR you live in an Apartment complex and its highly unlikely the landlords are going to install chargers. The only charger I have seen in the wild is some Tesla chargers at a local ISH Meijer but thats for Tesla only according to the signage, and Taylor Michigan is not necessarily the safest place, I woundnt want to linger.

[Kisai]

7 hours ago, Donut417 said:

Well last I seen inflation has taken hold and there are in some cases 2 years worth of inventory sitting in car lots waiting to be sold. People are not buying. I just paid off my Nissan Rouge last month. I wont be investing in a new car / EV any time soon. Furthermore a lot of people have no way to charge them. Lots of older homes are not set up for charging an EV. OR you live in an Apartment complex and its highly unlikely the landlords are going to install chargers. The only charger I have seen in the wild is some Tesla chargers at a local ISH Meijer but thats for Tesla only according to the signage, and Taylor Michigan is not necessarily the safest place, I woundnt want to linger.

This is why we're likely to see the scenario I mentioned earlier, where people have to park their cars on the periphery of large cities. Gasoline-only cars will have to be parked there because there will be no fueling stations. EV owners without a charging solution, likewise.

 

Like, let me show you something.

This is what charging presently looks like over here.

If you need to drive between Vancouver and Portland, you're covered. 

 

Meanwhile shell locations:

 

People like to whine and argue about EV's not being as good as ICE but the reality is that they last longer, far longer, than ICE, and a lot of that is due to innovations in ICE for saving fuel/emissions that put far more wear on the car (eg the auto stop feature) that EV's don't have to deal with. The market that EV's will have a hard time penetrating will be the "my first car"/"new driver" market because EV's don't come in "basic" models. Where as traditionally, you would have a 4-cyclinder 1-3L compact or small car with a manual transmission and no power steering as a first car for maybe the cost of year on a part time job.

 

Automakers aren't selling as many EV's because only the battery pack ever needs to be replaced, and that's when people trade them in. Car's require so many safety features today to be street legal that you will find it difficult to have a car "cheapened" to the point where you are buried in it in an accident. That said, North American's are super fond of their over-sized vehicles for "Safety" reasons, and that's another reason why we won't see an EV uptake in the rural markets. Battery fires scare away people, despite ICE's result in fireball in most of the same scenarios. The "self-ignition" and "ran over something and it punctured the battery pack" are scares against EV's.

 

My aunt has a Tesla, and has to drive all the way to Vancouver to get it serviced, 4 hours away. The problem is not going to be fueling, the problem is going to be servicing.

 

Anyway this is extremely off topic. 

 

[Donut417]

1 hour ago, Kisai said:

eople like to whine and argue about EV's not being as good as ICE

My issue is I have no way to charge in my area. I have not seen one EV charger in my area outside of the ones that are Tesla chargers at the Meijers. There's no way I sitting for a 10s of minuets to hours charging a car in a public place. Furthermore is the expense. People are keeping cars longer because prices are stupid.