Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?
This is your opportunity to ask questions. No question too simple or too silly.
Culture war topics are accepted, and proposals for a better intro post are appreciated.
Jump in the discussion.
No email address required.
Notes -
Does anyone have their own equivalent of a personal "antimeme", a concept you familiarize yourself with (potentially with difficulty) and then inevitably forget unless you make an intentional effort to look it up?
In no particular order:
It doesn't help that sometimes they refer to x86-64 as just x86 (assumes 32-bit address space are obviously deprecated) or AMD64 (as if that tells you anything about the instruction set). I suppose this is also a product of age and the computer market you grew up in. There was a time in the US when the IBM PC running an 8086 was the personal computer, and the fact that all other 86s descend from there feels natural. There was also a time when a 64-bit CPU felt like you were living in the future, e.g. Nintendo 64.
I leave this one to the physicians. Small related story, though. In an effort to get me to stop bothering them, I once had a circuits lab TA tell me to go off and build an ECG. I did at some point succeed at "building" the world's shittiest ECG; at least it made an appropriately squiggly-looking line (relying on the oscilloscope for 98% of the work, of course). I'm pretty sure that experience has only left me more mystified about what an ECG is supposed to do.
Two useful notes here.
This happens to me all the time, which calls back to my annoyance with LLMs sometimes. I'm sure it's partially a problem of imprecise prompting.
Often I will ask: "I'm trying to recall the name for something that is like X, Y, and Z. Can you help me determine what concept I'm looking for?"
Reply: "The concept is called XYZ and it works by X, Y, and Z." Entirely a hallucination when you then go to search for XYZ.
The heart goes through sequential contraction and relaxation phases, with the upper atria and lower ventricle being out of phase. This is governed by electrical waves propagating roughly top down. Since we're talking about a chemical process (ions crossing membranes), there's noticeable conduction delay.
Roughly speaking, it kicks off near the top of the heart, and has a "highway" of rapid conduction down the middle. There's increased latency the further you go.
We place multiple electrodes on the limbs and chest:
*The leads placed on the chest measure changes in voltage propagating perpendicular to the skin (front and lateral).
You draw a chart. Leads V1 and V2 focus on the anterior-right of the heart, 3 and 4 are a bit lower and right above the heart, so you get the anterior picture, 5 and 6 show you what's going on in the sides. The limb leads help figure out the inferior bit.
Once we have established a baseline, then we look at a patient's ECG for deviations from the norm. Too much or too little voltage, or an unusual delay between phases, these can all point to cardiac pathology, and we can localize based on which views are aberrant. For example, in a heart attack, the leads reading anteriorly will, badum-tss, be the ones most out of whack if the damage is on the anterior aspect of the heart (anterior myocardium/muscles), and so on. And those delays in conduction point towards something wrong with the inbuilt cardiac pacemakers or that highway I mentioned.
In effect, an ECG isn't just a single image, it's closer to tomography. The additional leads provide clear advantages over just attaching a potentiometer to someone's toes and fingers.
Of course, it gets much more complicated in practice. Especially when a patient has multiple heart conditions at once, I start sweating when I have to interpret those even when I'm fully up to speed. And it's all the worse in psychiatry, because you can't rely on the patients to be particularly cooperative. And it hurts when you pull off the adhesive on the cups and it takes chest hair with it.
But Pagliacci, I've tried clown therapy :(
3B1B is excellent, and his video on the FT is my go to. It's just that I forget the details beyond "you can decompose arbitrary analog signals into a sum of sine waves".
Which model? Hallucinations have become quite rare on the SOTA models, especially the ones with internet search enabled. It's not like they never happen, but I'm surprised that they're happening "all the time".
Yes, this does look like what you read when you look up what it's supposed to do.
My electrodes may or may not have been bare stranded copper wire duct-taped to myself. I still can't believe what we used to get away with.
It is in fact all linear-algebra all the way down.
I don't think it's so much a problem with the power of the model, but rather my own vagueness with recollection and prompting that I get back out what I put in.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link