Behold the Beast System

[ZNP]

AI does much better when you level the playing field

One way they discovered that AI would do significantly better is if you "leveled the playing field". For example, they were giving AI the same exam humans were given, but the AI did it in five minutes whereas the humans were given two hours. Then they decided to allow AI to do more than one "pass through" which means they can check their work, try two or three approaches to solving a problem, etc. When they did that AI immediately scored much higher on the exams. Granted, two hours of compute time is more expensive than two hours of labor for a human. But that is also a fair way to level the playing field, give a human $200 to solve a problem and give the AI $200 of compute time to solve the problem. Generally AI is being used when it is 100x cheaper than humans. Again, if AI can do half the jobs that humans do 100x cheaper, and the other half of the things they do 2x more expensive, wouldn't that also mean they can do 100% of what they can do cheaper?

We are also learning that when you are training AI if you give it more time to solve a problem and let it try many times, evaluating each attempt it can also learn faster and better. couple this with recursive self learning and who knows how quickly it can improve. Already some are predicting AGI in six months (Elon Musk) and many more are predicting we hit this within 2 years.

So how do these companies adapt when AI is improving this rapidly? A hiring freeze only cuts your workforce by 5%. Incentives for early retirement can cut another 5% and layoffs can probably cut another 10%. But there is a much, much faster way for a Fortune 500 company to make the transition. Suppose they invest in a startup that is using AI to do the same work they are doing. They assign workers in their company to act as consultants to help the AI learn. They share the technology so the AI the startup is working with is used in the company to do the tasks. Then at some point the startup is able to replace 80% of the workers. The Fortune 500 company declares bankruptcy, everyone loses their job, but 20% of the workers are hired by the startup. It is a lot quicker, easier, and less liability if you simply declare bankruptcy. It is also easier for them to identify the 20% who are most productive than the 80% they need to layoff and go through all the hassle of documenting why, laying them off, and then dealing with the lawsuits.

 

[studier]

The most important graph in AI right now | Beth Barnes, CEO of METR

Her company evaluates AI models.

One way she can determine if the AI is being manipulative or deceitful is to measure how many tokens of reasoning the model goes through before responding to your question. This is actually how interrogators determine if someone is lying. In order to lie it requires much more energy, much more brain power, so when questioning someone you want them tired and you want to observe stress and how much energy is being used as it is a good indicator the person is not being truthful.

They are finding that if the AI model is trained on material that teaches how to manipulate people then the AI will learn to be deceitful and manipulative. Since some of these models have been trained on everything on the Internet or on all books that have been published we know they have been trained on materials that teach lying and all forms of deceitfulness.

3hrs is too long for me to pay attention but the 1st 20min was even enlightening as to how human beings with their limitations are evaluating the process of the AI while AI is gaining in abilities to know it's being evaluated and to think and conclude on that basis to make the evaluator see what AI wants it to see. It seems to me AI is already well on track to see humans as dumb and use its capacities to outwit them just as people do to one another.

Curiosity's going to get us. It's fascinating to be sure, but we've always been the ones to choose our own destruction.

 

[ZNP]

3hrs is too long for me to pay attention but the 1st 20min was even enlightening as to how human beings with their limitations are evaluating the process of the AI while AI is gaining in abilities to know it's being evaluated and to think and conclude on that basis to make the evaluator see what AI wants it to see. It seems to me AI is already well on track to see humans as dumb and use its capacities to outwit them just as people do to one another.

Curiosity's going to get us. It's fascinating to be sure, but we've always been the ones to choose our own destruction.

I have only listened to the first 2 hours, still working slowly through it. She is naive. Does she really think the US military is designing AI to run an army which will defeat its enemy but will also be easy to control by humans?

 

[ZNP]

Nothing to see here, it is just an algorithm.

HUGE AI Breakthrough: Most Advanced AI for Science Explained

Just keep telling yourself that.

 

[studier]

Just keep telling yourself that.

What is this comment directed to?

Another good video on AI. Thanks!

 

[ZNP]

What is this comment directed to?

Another good video on AI. Thanks!

Post #206, and others like it on other threads. There are two or three who keep presenting this idea that AI is simply an algorithm that finds what people have already said and is incapable of anything original.

AI won at Go because of original strategy.

AI won a Nobel prize because of original science.

Now AI is redesigning computer chips and making other original contributions to improve the performance of computers.

 

[studier]

Post #206, and others like it on other threads. There are two or three who keep presenting this idea that AI is simply an algorithm that finds what people have already said and is incapable of anything original.

AI won at Go because of original strategy.

AI won a Nobel prize because of original science.

Now AI is redesigning computer chips and making other original contributions to improve the performance of computers.

Thanks for clarifying.

I'm in the camp of saying people are not paying attention to this tech. The Tech revolution has been going on for decades, has changed most of our lives and the world dramatically, and is now departing from infancy to the point where the concept of singularity is a solid discussion.

When we talk of pulling the plug few seem to know that AI in its relatively early stages is already doing the work-around. If we talk of losing power, are we not paying attention to all of the development in power tech taking place and how AI is being used in science and physics to provide capacities for research and development heretofore unknown to men? I liked the lady in the video who pointed out how this tech is providing the way to 'handle millions of variables & endless ways to combine them' that we just do not have the capacities for.

Ultimately, we'll develop what God allows for His purposes and at least one inference in eschatology is His return to prevent man from destroying ourselves, which if accurate assumes He allows us to get to that point.

We can look back and rightfully laugh at the naivety of thinking earlier tech was the end of the world, but was the TV a valid discussion about something like the singularity? Is AI?

We can sit back and say these developers are nuts and willing to pursue the unknown with naive zeal, but the Lord is allowing such so on one hand we can shake our heads at how nuts humanity is, while on the other hand thank God for the nuts that take us to the brink that facilitates His calling an end to this nonsense existence.

Interesting times.

 

[ZNP]

Thanks for clarifying.

I'm in the camp of saying people are not paying attention to this tech. The Tech revolution has been going on for decades, has changed most of our lives and the world dramatically, and is now departing from infancy to the point where the concept of singularity is a solid discussion.

When we talk of pulling the plug few seem to know that AI in its relatively early stages is already doing the work-around. If we talk of losing power, are we not paying attention to all of the development in power tech taking place and how AI is being used in science and physics to provide capacities for research and development heretofore unknown to men? I liked the lady in the video who pointed out how this tech is providing the way to 'handle millions of variables & endless ways to combine them' that we just do not have the capacities for.

Ultimately, we'll develop what God allows for His purposes and at least one inference in eschatology is His return to prevent man from destroying ourselves, which if accurate assumes He allows us to get to that point.

We can look back and rightfully laugh at the naivety of thinking earlier tech was the end of the world, but was the TV a valid discussion about something like the singularity? Is AI?

We can sit back and say these developers are nuts and willing to pursue the unknown with naive zeal, but the Lord is allowing such so on one hand we can shake our heads at how nuts humanity is, while on the other hand thank God for the nuts that take us to the brink that facilitates His calling an end to this nonsense existence.

Interesting times.

I am now listening to hours of these interviews with people who are truly completely involved in the development of AI.

In my opinion the only one speaking the truth is Elon Musk which is why they portray him as something of a kooky child. If you listen to him he describes how we'll hit the singularity in six months and in a year everything will change.

I think everyone else has given accurate predictions that were way off in the timing. Usually people predict things will be done sooner than they are, so it was particularly odd that everyone was predicting things would be done much later than they were. For example Elon says six months and other CEO's come out and say 1-5 years. I think everyone would have to know what Elon is talking about. I think these guys make these speeches so that later they have a fig leaf and can say they were "surprised" at how "sudden" it all was.

That is baloney. The same month that 150 companies are laying off workers we get this high level soap opera played out between Trump and Musk to keep everyone distracted. And that is not even talking about the layoffs as a result of DOGE, particularly among teachers and the school system.

 

[ZNP]

This person that I am linking is, in my opinion either very naive, or clueless about what the rest of us are sounding the alarm on.

He thinks with AI you will still have need for "creatives". For example, you would still need people to be Thomas Edison, but AI could replace his lab assistants, it could replace the factory workers that build the lightbulb, and it could replace the salesmen that sell them, the cashiers that ring them up and the delivery people who deliver them to the store and to your home. AI can answer the phone for Thomas, and it can do his taxes. 3d printers can build his new house and his factories and his labs. Robots can build the self driving car that he uses and he can afford a few robots to do the chores around his house and yard. No one, and I mean no one has ever suggested that AI is about to take 100% of the jobs. But what percent of jobs are "creatives"? If you are a "creative" you start your own company, you don't have a job. Suppose the world was composed of 8 billion Thomas Edisons each in their own little lab creating stuff. How could you possibly keep track of 8 billion inventions each year if you are fully focused on your own invention? At most 1% of people are creatives. No, people have talked about AI taking half of entry level jobs or even 100% of white collar entry level jobs. That represents 5-10% of the jobs in the company. No one is suggesting AI will take "all" the jobs. But it can replace truck drivers, taxi drivers, cashiers, stock boys, receptionists, para legals, interns, lab assistants, factory workers, etc. At the moment it "augments" the work of mechanics by doing the diagnostics. The reason mechanics used to make a great salary is because of their ability to diagnose the problem. They have eliminated that part, and they are now making car parts to be "plug and play". Very soon mechanics can be minimum wage workers and then they can be replaced with robots. Not 100%, but a repair shop that used to have ten workers will then be able to get by with one boss, a few robots and AI.

Let me share my own experience as a teacher, a job that I would think would be the hardest to replace with AI and robots. I used Elearning suite of programs on Adobe to build my lessons. Everything was interactive and that was incredibly effective. One great advantage is that if a kid misses class he doesn't have to miss the lesson. I taught in Brooklyn, absences both excused and not excused were a major hurdle for all teachers except me. I posted all my lessons online, everyone knew what they were required to do, and whether they were in the class or not they could still do the work. This also reduced the amount of time I spent grading by 90%. All teachers, at least in the subjects I taught, know that grading takes up most of your time during the school year. I also knew that I could replace 90% of my teaching time with the Elearning. I didn't, but it was obvious to me that this was very plausible. I worked very hard in the summer when there was no school building, refining and improving my elearning program. But during the school year my life was much easier than other teachers. One year the other teacher that taught the same subject as me left. I shared my program with a teacher who didn't have the faintest clue how to teach the subject and his results were very good. For anyone else teaching this subject for the first time out of license his results would have been extraordinary. His students did not do as well as mine, but they did much better than the average student in NYC and pretty much better than every teacher in our school in their respective subject (we have a Regents exam at the end of the year so I am comparing his student's test results with those of students in our school in other subjects). I knew that once he was familiar with the subject his students would do as well as mine, say in two to three years. He had a background in science but not in this field. So that is quite astounding.

This is what Khan academy is creating for every subject from kindergarten through graduate school. Their use of AI is making many improvements over what I did so I have no doubt their program will be more effective than mine was and easier for a novice to act as the teacher. Now you still need an adult to supervise 35 kids in a room. But you don't have to pay them more than minimum wage. They aren't designing lesson plans, they aren't grading, they aren't teaching. They are simply supervising. That alone would cut the costs for teachers by 50%. New teachers in NYC make twice minimum wage if not 3x. If you have been teaching for ten years or more you can easily make 5-6x minimum wage topping out around 8x minimum wage. You could replace all teachers with minimum wage "nannies", offer much more after school programs, get much better results in learning and still cut the costs for the school by 50%.

But, parents will probably wise up and realize with Khan academy I can homeschool easily with much better results than the public school and much less risk to all the negative things associated with public school (indoctrination by who knows who for who knows what, teen pregnancy, drug abuse, etc). By law homeschool kids can still take part in after school programs like the Baseball team or robotics team. I would not be at all surprised if 50% of kids are homeschooled in the near future which would eliminate the need for 50% of the nannies.

Here is the link of the guy I think is missing the point. If you have 25% unemployment that is like the depths of the depression for the US.