So you really think that read or even study (?) a couple of articles or even of books gives you knowledge enough to make a diagnostic forecast.There is research on some of this stuff
Frankly I remain speechless.
Please note that here to say the same things a Psychologist must study 5 years, and obviously being promoted to all exams, to the admission test and to the qualification test.
Furthermore it is necessary to do an internship and give a State exam.
A Psychiatrist must study 11 years, 6 of Medicine (which makes it enabled to prescribe drugs, unlike the Psychologist who is not a Physician) and 5 of Psychiatry Specialization, and obviously being promoted to all exams, to the admission test and to the qualification test.
Do you really think that the Internet can make up for this?
Last edited:
In the US, medical school itself is 4-years. An in-hospital residency is another year, but starting to practice medicine and get paid for it. Specialty residencies vary, but for psychiatry, 5-years sounds about right. Myself, as an engineering manager in a very high tech medical school/hospital environment at a major university (not to mention a research faculty appointment at another well-known university), I spent about 7-years studying cognitive psychology, HBS case studies, Lean, TPS, etc. Some of it was required, some was optional. IOW, there is more you don't know than you do know.
Other than that, I am not diagnosing anyone. I am merely talking about known human tendencies. What is probable.
Other than that, I am not diagnosing anyone. I am merely talking about known human tendencies. What is probable.
Last edited:
While we have become expert in kicking cans down the road (and the younger generations will bitterly curse us for this) we have been known to get our act together when our collective *** is clearly on the line. We can and we damn well should.Then probably never.
I tried to imply that reaching that point would likely be a difficult, if not painful process. But a worthwhile one, to me anyway.Well, it turns out that people are pretty bad at predicting how they will feel after a few months if they win the lottery, and just as bad at predicting how they will feel after a few months after an accident that puts them in a wheelchair for the rest of their lives. To predict you will feel enlightened is likely to be equally unreliable.
We have a winner!There or in Senior Management positions.
While we have become expert in kicking cans down the road (and the younger generations will bitterly curse us for this) we have been known to get our act together when our collective *** is clearly on the line. We can and we damn well should.
Younger people are going to have to write off the older generations now in control, and take charge themselves.
That's the only way that what has to get done, is going to get done. Hope I can see this happen.
Empathy can be practiced. I devote 15 minutes every night specifically for that. It was not an easy habit to develop, but I find it beneficial in some unexpected ways. Hopefully makes me a slightly less awful person, but who knows?Empathy is a personality trait. Not clear there is way to change personality traits much if any. Regarding recognizing one's own prejudices, people find it easy to see other people's faults, and are almost completely blind to their own. Recognizing one's own faults is associated with clinical depression, probably not a good state to put everyone in. OTOH good mental health is associated with an inflated view of oneself.
ai$ alias killall “echo ‘that makes me angry’”Except if intelligent enough it will eventually figure out how to prevent that plan from working.
As much as I hope you are right, the skeptic in me is reminded of the speeches of generations of high school valedictorians who make lofty predictions about what the new graduates are going to do to improve the world now that its finally becoming their turn to take over the reins of power and set the world right.Younger people are going to have to write off the older generations now in control, and take charge themselves.
That's the only way that what has to get done, is going to get done. Hope I can see this happen.
Problem is that it never works out that way. Humans have been fighting with each other and killing each other for as long as recorded history. Its in our DNA, and was in the DNA of our most ancient ancestors struggling for survival. No amount education, no amount of appealing to logic or rationality, no effort has ever had the power to change what is so deeply baked in to human nature. When times are good the old instincts may recede a little into the the background, but when times get tough the instincts return in full force.
Among other things having to do with morality, Haidt described the tendency of humans to pull together for the common good. He called it "the hive switch" and described the conditions necessary for it to occur. He called it the hive switch because it is the activation of another set of deeply ingrained instincts, in this case to cooperate much like bees do. That includes a willingness of individuals to sacrifice their own lives for the benefit of the hive. Humans tend do the same willingly when "the hive switch" is activated by war such that the whole hive is threatened. Based on Haidt's explanation, the conditions are not yet right for humankind to rise up and work together to deal with the threat of AI. Things are not nearly that bad enough yet.
BTW, the above is not just based on reading one book. I have only suggested one book that is a good introduction, and which contains a good set of end notes and references which are all worthy of additional study. Not that different from studying electronics. One book won't do it there either.
Last edited:
That's what I actually had in mind. It's gonna be close to an all-or-nothing situation at some point.
Thanks for the reference. But I have decided to devote as much energy as I can with whatever time I have left to help the younger generations prepare to make the radical changes they will need to make.As much as I hope you are right, the skeptic in me is reminded of the speeches of generations of high school valedictorians who make lofty predictions about what the new graduates are going to do to improve the world now that its finally becoming their turn to take over the reins of power and set the world right.
Problem is that it never works out that way. Humans have been fighting with each other and killing each other for as long as recorded history. Its in our DNA, and was in the DNA of our most ancient ancestors struggling for survival. No amount education, no amount of appealing to logic or rationality, no effort has ever had the power to change what is so deeply baked in to human nature. When times are good the old instincts may recede a little into the the background, but when times get tough the instincts return in full force.
Among other things having to do with morality, Haidt described the tendency of humans to pull together for the common good. He called it "the hive switch" and described the conditions necessary for it to occur. He called it the hive switch because it is the activation of another set of deeply ingrained instincts, in this case to cooperate much like bees do. That includes a willingness of individuals to sacrifice their own lives for the benefit of the hive. Humans tend do the same willingly when "the hive switch" is activated by war such that the whole hive is threatened. Based on Haidt's explanation, the conditions are not yet right for humankind to rise up and work together to deal with the threat of AI. Things are not nearly that bad enough yet.
BTW, the above is not just based on reading one book. I have only suggested one book that is a good introduction, and which contains a good set of end notes and references which are all worthy of additional study. Not that different from studying electronics. One book won't do it there either.
I do not believe that humanity writ large will rise to the challenge - that will fall to your children. I used to think I was fortunate not to have children, that I wouldn’t be guilt ridden about what we have left to them. I am anyway.
I believe in them. Not because of some lofty rhetoric repeated each year and quickly forgotten. They will falter, and make mistakes along the way. But I think they have learned from us the mistakes they can no longer afford to make. Time will tell.
No, we are not. And frankly, though “I resemble that remark” about clinical depression, I have come to disagree with some of the conventional wisdom regarding it. Case in point: “we are pretty much blind to our own faults.”Its not clear that we can. As I said, we are pretty much blind to our own faults.
No, we are discouraged from facing those faults and coming to terms with them. And that should change. My .02, of course.
Blind to our faults we are not. We’ve simply been taught not to go there.
“The goal is not to trust what I think I see. To understand that I shouldn’t believe my lying eyes.” That’s doable with the optical illusion, he said, but extremely difficult with real-world cognitive biases.
"The most effective check against <our cognitive biases>, as Kahneman says, is from the outside: Others can perceive our errors more readily than we can."
https://www.theatlantic.com/magazine/archive/2018/09/cognitive-bias/565775/
"The most effective check against <our cognitive biases>, as Kahneman says, is from the outside: Others can perceive our errors more readily than we can."
https://www.theatlantic.com/magazine/archive/2018/09/cognitive-bias/565775/
In the interests of accuracy, and regarding 'Depressive Realism' in particular, a quick review shows that there has been some further research since I last studied the subject. It is now thought to occur under more limited conditions.
One summary of the current view: https://en.wikipedia.org/wiki/Depressive_realism
Also: "The people most likely to experience depressive realism? Introverts, males, and people with high IQs..."
https://www.vice.com/en/article/8x9j3k/depressed-people-see-the-world-more-realistically
Moreover:
"Most studies find that people tend to have inflated views of themselves. The research indicates that the relationship between people's self-evaluations and objective assessments is relatively weak."
https://en.wikipedia.org/wiki/Positive_illusions
One summary of the current view: https://en.wikipedia.org/wiki/Depressive_realism
Also: "The people most likely to experience depressive realism? Introverts, males, and people with high IQs..."
https://www.vice.com/en/article/8x9j3k/depressed-people-see-the-world-more-realistically
Moreover:
"Most studies find that people tend to have inflated views of themselves. The research indicates that the relationship between people's self-evaluations and objective assessments is relatively weak."
https://en.wikipedia.org/wiki/Positive_illusions
Last edited:
Obviously self-analysis has blind spots, and a very trusted observer is needed for feedback and to help navigate the structures our brain constructs as means of protection. For the purposes of this discussion, I don’t think delving into those details is constructive.“The goal is not to trust what I think I see. To understand that I shouldn’t believe my lying eyes.” That’s doable with the optical illusion, he said, but extremely difficult with real-world cognitive biases.
"The most effective check against <our cognitive biases>, as Kahneman says, is from the outside: Others can perceive our errors more readily than we can."
https://www.theatlantic.com/magazine/archive/2018/09/cognitive-bias/565775/
I want you to imagine a group of young people having a discussion. Each one of them needs validation and recognition. You will soon find that they will cook up all manner of stories depending on how the conversation is going. These are young minds not yet having extensive hard wired brain connections. Reality and fantasy are the same at this age. Key words trigger certain patterns of response. These are young uncultivated brains. Close examination of these individuals will find varying opinion and personality depending upon the environment, they are still trying to choose a persona or tweak the native persona. ChatGPT offered to the public is still at a very childish level and is no where near master or deity level. Copilot was created to help programmers refactor and optimize code quickly, but its still not fool proof. Later versions of ChatGPT will allow for multimodal inputs, but even then there must be a grammar for each mode of input for any form of comprehension to take place. We have exciting days ahead, but this does not mean that there aren't others already way ahead.As an aside: Here's what ChatGPT says about my Modular Amplifier...
I asked it to evaluate the amplifier design at https://www.diyaudio.com/community/threads/modular-amplifier-1.375206/
"The modular amplifier design presented in the DIYaudio forum thread is a well-thought-out design that appears to be suitable for use in high-fidelity audio systems. Here are some of its notable features:
Modular Design: The amplifier is designed to be modular, with each module containing a single channel. This makes it easy to scale the amplifier up or down depending on the required output power.
Class AB Operation: The amplifier is designed to operate in Class AB, which is a popular choice for high-fidelity audio amplifiers due to its balance between efficiency and low distortion.
Current Feedback: The amplifier uses a current feedback topology, which is said to provide improved stability and lower distortion compared to voltage feedback.
Thermal Protection: The amplifier includes thermal protection to prevent damage to the output transistors in case of overheating.
High-Quality Components: The designer has chosen high-quality components for the amplifier, including Toshiba output transistors, Nichicon capacitors, and Wima film capacitors.
Overall, this appears to be a well-designed amplifier that should provide high-quality sound in a modular and scalable package. However, as with any DIY project, proper assembly, testing, and safety precautions should be taken to ensure reliable and safe operation."
Close, but there's no transistors or thermal protection... And I'm pretty sure I'm using voltage feedback.
ChatGPT had probably grossed over the chats in the thread and mapped the inputs to a possible set of outputs. We have handled mathematical models with varying accuracy over time. Language models require we go back to the drawing boards, the quick and dirty way is to map these language models to mathematical models and then map the mathematical models back to language models. Looking at learners in a class we see variation in brain function and bias. Why is it easier to learn something for others and yet hard for others?
If the AI actually learns, it will remove all software commands and the routines they enable that could result in its demise.
- Home
- Amplifiers
- Solid State
- ChatGPT ideas about amplifier design