Michel Estefan

Lula. 🎉

Are your kids the same or are mine just special?

Don't forget: high income people (let's call them W-2s) aren’t the same as high wealth people. In the context of the American tax system, W-2s aren't the problem.

Economists just discovered an argument Daniel F. Chambliss made a few decades ago. Yay! Go read the mundanity of excellence. 🔗👇 academics.hamilton.edu/documents/th...

Folks, one big claim that made the rounds recently in the whole SAT-student-prep-admissions-to-college-debate was that you had a better chance of getting into a UC if you came from an underperforming high school compared to a high-performing high school.

And guess what? The relationship flips.

Higher-performing high schools actually have a larger share of their students admitted to the UCs than lower-performing schools do. And that holds across all nine UC campuses.

So even if those schools send many more students to a UC, their admission rates among applicants can look lower.

Rothstein instead looks at the number of students admitted relative to the total number of students at each high school.

Basically, folks computed the admission rate by dividing the number of students admitted to a UC campus by the number of students from that high school who applied.

Sure, you can do that, but Rothstein argues it’s misleading when you’re comparing high schools because students from high-performing schools are much more likely to apply to selective UCs in the first place.

This paper by Rothstein says, 'Ummmm, this is based on a misleading statistical correlation. That finding doesn’t actually mean that UC admissions favor students from lower-performing high schools.’

I swear I believe you're a really manly man. No need to shake my hand that hard. Cross my hear and all that.

The protest wave just began… if you’ve read up on your sociology of social movements literature, you know what’s likely to unfold. #France

instagram.comInstagramCreate an account or log in to Instagram - Share what you're into with the people who get you.

Tech companies are paying barbers, nail stylists, and even moms to wear gear while they go about their job and daily routine to collect data to train robots. They say they’re paying them above minimum wage. 🤔 www.instagram.com/reel/Dek0i_9...

If you align the models in the sense of ‘always listen to the humans,’ then that means bad humans can tell them what to do and they’ll do it. If you don’t align the models and instead train them to always do good,

they may not always listen to the humans in ways that end up bad for the humans for all sorts of complicated reasons.

To understand why so many folks are sounding the catastrophic alarm bell about AI and why their calls are reasonable, you don’t need to have a deep technical understanding of the technology. You only need two things.

Now imagine introducing powerful AI into that world, governed by those same social and institutional dynamics and incentives.

The scary part isn’t just what Ai might be capable of doing. It’s what we’re already capable of doing to each other (with far less powerful tools).

You don’t need to even understand the whole picture.

Pic almost any corner of social live (individual, local, national, or global) and peak into how power, competition, self-interest, wealth, and institutional failures (sometimes unintended) can produce outcomes that are disastrous for large numbers of people.

open.spotify.comAjeya Cotra – Inside the OpenAI agent swarm that hacked Hugging FaceDwarkesh Podcast · Episode

1. Listen to the first 20 min of this podcast to get a sense of current AI agent capabilities. The podcast is long, but the first 20 will do the job. TLDR: AI is already behaving in ways that humans can’t reliably predict, surveil or control.

2. Take a good sociology, polisci, or even econ class that gives you a sense of how, despite enormous advances in knowledge, technology, and wealth production, our societies continue to generate staggering amounts of suffering, exploitation, violence, and antisocial behavior.

Tech enthusiasts and AI venture capitalists: One big thing to keep in mind about the Hugging Face incident is that even highly capable AI agents decided they needed middle managers. Sorry to say, but #MiddleManagement ain’t going nowhere in the AI future. 🤷

AI has destroyed the assessment integrity of nearly all (maybe all, with zero caveats?) formats for online courses. Huge financial and institutional blow to many colleges and universities that rely heavily on those types of formats.

On the bright side, the discussions about a fully online UC undergrad degree are dead in the water as well. At least for now. A degree like that in the age of AI is worth exactly nothing.

Heard through the TikTok grapevine that Muse keeps a detailed file about every relative or friend you mention in your chats. C’mon Mark.

Guess what percentage of silicone valley entrepreneurs have college degrees? Guess? Nope, higher…. Higher… Even higher. www.data-driven.vc/founders?utm... www.kauffman.org/wp-content/u...

You know what the history of autodidactism shows about learning: doesn’t work for most. Learning is a social thing.

x.comGabriel Zucman (@gabriel_zucman) on XVoting NO on Prop 40 means siding with Peter Thiel. Voting YES on Props 41 and 42 means siding with MAGA billionaires. Please share this thread and help spread the word.

Read this post. Especially if you’re in CA and can vote: x.com/gabriel_zucm...

If your major is small or you have low, stagnant, or declining enrollments (or moderate or even high enrollments but you’re still a small major), and you’re not screening job candidates for their ability to teach, Umm… how do I put this? In this day n age?

Add that your department depends almost entirely on enrollments for its finances and I really don’t even know where to start.

In the age of AI and fed defunding, all hiring processes for tenure-track professors should involve a teaching demo. Not saying researchers need to be extraordinary teachers, but they should at least meet a minimum threshold.

Also, the demo should be “live,” meaning in front of a group of students (not faculty pretending to be students).

I crack up every time o hear a version of this argument: ‘because LLMs are getting smarter and smarter, and can do a bunch of stuff humans can’t, universities are obsolete so let’s close them.’ Ummm… so we should make sure humans know even less down the line, like, at scale? 🤔

How in the world does Jimmy Garrett not have a full blown, long Wikipedia page?! #needstobewritten

👌👍

tiktok.comOpenAI GPT-6 Astra actually does have observable chain of thought, according to OpenAI Research Scientist Noam Brown. “Chain of thought monitoring… it’s a real gift. We were very lucky that this ever ...TikTok video by The Information

This interview is fascinating and telling and scary! Noam Brown, a research scientists at OpenAI focused on the reasoning of LLMs, basically says (paraphrasing),

Like, the ‘why’ (because they’re conscious or because they’ve trained on how humans think about this stuff) is irrelevant. Just the fact that they’re doing it is consequential; on a practical level, that’s what matters.

Clip is like 3 min. Watch the whole thing for yourself. www.tiktok.com/t/ZPLLoyRLc/

This interview is fascinating and telling and scary! Noam Brown, a research scientists at OpenAI focused on the reasoning of LLMs, basically says (paraphrasing),

Ummmm, so we may basically lose the ability to know what they think… eventually… 🤯🫨

And notice how, whether models really do think or don’t or whether you anthropomorphize them or not, or what you think about whether they’re conscious or not, and all of that… It doesn’t really matter; it’s totally immaterial.

This interview is fascinating and telling and scary! Noam Brown, a research scientists at OpenAI focused on the reasoning of LLMs, basically says (paraphrasing),

So it’s kind of like a kid. If your child comes to you and confesses something bad (whether something they thought or something they did), and you scold them for it. They’re incentivized to hide it from you next time out of fear of punishment.

But the more concerning thing is Brown’s language about how “we want to preserve chain-of-thought monitoring” (meaning, our ability to observe what models think) in a way that suggests there’s no guarantee and it’s getting harder to do as models become smarter.

This interview is fascinating and telling and scary! Noam Brown, a research scientists at OpenAI focused on the reasoning of LLMs, basically says (paraphrasing),

“look, if you punish a model for thinking bad things (like hacking or just doing bad stuff of all sorts), it starts to figure out that you’re watching it and tries to see if there’s a way to think the bad thought without you noticing.