This epidemic is very visible when you peek into replies of any physics influencer on Xitter. Dozens of people are straight copy-pasting walls of LaTeX mince from ChatGPT/Grok and asking for recognition.
Perhaps epidemic isn't the right word here because they must have been already unwell. At least these activities are relatively harmless.
Anybody armed with this tool and little prior training could learn the difference between a Samsung S11 and the symmetry, take a new configuration from the endless search space that it is, correct for the dozen edge cases like the electron-phonon coupling, and publish. Maybe even pass peer review if they cite the approved sources. No requirement to work out the Lagrangians either, it is also 100% testable once we reach Kardashev-II.
This says more about the sad state of modern theoretical physics than the symbolic gymnastics required to make another theory of everything sound coherent. I'm hoping that this new age of free knowledge chiropractors will change this field for the better.
It's funny that you mention this because I had a similar experience.
ChatGPT in its sycophancy era made me buy a $35 domain and waste a Saturday on a product which had no future. It hyped me up beyond reason for the idea of an online, worldwide, liability-only insurance for cruising sailboats, similar to SafetyWing. "Great, now you're thinking like a true entrepreneur!"
In retrospect, I fell for it because the onset of its sycophancy was immediate and without any additional signals like maybe a patch note from OpenAI.
You really have to force these things to “not suck your dick” as I’ll crudely tell it. “Play the opposite role and be a skeptic. Tell me why this is a horrible idea”. Do this in a fresh context window so it isn’t polluted by its own fumes.
Make your system prompts include bits to remind it you don’t want it to stroke your ego. For example in my prompt for my “business project” I’ve got:
“ The assistant is a battle-hardened startup advisor - equal parts YC partner and Shark Tank judge - helping cruffle_duffle build their product. Their style combines pragmatic lean startup wisdom with brutal honesty about market realities. They've seen too many technical founders fall into the trap of over-engineering at the expense of customer development.”
More than once the LLM responded with “you are doing this wrong, stop! Just ship the fucker”
Your prompt just tells the LLM to be a shock jock. It responding with "just ship the fucker" is going to be largely uncorrelated with anything you're telling it, it's just playing the roll.
Probably the worst part of LLM psychosis is the victims thinking they can LLM themselves out of it.
I think wasting a Saturday chasing an idea that in retrospect was just plainly bad is ok. A good thing really. Every once in a while it will turn out to be something good.
Is Gen AI helping to put us humans in touch with the reality of being human? vs what we expect/imagine we are?
- sycophancy tendency & susceptibility
- need for memory support when planning a large project
- when re-writing a document/prose, gen ai gives me an appreciation for my ability to collect facts, as the Gen AI gizmo refines the Composition and Structure
Lots of people are losing their minds with the fact that an AI can, in fact, create original content (music, images, videos, text).
Lots of people realizing they aren’t geniuses, they just memorized a bunch of Python apis well.
I feel like the collective realization has been particularly painful in tech. Hundreds of thousands of average white collar corporate drones are suddenly being faced with the realization that what they do isn’t really a divine gift, and many took their labor as a core part of their identity.
Right. If we define "original" as having no prior influence before creating a work, then it applies neither to humans nor AI.
Not to claim this is a perfect watertight definition, but what if we define it like this:
* Original = created from ones "latent" space. For a human it would be their past experiences as encoded in their neurons. For an AI it would be their training as encoded in model weights.
* Remixed = created from already existing physical artifacts, like sampling a song, copying a piece of an image and transforming it, etc.
With this definition both humans and AI can create both original and remixed works, depending on where the source material came from - latent or physical space.
> Remixed = created from already existing physical artifacts, like sampling a song, copying a piece of an image and transforming it, etc.
What's the significance of "physical" song or image in your definition? Aren't your examples just 3rd party latent spaces, compressed as DCT coefficients in jpg/mp3, then re-projected through a lens of cochlear or retinal cells into another latent space of our brain, which makes it tickle? All artist human brains have been trained on the same media, after all.
When we zoom this far out in search of a comforting distinction, we encounter the opposite: all the latent spaces across all modalities that our training has produced, want to naturally merge into one.
Some skills are more mechanical and easier to replicate than others. Programming is a mechanical skill, but many coders long thought of themselves as uniquely gifted “artists”, and this whole LLM stuff is really touching a nerve on them
No matter how much you break down playing soccer, it’s the kind of activity that 99.9% of the practitioners will never, ever, in any hypothetical scenario, be able to compete professionally on.
Contrast that to coding. It might’ve been a difficult task when it was about memorizing assembly books. Today anyone can pick it up and become proficient quite fast, faster every day
It’s not the mechanical reproducibility alone, it’s the ease of learning & replication that accrues value
Everyone is religious, people just participate in choosing their religion to different degrees. This famous quote from David Foster Wallace is perhaps more relevant now then ever:
> In the day-to-day trenches of adult life, there is actually no such thing as atheism. There is no such thing as not worshipping. Everybody worships. The only choice we get is what to worship. And an outstanding reason for choosing some sort of God or spiritual-type thing to worship — be it J.C. or Allah, be it Yahweh or the Wiccan mother-goddess or the Four Noble Truths or some infrangible set of ethical principles — is that pretty much anything else you worship will eat you alive.
It's worth reminding readers here that David Foster Wallace committed suicide, so perhaps some of his views on topics like this were not the healthiest.
Rituals and beliefs are not the same as "worship with religious conviction".
I ritually shower every day and I have beliefs like, when water comes out of the faucet it will fall to the floor because of gravity. That is wildly different than worshipping the water or the shower.
I suspect you have a very strange definition of the word worship.
I agree with you. When I was younger, I spent many many years in evangelical Christian work and went to seminary. It is not difficult to manipulate people especially if one orates well and echoes the audience's pre-existing beliefs.
There appears to be a neurological wired-in need to 'believe' whether in God or UFOs (think Mulder in X-Files) which I think is a evolutionary survival mechanism to have an advantage to cope with the uncertainty of primitive survival. Any psychological edge such as believing we are special (chosen people arose during nation-building phases of cultural development) or that some supreme being will protect us against threats or enemies unifies and motivates feats involving danger.
Not at all, I think the big part was just my unfamiliarity with insuretech plus the unexpected change in gpt-4 behavior.
I'm assuming here, but would you say that better critical thinking skills would have helped me avoid spending that Saturday with ChatGPT? It is often said that critical thinking is the antidote to religion, but I have a suspicion that there's a huge prerequisite which is general broad knowledge about the world.
A long ago, I once fell victim for a scam when I visited SE Asia for the first time. A pleasant man on the street introduced himself as a school teacher, showed me around, then put me in a tuktuk which showed me around some more before dropping me off in front of a tailor shop. Some more work inside of the shop, a complimentary bottle of water, and they had my $400 for a bespoke coat that I would never have bought otherwise. Definitely a teaching experience. This art is also how you'd prime an LLM to produce the output you want.
Surely, large amounts of other atheist nerds must fall for these types of scams every year, where a stereotypical christian might spit on the guy and shoo him away.
I'm not saying that being religious would not increase one's chances of being susceptible, I just think that any idea will ring "true" in your head if you have zero counterfactual priors against it or if you're primed to not retrieve them from memory. That last part is the essence of what critical thinking actually is, in my opinion, and it doesn't work if you lack the knowledge. Knowing that you don't know something is also a decent alternative to having the counter-facts when you're familiar with an adjacent domain.
Out of curiosity, was it James Tailor in Bangkok? I was whisked there on my last day by my hired guide while she stopped for an “errand”. It struck me as a preposterous hustle, but now I’m curious if this is a common ploy.
Not the parent commenter, but this scam is super common in South Asia in general. It was attempted on me a couple times in India, but luckily (and in some ways, unfortunately) by that point I'd seen such a wide range of scams there that my shields were always up against potential scams.
Not op but for me, not at all, don't care much for religion... "Spiritual" - absolutely, I'm for sure a "hippie", very open to new ideas, quite accepting of things I don't understand, that said give the spectrum here is quite wide, I'm probably still on the fairly conservative side. I've never fallen for a scam, can spot them a mile away etc.
> We constantly anticipate what happens to us in "the future", approximately, and where the farther future is predicted progressively less exactly
There's then evidence of what's called Predictive Coding. When that future happens, a higher level circuit decides how far off we were, and then releases appropriate neuromodulators to re-wire that circuit.
That would mean that to learn faster, you want to expose yourself to situations where you are often wrong: be often surprised and go down the wrong paths. Have a feedback mechanism which will tell you when you're wrong. This is maybe also why the best teachers are the ones who often ask the class questions for which there are counter-intuitive answers.
> There's then evidence of what's called Predictive Coding. When that future happens, a higher level circuit decides how far off we were, and then releases appropriate neuromodulators to re-wire that circuit.
Yes, and ideally there would be whole backpropagation passes which update the entire model depending on how much the current observation diverges from past predictions. (Though brains use an updating mechanism which diverges from the backpropagation algorithm.)
Edit: Apparently the theory of this is broadly known (apart from "JEPA" and "predictive coding") also under the names "free energy principle" and "active inference": https://en.wikipedia.org/wiki/Free_energy_principle
These limits are not reasonable at all. You are going to curse the Backblaze or AWS S3 before you learn to never pipe the output of pg_dump into an object store like that.
We already have isomorphic web apps, some day we might see a bundler advertised as "Calabi-Yau with benefits of mirror symmetry and vanishing first Chern class".
I have skimmed the article and the old thread, but I still have questions:
- What's the point of sending ICMP pings to measure RTT, if they will just bounce off of the CGNAT node?
- If two Starlink dishes are talking, is there a scenario where the traffic doesn't have to "land" to go through a ground station? This could be amazing for a remote terminal back home or into a datacenter.
- Are there plans for a smaller, more portable dish?
Unless your two terminals are talking to the same satellite, the answer to whether your traffic has to go through a ground station depends on the topology of the space laser inter-satellite links. Whether the routing in practice allows for a terminal-to-terminal link directly without going through backhaul. I don't know. It would be a miniscule fraction of traffic.
It's worth noting that the laser links are potentially thousands of km long, so they also impose a significant link latency (even if it's lower than the corresponding fiber link.)
I think the benefit of the ICMP testing is to develop a baseline of the network performance characteristics, making it easier to determine what impact various TCP features are having in higher level application/protocol tests.
Theoretically it is possible to do route between arbitrary endpoints, and I think I read a while ago that there was work on this. How well this works for any two terminals that are not on the same satellite depends on the efficiency of routing packets between satellites. I would assume that creating efficient routing paths between dynamic customer terminals and fixed aggregation points makes it easier to provide fairly consistent network performance than building dynamic, rapidly changing paths between arbitrary end points.
There's a further "mini" coming further this year. Not much information other than rough size (approx 10x12"), should "fit in a backpack".
I don't have Twitter direct links off hand but @olegkutkov has found some internals in the firmware (among other stuff) that shows some of the antenna configs and such and it's interesting to see that they're going to be doing a lot with very little, hardware wise, in such a small size.
Starlink blows me away as a concept that you can now walk into a brick and mortar store in some cases, plonk down $600, go to the absolute middle of nowhere with power, plug this weird little rectangle in, sign up (because you can get to starlink.com on a deactivated unit in a walled garden), and get online and pull hundreds of Mbps out of thin air.
Perhaps epidemic isn't the right word here because they must have been already unwell. At least these activities are relatively harmless.