More like: the experiment's been tried and kept in other places. Michigan's not doing something unprecedented (nor are they any slouch of a program). If it was that bad, you'd expect the folks who've done it to see some poor performance from it. Instead, we see two of the most elite engineering schools (Caltech and MIT) both known for their extremely demanding curriculum, who decide that removing that grade stress on freshman improves the end result. So, more "If the experiment was bad, you'd expect that the programs who've implemented it would have trouble with it" and it doesn't seem like they're suffering from it.

Ask MIT. Been their standard for years.

Obligatory *re*mention: MIT and others already do this. It's deeply disingenuous to act like this is some sort of wacky experiment that nobody with any respect for their students' capability would never consider. If anything, giving students the permission to actually take risks and fail lets you push them harder during classes. If they're spending all their time grade-scumming and afraid to take material they're actually interested in, hone in on their major, etc. they're really not learning that much. The grades provide a false "rigor" for folks on the outside that like to assume that education needs some standard of "toughness" that is best reflected in grades. Frankly, what that results in is straight-A students who are risk-averse glass cannons that crumble on actual project work. When I was in graduate school, the absolute toughest class I had refused to discuss grades. The agreement was that we'd all plow face-first into the bird-unsafe-glass window of the topic and so long as we failed spectacularly. The topic was phenomenally hard and I learned so much, but I still agree with the professor that focusing on the grade would only have held me back in actually learning the material.

With the scale of how much testing is disrupting my kid's education, I'm having a hard time taking any research seriously that thinks that the key problem is that we don't test kids enough. We have all these stressors, insecurity, and funding issues and the main concern is that we're not measuring them enough? What would increased 'accountability' do if we systematically won't fund and support education, itself?

...but I'm doing my best to keep you as far away from the actual controls as possible.

"Meta Research is now recommending that meta-Zuck be given signature authority in management decisions, this following an analysis that showed, surprisingly, that meta-Zuck is not only better-regarded than his original, but is known for actually completely reading and considering the reports he's given." "Staff are now reporting that they've had zero incidents stemming from a random wild hare that Zuck was known for chasing, and that project prioritization discussions are much more consistent." "At least we're going to be able to avoid another Metaverse paradigm shift."

They're also not considering that forgetting may also be computationally beneficially for AIs.

Just because humans are capable of symbol processing doesn't mean that we are, at root, symbol processors.

Everyone's stuck between rising prices and Apple suddenly competing on entry-level options. That's going to take a bite out of everyone simply for *existing* because the "not having Apple money" is no longer an excuse. Even if it's underpowered or has other technical limitations, they're competing at a price point where those sorts of things are common and if they can execute half-decently, they're going to be viewed more favorably and get more credit for their accomplishments in the space, even if they're only moderate. High upside for Apple on this one.

I remember seeing that Moltbook's most famous posts were faked by people, anyways.

As determined by the beliefs of the person giving the order, prior to validation by a court of law, some half decade down the road. Meanwhile, contractors have no right to set terms on how their services are used or abused, essentially removing their first amendment rights. However the Pentagon can use your tool is valid and you're obligated to support it.

...and is now enthusiastic about new, innovative methods of squeezing their client base for revenue streams.

I'd think it's more Goodhart's law ( https://en.wikipedia.org/wiki/... ) If the AI were to be able to make them more effective, that should be sufficient without measuring which tool they chose. Now they prioritize the method over the outcome.

I'm sure none of those sources needed to be consistent and complete. An AI can be handy, especially for retrieval when your parameters are a bit fuzzy for proper search, but the ability to realllllly point where you got the claim from is pretty damned important if you want to build anything reliable on it.

Also, there's no way to just jump from L2 to L4. Part of the cost and challenge for L3 involves R&D work that's critical for the success of L4. The amount of work and effort to move from L3 to L4 is potentially greater than the work to move from L2 to L3.