Enjoy!
Lots of interesting thoughts and ideas there.
A couple of quick reactions.
1. A machine will never be able to perform all the intellectual functions a human could. A machine could provide a material assist, and speed up the process, but no machine can make decisions in the manner a human brain could.
2. #1 depends on educating and developing the human brain. We are failing in that endeavor. We need to materially up our education game and that ain’t happening.
@co-hoosier what’s strange is that it gets parts of answers wrong. It also seems vulnerable to leading questions
@co-hoosier I agree with your basic idea.
To me the bigger concern is how criminal humans could abuse AI to sabotage or manipulate our primary infrastructures, financial, banking, transportation, communication, military, and so on.
To me the bigger question is, will civilized humans manage to use the power of AI to undermine the criminals?
Had a fascinating call yesterday on a project that will be using an absolutely massive supercomputer (massive in terms of computing power). There are a very small number of computers in the world with this level of computing power. This level of processing power is believed to match that of the human brain at the neural level. And no, it has absolutely nothing to do with AI (at least not AI like ChatGPT, etc.), and I do not believe this supercomputer is (or ever will be) used for AI purposes.
But having the same processing power as the human brain does not mean it can duplicate the capabilities of the human brain.
Cars in the NASCAR Cup Series all generally have the same capabilities (processing power). Yet, there are vast differences in drivers' performance in races. Take Shane van Gisbergen (Kiwi). In the four road/street course races this year he has won three (and finished 6th in the other one). His average place on ovals? 27th (and most races have fewer than 40 cars entered). But, no one has won more races this season (Hamlin and Larson have also won 3 each), and van Gisbergen will obviously be in the playoffs. (Unfortunately for him, there are no road courses until the 4th playoff race, and they cut from 16 to 12 drivers after the first three playoff races.)
Like NASCAR race cars, simply having the same processing power does not mean the system can match the human brain's capabilities. You also need the right driver (i.e., the software).
Another angle on criminal abuse is China. They have serious control over their internet. Maybe more significant, tight control of their workforce. There’s a world in which they not only leapfrog the US but also leave the world in their dust.
AI isn't intended to simulate the human brain. It will have it's own way of thinking. It's very unlikely that there is anything humans can do now that AI won't eventually find ways to do better.
We don't even know how LLMs are working this stuff out:Had a fascinating call yesterday on a project that will be using an absolutely massive supercomputer (massive in terms of computing power). There are a very small number of computers in the world with this level of computing power. This level of processing power is believed to match that of the human brain at the neural level. And no, it has absolutely nothing to do with AI (at least not AI like ChatGPT, etc.), and I do not believe this supercomputer is (or ever will be) used for AI purposes.
But having the same processing power as the human brain does not mean it can duplicate the capabilities of the human brain.
Cars in the NASCAR Cup Series all generally have the same capabilities (processing power). Yet, there are vast differences in drivers' performance in races. Take Shane van Gisbergen (Kiwi). In the four road/street course races this year he has won three (and finished 6th in the other one). His average place on ovals? 27th (and most races have fewer than 40 cars entered). But, no one has won more races this season (Hamlin and Larson have also won 3 each), and van Gisbergen will obviously be in the playoffs. (Unfortunately for him, there are no road courses until the 4th playoff race, and they cut from 16 to 12 drivers after the first three playoff races.)
Like NASCAR race cars, simply having the same processing power does not mean the system can match the human brain's capabilities. You also need the right driver (i.e., the software).
Here's a more detailed explanation: "
-
Mechanics are understood:At a basic level, LLMs are neural networks that process and predict the next word in a sequence. They use techniques like word embeddings, where words are represented as vectors, and transformers, a type of neural network architecture well-suited for sequential data.
-
The "black box" problem:Despite understanding the mechanics, we struggle to understand why LLMs perform certain tasks with such proficiency. We can see the numerical operations within the model, but we can't pinpoint the specific concepts or reasoning processes they use to generate accurate or creative text.
-
Emergent abilities:LLMs exhibit abilities that were not explicitly programmed or even anticipated during training. For example, they can perform complex reasoning tasks, solve puzzles, or even generate creative content, which are not directly taught but emerge from their training data and architecture.
-
Lack of internal reflection:LLMs don't have a sense of their own inner state. You can ask them why they gave a certain answer, and they might provide a plausible explanation based on their training data, but they don't have true self-awareness or understanding of their own reasoning processes.
-
The "thinking" question:The biggest mystery is how LLMs achieve what appears to be genuine understanding and reasoning. While they can mimic human-like language and problem-solving, it's not clear if they truly grasp the concepts they are using or if they are simply very good at pattern recognition and prediction.
-
The "hallucination" problem:LLMs can sometimes generate false or nonsensical information, known as hallucinations. This highlights the gap between their ability to process and generate text and their actual understanding of the world.
The fraud implications are also undersold.
Similarly, We also know much about how the human brain is built and how it is organized. We know very little about how it creates consciousness or how it process information.
Similarly, We also know much about how the human brain is built and how it is organized. We know very little about how it creates consciousness or how it process information.
There's an old Frank Herbert sci-fi novel called Destination Void, where they put expendable clones of a crew on several rounds of a spaceship mission, then have the central computer crap out so the crew has to develop AI to run the various life functions to survive.
After several missions, where the ship blows up, in the final flight, which the novel is about, the crew is successful, but the AI computer becomes omniscient and omnipotent.
The crew is tasked with how to properly worSHIP.
.


