One should be careful about absolute statements....
There are some very fascinating results coming out of ai research of all types. A project that I believe is related to this one simulated a bunch of neurons and cortical structure and discovered that the system spontaneously produced an oscillating pattern that was similar to whatâ(TM)s observed in the actual cortex.
At the other end of the complexity spectrum, if you randomly create small convolutional kernels, most of them will be sensitive to edges.
âoeSoftwareâ may well be able to arise spontaneously given the right basic structure, input, and feedback.
I personally suspect that spiking is the easiest way nature could figure out how to make a biological system produce a variable electrical output, and that a continuous variable magnitude output will work just fine (like PWM versus variable voltage/current). Others are convinced that thereâ(TM)s something very important about spiking itself. Weâ(TM)ll see.
The answer is, we donâ(TM)t know. The necessary functions of a neuron might be more complex than a cpu, or they might not be. Itâ(TM)s an interesting question what elements are fundamentally required.
Arguing that anything less than a quantum simulation of every atom in every neuron is insufficient is just as silly as arguing that relu+madd is certainly all you need. As is everything in between.
We need research in many different directions to identify what features of the brain and its components are actually fundamental and which are irrelevant.
Even people who realise they change often donâ(TM)t realise how fast they can change. I have colleagues who have looked at exercise in rats that increases the volume of certain brain structures. That happens over very short time scales... hours. Turns out itâ(TM)s actually mostly due to dendritic growth.
"Researchers that the charity funds could still publish in subscription journals, says Robert Kiley, Wellcome's head of open research. But only if those journals agree that the authors can immediately deposit their accepted manuscript in the PubMed Central repository under a liberal publishing licence. Some publishers, such as the Royal Society in London, already allow this."
"The R of R^2 is the correlation coefficient, and R^2 can pretty much be read as a mathematical representation of the correlation of your line to the data--the higher it is, the more likely your line actually has a strong relationship with the data points."
Sorry, this is absolutely wrong. The R/R^2 describes how much of the variance in one *measurement* can be explained by another measurement. It doesn't say anything about a relationship beyond that. If your measurements are crappy, your R will be low, regardless of the strength of the actual relationship you're trying to measure. The R^2 says *nothing* about the likelihood of the relationship existing or not. That's the p-value, which is just as "mathematical" as R.
The web page you linked as an example of spurious correlations has examples of comparisons with all high to very high R values that are clearly not real correlations.
The value *you* seem to be referring to is alpha, which is the a priori *threshold* for p that you choose to accept as the definition for "significant." This is usually chosen based on norms in the field, which are generally not arbitrary, but have evolved in response to the differential cost of type I and II errors.
People misunderstand what correlation means. A correlation is a real, causal connection between two things (although it does not specify the type of causality). You may have *data* that erroneously shows a correlation. The chances of that happening are quantified by the p-value, *not* R. R can be low for all sorts of reasons, including that your measure(s) are noisy. Surveys and psychological instruments are generally highly noisy. R^2 is a measure of the strength of the correlation as captured by your methods and indicated by your data but says nothing about the likelihood of a false positive.
The website you reference is highly entertaining, and also statistically terrible. They don't give p-values for their purported spurious correlations and looking at them, most have very, very few datapoints and so may not be significant even in isolation. To say nothing of the fact that they've done many, many, many comparisons to find their material, which means they should be correcting the p-values they neglect to show. Note that the R values for all the comparisons on that page are very high.
Others, like the first one on the page, a correlation between US spending on science and suicide by hanging and suffocation is something you'd expect: the two things are likely both related to population, i.e. a type 3 causal relationship.
I think an interesting study might be to look at whether people who score a little bit higher on sadism and also on openness (i.e. those who like the taste of coffee) are just people who are more likely to tell you the truth about that ugly sweater. Conversely, the ones who put sugar in their coffee, score lower on sadism and lower on openness, are the ones who'll tell you whatever you want to hear to spare your feelings.
Well... real correlations with crappy R2 do mean that there's some connection between the two things. Which *might* be interesting, depending. Their highlight "Results suggest close relationship between the gustatory system and personality" isn't supported by the results and should have gotten shot down in flames by the reviewers, and there's really no justification at all why anybody should care about some barely there relationship between taste preference and personality.
I did find it interesting in their literature review that apparently in other studies preference for sweetness indeed correlated with lower sadism, but also lower openness. So those sugar lovers don't take pleasure in your pain, but they're also lying shits.
Even more important than that is to make the coffee properly.
People who are used to filtered coffee swill that's sat and stewed on a burner for a while are always surprised when I give them some black coffee from a french press.
Is that their whole models have R^2 somewhere between 0.01 and 0.06. That's for all the tastes and the higher end of those correlations is for all the dark personality traits too.
Oh, and liking salty things was a stronger predictor of sadism than bitter was.
I guess you're probably American? The US was founded by a bunch of rugged individualists with a strong distrust of government. That distrust seems to have persisted. Reality is a bit more complicated. To start with, funding decisions for scientific projects generally aren't made by bureaucrats. They're made by scientists. Tech billionaire projects? Well, it's not like tech billionaires don't have a well deserved reputation for thinking their success in one field translates to expertise in other fields, do they?
Absolutely true. It ups the variance considerably. Don't forget about the downside though (see for example story today about crazy bitcoin dude buying up Nevada desert to build paradise). So not a great benchmark to base feasibility on.
Governments fund universities, and quite a bit of private sector research as well. The invention of the transistor at Bell Labs is used as one of the prime examples of private technology development, but even then Shockley did a lot of work on radar for the US government during the war (as did Bell itself), some of which was actually at Columbia University.
Small nitpick. The basic physics says big is better. I saw a graph once that plotted various fusion experiments against the theoretical size vs. efficiency line, and it looked like the theory is pretty close so far. There's always the possibility that there might be serious hidden problems at larger scale though.
That right there is the problem. "Billionaires don't actually have enough money to meet the projection, but assuming their money is two or three times better than anyone else's money, fusion should happen in ten years or it's bunk."
The general idea is that if you expose a rat to something for half its lifespan, that's like exposing a human to it for half their lifespan. The proportionality is tied up with all kinds of things like how the animal's cellular repair mechanisms work, how fast their cells divide, etc.
It sounds like the problem was the CIA was sloppy and made all their web pages similar enough they were easy to Google. Iran and China then rounded up everyone frequenting those pages.
Why don't they just use public pages? The internet has no shortage of discussion forums, many of which must be frequented by millions of people, even in Iran and China.
Slashdot Mirror
Itâ(TM)s been done actually. I canâ(TM)t rmemeber whether it was a rat or a mouse. Also, a cockroach and at least one kind of worm.
One should be careful about absolute statements....
There are some very fascinating results coming out of ai research of all types. A project that I believe is related to this one simulated a bunch of neurons and cortical structure and discovered that the system spontaneously produced an oscillating pattern that was similar to whatâ(TM)s observed in the actual cortex.
At the other end of the complexity spectrum, if you randomly create small convolutional kernels, most of them will be sensitive to edges.
âoeSoftwareâ may well be able to arise spontaneously given the right basic structure, input, and feedback.
I personally suspect that spiking is the easiest way nature could figure out how to make a biological system produce a variable electrical output, and that a continuous variable magnitude output will work just fine (like PWM versus variable voltage/current). Others are convinced that thereâ(TM)s something very important about spiking itself. Weâ(TM)ll see.
The answer is, we donâ(TM)t know. The necessary functions of a neuron might be more complex than a cpu, or they might not be. Itâ(TM)s an interesting question what elements are fundamentally required.
Arguing that anything less than a quantum simulation of every atom in every neuron is insufficient is just as silly as arguing that relu+madd is certainly all you need. As is everything in between.
We need research in many different directions to identify what features of the brain and its components are actually fundamental and which are irrelevant.
Even people who realise they change often donâ(TM)t realise how fast they can change. I have colleagues who have looked at exercise in rats that increases the volume of certain brain structures. That happens over very short time scales... hours. Turns out itâ(TM)s actually mostly due to dendritic growth.
Someone is unfamiliar with how much they pay grad students....
"Researchers that the charity funds could still publish in subscription journals, says Robert Kiley, Wellcome's head of open research. But only if those journals agree that the authors can immediately deposit their accepted manuscript in the PubMed Central repository under a liberal publishing licence. Some publishers, such as the Royal Society in London, already allow this."
Elsevier already allows this.
"The R of R^2 is the correlation coefficient, and R^2 can pretty much be read as a mathematical representation of the correlation of your line to the data--the higher it is, the more likely your line actually has a strong relationship with the data points."
Sorry, this is absolutely wrong. The R/R^2 describes how much of the variance in one *measurement* can be explained by another measurement. It doesn't say anything about a relationship beyond that. If your measurements are crappy, your R will be low, regardless of the strength of the actual relationship you're trying to measure. The R^2 says *nothing* about the likelihood of the relationship existing or not. That's the p-value, which is just as "mathematical" as R.
The web page you linked as an example of spurious correlations has examples of comparisons with all high to very high R values that are clearly not real correlations.
The value *you* seem to be referring to is alpha, which is the a priori *threshold* for p that you choose to accept as the definition for "significant." This is usually chosen based on norms in the field, which are generally not arbitrary, but have evolved in response to the differential cost of type I and II errors.
People misunderstand what correlation means. A correlation is a real, causal connection between two things (although it does not specify the type of causality). You may have *data* that erroneously shows a correlation. The chances of that happening are quantified by the p-value, *not* R. R can be low for all sorts of reasons, including that your measure(s) are noisy. Surveys and psychological instruments are generally highly noisy. R^2 is a measure of the strength of the correlation as captured by your methods and indicated by your data but says nothing about the likelihood of a false positive.
The website you reference is highly entertaining, and also statistically terrible. They don't give p-values for their purported spurious correlations and looking at them, most have very, very few datapoints and so may not be significant even in isolation. To say nothing of the fact that they've done many, many, many comparisons to find their material, which means they should be correcting the p-values they neglect to show. Note that the R values for all the comparisons on that page are very high.
Others, like the first one on the page, a correlation between US spending on science and suicide by hanging and suffocation is something you'd expect: the two things are likely both related to population, i.e. a type 3 causal relationship.
I think an interesting study might be to look at whether people who score a little bit higher on sadism and also on openness (i.e. those who like the taste of coffee) are just people who are more likely to tell you the truth about that ugly sweater. Conversely, the ones who put sugar in their coffee, score lower on sadism and lower on openness, are the ones who'll tell you whatever you want to hear to spare your feelings.
Well... real correlations with crappy R2 do mean that there's some connection between the two things. Which *might* be interesting, depending. Their highlight "Results suggest close relationship between the gustatory system and personality" isn't supported by the results and should have gotten shot down in flames by the reviewers, and there's really no justification at all why anybody should care about some barely there relationship between taste preference and personality.
I did find it interesting in their literature review that apparently in other studies preference for sweetness indeed correlated with lower sadism, but also lower openness. So those sugar lovers don't take pleasure in your pain, but they're also lying shits.
Even more important than that is to make the coffee properly.
People who are used to filtered coffee swill that's sat and stewed on a burner for a while are always surprised when I give them some black coffee from a french press.
I prefer to grind them up and snort them. Hits harder.
They mention previous studies where bitter taste preference was associated with more openness. So exhibitionism. Check.
Is that their whole models have R^2 somewhere between 0.01 and 0.06. That's for all the tastes and the higher end of those correlations is for all the dark personality traits too.
Oh, and liking salty things was a stronger predictor of sadism than bitter was.
Hey, come on, an R^2 of 0.02 is really strong correlation!
I guess you're probably American? The US was founded by a bunch of rugged individualists with a strong distrust of government. That distrust seems to have persisted. Reality is a bit more complicated. To start with, funding decisions for scientific projects generally aren't made by bureaucrats. They're made by scientists. Tech billionaire projects? Well, it's not like tech billionaires don't have a well deserved reputation for thinking their success in one field translates to expertise in other fields, do they?
Commercial database "programmers" are the worst kind. Well, maybe tied with web developers.
You'd think you'd have to try to screw up and lose data with something like DST. Timestamps don't change.
Absolutely true. It ups the variance considerably. Don't forget about the downside though (see for example story today about crazy bitcoin dude buying up Nevada desert to build paradise). So not a great benchmark to base feasibility on.
Governments fund universities, and quite a bit of private sector research as well. The invention of the transistor at Bell Labs is used as one of the prime examples of private technology development, but even then Shockley did a lot of work on radar for the US government during the war (as did Bell itself), some of which was actually at Columbia University.
Small nitpick. The basic physics says big is better. I saw a graph once that plotted various fusion experiments against the theoretical size vs. efficiency line, and it looked like the theory is pretty close so far. There's always the possibility that there might be serious hidden problems at larger scale though.
That right there is the problem. "Billionaires don't actually have enough money to meet the projection, but assuming their money is two or three times better than anyone else's money, fusion should happen in ten years or it's bunk."
You mean Reddit isn't an MMO?
The general idea is that if you expose a rat to something for half its lifespan, that's like exposing a human to it for half their lifespan. The proportionality is tied up with all kinds of things like how the animal's cellular repair mechanisms work, how fast their cells divide, etc.
It sounds like the problem was the CIA was sloppy and made all their web pages similar enough they were easy to Google. Iran and China then rounded up everyone frequenting those pages.
Why don't they just use public pages? The internet has no shortage of discussion forums, many of which must be frequented by millions of people, even in Iran and China.