Author’s central point is that an LLM answer “is optimized for arrival, not for becoming” (to paraphrase from the Google “Lucky” part).
So a reasoning LLM that does the comparisons and checks “like a human” still fails the author’s test.
That said, this still feels like a skill issue. If you want to learn, see opposing views gather evidence to form your own opinions about, LLMs can still help massively. You just have to treat them research assistants instead of answer providers.
Imagine you are a salesperson for the hardware division of such a company. Imagine you did your work well through the entire year. How would you answer your question?
That's with a naive stereo split. Many would still put the bass on one side, with the binaural processing so it's still heard on the right, but quieter and with a tiny delay.
Hard panning isn't naive. It's just a choice that presumes an audio playback environment.
If you're listening in a room with two speakers, having widely panned sounds and limited use of reverb sounds great. The room will mix the two speakers somewhat together and add a sense of space. The result sounds like a couple of instruments playing in a room, which is sort of is.
But if you're listening with a tiny speaker directly next to each ear canal, then all of that mixing and creating a sense of space must be baked into the two audio channels themselves. You have to be more judicious with panning to avoid creating an effect that couldn't possibly be heard in a real space and add some more reverb to create a spatial environment.
Maybe I'm misunderstanding him but I think he says the music track can have hard panning, and it's the headphone playback system that should do some compensatory processing so that it sounds as if it was played on two speakers in a room.
Don't ask me how it works but I know gaming headsets try to emulate a surround setup.
Yes, these sorts of compensation features have become common on higher end headphones.
One example:
> The crossfeed feature is great for classic tracks with hard-panned mixes. It takes instruments concentrated on one channel and balances them out, creating a much more natural listening experience — like hearing the track on a full stereo system.
Your comment wasn’t wrong. Neither is the reply wrong to be frustrated about how the world understands this complex topic.
You’re talking about autism. The reply is about autism spectrum DISORDER.
Different things, exacerbated by the imprecise and evolving language we use to describe current understanding.
An individual can absolutely exhibit autistic traits, whilst also not meeting the diagnostic criteria for the disorder.
And autistic traits are absolutely a variant of normalcy. When you combine many together, and it affects you in a strongly negative way, now you meet ASD criteria.
I think a very useful term here is Broad Autism Phenotype (BAP) - subclinical ASD - you have significantly more of the traits than the average person does, but they are not strong enough or not disabling enough to merit a clinical diagnosis of ASD.
BAP is very common among (1) STEM professionals, (2) close blood relatives of people with clinical ASD (if you have a child or sibling with an ASD diagnosis, then if you yourself don’t have ASD, odds are high you have some degree of BAP), (3) people with other psychiatric diagnoses (especially those known to have a lot of overlap with ASD, e.g. ADHD, personality disorders, PTSD, OCD, eating disorders, the schizophrenia spectrum), (4) certain LGBT subgroups (especially transgender people) - all of whom have heightened odds not just of having BAP / subclinical ASD, but clinical ASD too
Like ASD, BAP skews male, but women can have it too. (The average man is a little bit more autistic than the average woman.) Also, autistic traits are positively correlated between romantic partners, so a woman in a relationship with a man with BAP or ASD is more likely to have some degree of BAP herself (as well as being more likely to have clinical ASD)
BAP itself is a matter of degree… autistic traits is a continuum and we are all somewhere on it (actually a one-dimensional continuum is a simplification, it is a multidimensional construct-but a useful simplification) - and clinicians draw a line at some point (they don’t all draw it at the same place, and its location varies across time and space and culture and even clinical subcultures) and if you are on one side of that line you have clinical ASD, if you are on the other you don’t-if you are on the non-clinical side of the line, but nearing it, you have BAP… but “nearing” it subdivides into people who are closer and people who are further away
I'm pretty sure it was that series that also described https://en.wikipedia.org/wiki/Liquid_droplet_radiator , with the side effects of different ships having very distinct heat patterns because of their radiator patterns. And that if a ship ever had to make a turn while they were active, big glowing arcs of slowly-cooling droplets would be flung out into space and leave a kind of heat plume.
So a reasoning LLM that does the comparisons and checks “like a human” still fails the author’s test.
That said, this still feels like a skill issue. If you want to learn, see opposing views gather evidence to form your own opinions about, LLMs can still help massively. You just have to treat them research assistants instead of answer providers.
reply