cs_throwaway

I wonder how many math professors think they were actively working on well-known unsolved problems. Is it just a small elite club, or is it every tenured / tenure-track professor in all R1 schools?

UBI is politically dead on arrival in the U.S.

A lot of people put a lot of their self-worth into their work and that’s a good thing that won’t change.

There is no nice way to tell someone that you’ve scooped them, and this is industrial scale scooping.

A few papers have been retracted, but it looks like many are withstanding intense scrutiny. Lean is making the results more likely to be correct, but I think making them harder to understand.

The world has changed and you’ll know a math department is making a serious attempt to adapt when it teaches a required Lean course in freshman year.

It’s just as unlikely that this will cease. hiring a professor at a top school because they were an excellent math teacher… It won’t happen.

Let us know when it is clear that UCLA does not hire the candidate with the most top-tier journal papers.

It’s more likely that instead of spending a 100K/year direct grant on two PhD students, PIs will hire 1 and have the student spend 50K on AI.

Maybe someone on the NYU team forgot to opt out of “improve the model for everyone”.

If you know how to use an agent, and I do, you can have do in minutes what would take a undergrad all summer to do.

But undergrad research has always served two roles: getting my work done and the pleasure of giving the best students the chance to go beyond the curriculum, which is designed for the average student.

I think the impact of AI on undergrad research opportunities will be zero. The bigger impact is reduced science funding. I have directed lots of PhD student grant funding to undergrads who deserved it. I don’t see myself doing that in the near future.

This is a good use of TIP division funding. Whatever they were doing before seemed pointless.

Right. Most people are deeply uninterested in learning anything, not just in K-12, but at pretty good colleges like mine.

Modern AI has made me a more productive teacher—-I produce higher quality material and have more time for research.

But the impact on most students is negative. It is another thing to engage with, which they won’t unless forced. The only way to learn is to do the work yourself. An AI tutor can get you unstuck faster, but that’s typically bad. Learning to be productively stuck on something for days without making visible progress is an important skill that most people never learn.

I have always thought of it like this: U.S. citizens have the right to marry and bring home anyone they want. It is not about the immigrant. For example, if you're stationed on a military base on Japan or Germany, you can meet a local girl, fall in love, and bring her back home.

"Chain migration" however is more questionable.

It is counter-intuitive that the more accomplished you are, the more evidence you need to provide. The part about the child not getting a naturalization certificate even though they are naturalized is very weird -- that should be fixed administratively.

Anyway, I don't think the O-1 / EB-1A is the easiest setting. An even easier setting is to become a tenure-track professor at a reasonable university in a technical field, e.g., computer science. That gives you an H1-B without any drama. An EB-1B green card requires a lot of evidence, but maybe a few pages less than an EB-1A green card.

Finally, getting citizenship is trivial. It's the green card that is hard to get.

The computer science that matters the most today —- machine learning, vision, NLP —- is open access without the fees because the main confs are not ACM. (Vision has some in IEEE.)

I guess the ACM fees are paying for stupid things like the new AI summaries.

Surprising it is necessary, given no such fees for machine learning and associated areas. (Which are all not ACM.)

More money, more income. That's why flood of foreign money is good for a university. But, it is a fallacy to think that this has no cost.

In my experience, the large influx of foreign students are typically at the masters level. MS classes are typically (not always lol!) more advanced than undergraduate classes. So, you need more qualified instructors, such as your tenured/tenure track faculty to teach them. When you take T/TT faculty out of undergraduate classes and replace them with teaching faculty, you lose a lot. (Let me know if you need what's lost to be spelled out.)

Exactly the same.

I don’t understand why we cannot just blacklist these consulting companies as a first step. The 100K fee may effectively do that so it sounds good to me.

I was not accurate in my previous post. The O-1 isn’t on an academic’s radar because they are not subject to H1B caps. So might as well do an H1B with minimal effort.

Yes. As far as I understand it, the EO is not a rule, but will impact upcoming rules. H1B already has exceptions, so there is nothing new being said. It already has a cap exception for doctors and professors, so why not for critical private sector industries.

The "kids" in the original article are likely PhD students. They are not paying tuition. You and I are likely paying for their education with taxes. (This is likely a small part of the overall higher education picture.)

I think a lot of people arguing about the H1B visa are talking past each other.

- There is no doubt a large volume of abuse by tech consulting companies. It's likely even worse than it looks, because the H1Bs in the U.S. are to support even larger teams offshore. I don't understand why we can't just blacklist these companies.

- Some of medium-skill hires, e.g., did a 2 year MS degree from random university in the U.S., are also a bit sus, in my opinion.

- I'll bet several of Zuck's recent $10M Superintelligence Team hires were at least briefly on an H1B before getting their EB-1A Green Cards.

- Same for a lot of faculty in computer science -- you can get an EB-1B Green Card quickly, but you have to spend some time on an H1B. You cannot convert directly from a student visa. The O-1 exists, but is not on most people's radars in academia. I think likely because the legal fees are prohibitively expensive. (I have heard $40K+)

Curious what this will do for faculty. Common to use H1B as a bridge for a few months before green card. New CS faculty salaries cap out at 180K at the high end.

This won't convince anyone who wants to pause all immigration.

However, if you want to allow some immigration, you can make a case PhDs in computer science from Carnegie Mellon, which is what he's talking about.

These are kids who were already world-class coming in and become even better by the time they graduate. It is paid for by taxpayers, for which they should be grateful, and it is done in a context that builds admiration for the country.

> It used to be that if you lost your job as an H1-B, you had 30 days to uproot your life and get out of the US otherwise you'd be in violation of immigration laws.

This is still true, right?

Overall, the only hard requirement of the H1B seems to be "can you hold down a job 100% of the time, until you choose to depart or receive a green card?" It is quite hard to think of other requirements that are possible to implement at scale, but I do wonder.

As I said, they already cost more than smaller and cheaper alternatives. But people buy them anyway.

Heavily vehicles may be cheaper for whatever reason you cite. But they are still much more expensive than smaller sedans. People still buy the heavier vehicles.

Evidence suggests people want these cars. They already cost more than smaller and cheaper alternatives. But people buy them anyway.

It is hard to take the ACM's take on publishing seriously, when virtually all significant publications in machine learning, natural language processing, and computer vision are not being published in ACM venues.

> The risk of products like Study Mode is that they could do much the same thing in an educational context — optimizing for whether students like them rather than whether they actually encourage learning (objectively measured, not student self-assessments).

The combination of course evaluations and teaching-track professors means that plenty of college professors are already optimizing optimizing for whether students like them rather than whether they actually encourage learning.

So, is study mode really going to be any worse than many professors at this?

I agree. We are talking about technical, mathy stuff, right?

As long as you can tell that you don’t deeply understand something that you just read, they are incredible TAs.

The trick is going to be to impart this metacognitive skill on the average student. I am hopeful we will figure it out in the top 50 universities.

A lot of this boils down to what you can convince the prof to teach. For most professors, teaching CS1 is a service and does not advance their research. Fewer faculty today are willing to learn Scheme just because it may be marginally better for students, according to some people.

Don’t worry, aforementioned kid is doing just fine, machine learning PhD IIRC. There is a lot of other remarkable stuff in modern computing.