It's about what university you go to. Liberal arts schools may have a perfectly respectable physics program, but their whole science program may be housed under the College of X, which gives out BAs rather than BSs. The reverse also happens; had I wanted to, I could have gotten a BS in Latin from Caltech, because they only give out BS degrees, regardless of the field.

That is a pretty bizarre stance. Are you really suggesting that no one employed by an AI company actually does research? I rather wonder how new models keep improving, then. It's also odd to suggest that none of those researchers would ever talk to the press, especially "ex-"researchers who don't have to worry about getting fired.

There's some SERIOUS survivor bias in there, literally. Just because those previous "scary" technologies did not break the world does not mean that no technology ever will. This generation of AI won't, and I sincerely doubt any future evolution of AI along the current LLM model ever can. But burning hydrocarbons is and will have severe effects on the world's economy and resource politics. Massive buildup of nuclear weapons damn well could have destroyed society.

AI doesn't "understand the concept of a reward", nor does it need to. It's not functionally different than a classical nonlinear likelihood maximization. An arbitrary function is evaluated at different points in some parameter space, and the answer is the point that yields the minimum. If you want to train an AI that prioritizes correct answers over acceptable methods, you simply change how much each of those criteria is weighted in your objective function. AIs can effectively work in much higher parameter spaces than you can evaluate with classical minimizers, but that's about the only real difference.

Think about navigation. The objective function is minimum travel time between two points. Usually you can set options to avoid highways or tolls entirely. But you could just as easily have an option like "Avoid tolls limited by adding no more than 10 minutes per $1 saved". Then the objective function is just (total time + 10m*$1) and you get a different route.

10% chance, huh? When a real scientist tells you a number, it's the result of a calculation/measurement/model, not "sure, that seems about right".

You know why it's really dead? Because broadcast TV is dead. Instead of being freely available for anyone with a TV and rabbit ears to watch, the show about the perfect socialist utopia where no one even remembers what money is is locked behind a Paramount subscription. Even as we move closer to a world where human labor is obsolete, we move further from a society where anyone but the billionaire owners benefit from that technology.

If the only thing that happens is something coming along to build something better, shouldn't we now have the best possible everything? Was any "social media" platform ever better than ICQ?

Serious question: how is that stat affected by people working multiple jobs? I assume salaried "exempt" workers who put in 60+ hours/week are just counted as full time 40, but what about people working multiple crappy jobs working 60-90 per week? Overall it probably wouldn't affect things much, but I would guess once you start slicing into demographics you'd get a lot of bias in low income groups.

Invoking your second amendment rights in response to police talking to you is called suicide by cop.

Where is the "+1 Groan" option?

Except these are never going to totally meet your daytime demand (max 1.2 kW). It generates the most power when your air conditioner is working hardest (we are talking about California here). They're appealing because you don't have to fuss with a battery and inverter and all that, you take it out of the box, clamp it to the balcony railing, plug it in and done. If you really want to completely offset your total power use you get a real system.

I'm sure it's at least partly a pebkac error on my part, but all of the coding agents I've ever worked with want to write an entire new function call for every single line of code. Even when that exact function with a different name already exists, often in the same file, written by the agent in the same session. Even when they can recognize existing functions, they almost never modify existing functions to take an extra argument when that would be appropriate, but make a new one. And when a function does take multiple arguments, they want to write a wrapper function for every invocation. Instead of calling `foo(a, b)`, it's always `def foo_b(a): return foo(a, b)`. Then when that bloats to infinity, it will start forgetting entire modules exist and rewriting them. DRY is a completely alien concept; all data structures by default will be defined three times, once for the database, once for the API, and once for live data. So yeah, writing 100k LOC doesn't take long at all if you're willing to ignore all good practices and don't care that it should have been maybe 5000.

thank you for providing the clarification that I was too lazy to spend 30 seconds googling =)

An interesting read, but the authors' main argument is that, because we ingest lots of nitrates/nitrites in fruits and vegetables and there are health benefits in that context, nitrate additives should be less regulated and even considered as required nutritional intake. They imply that all nitrates consumed are equivalent, but other research does not support it. Much like fructose in fruit has significantly different response than fructose delivered in other ways, nitrate additives are different from naturally-occurring nitrates in fruits and vegetables. Oh, and the authors have financial interest in a company selling nitrate supplements, which was slapped by the FDA for making exaggerated and baseless claims of health benefits based on that paper.

That said, nitrates was probably the wrong example. A high bacon diet will kill you from heart disease before you have to worry about nitrates. Let's use the "cars" example instead. The negative mental health effects of social media are roughly as well established now as the effects of leaded gasoline were when it was banned. We now know that it was even worse than anyone suspected. I don't think "doomscrolling" per se has been isolated as problematic, but it is certainly at least related to intentionally-addictive designs that *are* causatively linked to negative mental health effects.

It needs to be made a national holiday. Just making it Saturday benefits nine-to-fivers over lower-income voters who are more likely to be working Saturdays due to multiple jobs or customer-service oriented jobs.

As for mandatory voting, I fail to see how getting more uneducated voters into voting boxes helps anything. Not uneducated in terms of schooling, but uneducated about what they're actually voting on. Anyone who is only voting because they are forced to is not going to spend the time to learn about the issues; they'll vote for the guy who bought the most billboards.