It's background education in the basics so you can understand what drug addiction is and the neurological differences in the active populations for wanting versus liking. I guess I can spell it out.
Addictive drugs directly increase wanting via directly activating the downstream targets of dopaminergic populations which predict the valence of stimuli and control of wanting and motivation. By taking a chemically addictive drug you don't even have to enjoy the stimuli related to it. You will still be conditioned to want it and be motivated to re-experience the stimuli surrounding it.
This is vastly different in mechanism and result than simply seeing or hearing a screen. These things cannot directly increase incentive salience regardless of actual valance of the stimuli. You have to actually enjoy the thing and the experiences to form habits.
Do you see the difference now? One thing, the chemical drugs, are addictive. The other things are enjoyable. One will addict everyone because they're addictive. The other only leads to addiction-like behaviors in the context of say, random interval operant conditioning, if you actually enjoy the thing intrinsically first and are of the fairly small subset of that subset that is predisposed to behavioral addictive behaviors.
You're right in an important sense. There's not a complete difference in outcome between direct manipulation of wanting with drugs and using enjoyable stimuli in some form of unethical non-consensual conditioning program (aka advertising). It is one of many scales of magnitude and a lot of abstraction but that's still bad.
What I am trying to get across, and what I'd hoped all the conditionals and premises I laid out in my original comment made clear, is an additional consideration:
Screens are not drugs. They are not somehow uniquely and magically addictive (like drugs actually are). The multi-media is not the problem and not the device to be regulated. The corporate structure and motivations are the problem. This issue literally applies to any possible human perception even outside of screens. Sport fishing itself is random interval operant conditioning in the same way that corporations use. And frankly, with a boat, it's just as big of a money and time sink.
We should not be making laws regulating screens themselves because we think screens are more addictive than, say, an enjoyable day out on the lake. They're not. You could condition a blind person over the radio with just audio. The radio is not the problem and radios are not uniquely addictive like drugs.
I am saying it's important not to think of screens as the problem. The problem is the corporations' behavior and scale. That's a big difference in terms of the legislation used to mitigate the problem and the people to which it will apply. The Digital Markets Act in the EU is a good template to follow with it only applying to very large incorporated entities acting as gatekeepers.
I mean, yes, trivially? That only hinges on two factors: what share of a fast food business' overall expenses actually go to labor costs, and, y'know, how much extra demand is enabled by ensuring even the poorest workers make enough to afford fast food once in a while.
Takes like yours used to baffle me, until I realized that the US was founded on enslaved labor and to this day there remains a silent expectation in some circles that there must be a laborer class which should be as inexpensive and disposable as possible, and is fundamentally distinct from the consumer class. A lot becomes clearer all at once when you realize that to some, there's a whole segment of the population that is not expected to benefit from the economy, only serve it.
Historically, such worldviews have in the long term tended to bring sharp misfortune to those holding them. I'm hoping for a better outcome here, though.
Fast food workers are included in the consumer class.
As for slavery, the poorly educated believe that it was a uniquely American phenomenon. Slavery was a global institution practiced by every civilization, nation, and culture on earth. In fact, it’s still alive and well in multiple places. The US abolished it fully in 1865. Products produced by slaves accounted for around 15% of our GDP at its peak.
GLM-4.7 (specifically this version) repeats the guardrail prompt injections from 3.0 Pro, word-by-word, and never follows them, which is consistent with training on a reward-hacked CoT. Gemini 3.0 only discusses snippets from this injection in its native CoT (hidden by default, trivial to uncover), but GLM-4.7 was able to reconstruct it in full during training. The only possible reason for this is direct training on a large amount of examples of Gemini's CoT. Its structure and a lot of replies were identical in GLM too.
Gemini 2.0 Exp 1206 was reported to be indirectly trained on Claude's outputs with humans in between [1], which was pretty consistent with its outputs at the time. No other Gemini versions except two experimental ones were similar to Claude.
This absolutely fits my observations and it's got to be one of my favorite secret things about the industry. More generally, the higher the skill level at a given org, the more trans furries you'll find, it seems like. There was a time you couldn't throw a stuffed fox across a Google SRE office without hitting one.
I wonder if this holds well enough that you can use it as a proxy metric to assess the technical chops at a new company.
The specifics depend on your country, but your union should have labor law attorneys who can tell you what to do to maximally discourage holding your diagnosis and its consequences against you. In most civilized countries, discrimination based on medical conditions, or even the appearance of it, can get an employer in uncomfortable waters.
Meanwhile, do shit that you love. Try new shit that you may end up loving too. Learn a new music instrument and suck at it with great joy. Go to a furry convention. Learn how to make pizza! Look, no one knows how long we've got, and the most anyone gets, goes by in an eyeblink anyway. But you can do more, or less living during that time, mostly regardless of how long it lasts, and all told I'd recommend doing more living because this is the one shot you get at it anyway.
I wish you as much luck with your tumor as I had with mine, and enduring joy every day no matter what.
The idea that there may be a cognitive lingua franca hiding in the layers is fascinating and gives me hope for a neat idea: pluggable knowledge banks.
MoE notwithstanding, a model trained on the whole Internet and a few hundred thousands stolen books carries way more knowledge than is actually needed for any given workflow. It would be great if we could ship slimmed down models into which we'd plug the knowledge banks useful for today's work, and only those.
It would also mean that you could keep a model's knowledge fresh without retraining the whole of it.
Rare earth minerals are not consumed in the process of generating solar energy, whereas once you've burned your oil to generate energy, it's gone and you need to buy more. That makes a pretty damn major difference.
reply