That's a bit too simplistic -- if there is a small group that really pushes things forward in a big way, then maybe not, but if this result builds upon decades of prior work, then Cook and Levin might be equally or even slightly more famous than the solver group after the dust settles.
But it is a moot point anyway. Cook and Levin are very well known already in TCS, and credit is not directly enumerable like money, so "more than a lot of credit" doesn't make too much sense.
For this problem in particular, asking the right kind of question was really important for the field and led to a lot of discoveries even before it will be answered.
> If the answer is "no one really", entropy will overwhelm your codebase sooner or later. Otherwise, you need to read the code
I think about this on the regular -- I know the answer is currently "you own the code, so you have to understand it", but to unlock the true productivity multiplier, in the future, the answer has to be "no one really".
I think about it using the concepts from my job (academia) -- to actually have PhD student-level intelligence means that you have to trust that it does a good enough job that you can focus on other stuff. Professors often bring the correct ideas or intuitions, but they have to trust the PhD student to write the code and/or fill in the gaps in the proofs -- they can advise them on the high-level issues during a consultation, but that's about it.
I am pretty bad at working in the current LLM workflow -- it is tough for me to focus on reading a TCS paper for review, keeping all the details and invariants in my head, but every 5-10 minutes go to my PC, completely switch contexts/projects, read the code and think about the LLM's comments, suggest the next step, and then go back to reading.
> Learning to do more with less money isn’t as bad as many people think.
We are wading into philosophy here, but I believe this analogy doesn't track in this case -- my suspicion from this blog post and others is that already today, the Pro level thinking models are a positive multiplier to your research output similar to how the models one level lower are a multiplier to one's programming output.
Maybe one can someday use the cheaper models similar to how you can use cheaper models than Opus/5.5 and still be nearly as productive as a programmer -- but I am trying and failing doing exactly that for research questions.
I will leave the contact up for a bit longer if people want to get in touch and share their experience with the research gap of the models -- or anything, really -- but I do not think there is any need of further support. Like I said elsewhere, the offer of support made my day and the gesture is enough.
An aside: It was a very nice gesture and completely unexpected by me, so even if it doesn't work out, it made my day. I personally believe that kind gestures have a lot of power.
Back on topic: There is a real danger of the gap between rich and poor universities significantly widening in all fields if the rich can afford Pro level models, or even hardware that can run their own comparable models, and this being fiscally inaccessible to the rest.
One can sweep this under the rug by blaming the educational funding but this just shoots down all discussion. Even if GDP of a country goes up by a lot -- such as Poland -- it takes time before any budget benefit trickles to the education budget, and with some governments it might never do.
I believe Microsoft et al do have the most power here to boost affordable access to AI for researchers on a large scale; the fact that they cut some too expensive models (Opus, 5.5) from their academic benefits package is a grim omen. I do realize they would like universities to pay them also, and ultimately the universities should do that -- but then we are back at the institutional level of the problem.
Can you tell me what is the budget necessary to supply AI tools capable of substantial research assistance to all academic staff at a university?
You seem to have a good estimate in your head; I definitely do not.
From personal experience, ChatGPT 5.5 (the Plus tier) is excellent for programming tasks and also for various teaching related tasks but I have not observed the research benefits that Tim Gowers has when I asked it questions in my area of expertise. So the costs are definitely higher than a few dozen $ a month per PhD/professor.
You might be right that universities should immediately spring into action and demand funding for research level AI resources and hardware. One thing you might be mistaken in is that public universities are unfortunately very inflexible institutions; one reason for this is that they have a large internal leadership structure AND they are funded by the state, so even if the entire university agrees on something, the funding is at the whim of the ministry of education and thus the current political leadership.
There is a significant gap between what academics are paid across European countries, and since most top universities here are public institutions, you are right -- Eastern European government employees tend to be on the poorer side.
There are several other philosophical arguments against what you propose but I do not wish to go down that route.
As a TCS assistant professor from Eastern Europe, I always am a little jealous of the biggest names in math having such an easy access to the expensive, long thinking models.
Paying for Pro from any of my current academic budgets is completely ouf of the field of reality here -- all budgets tend to have restricted uses and software payments fit into very few categories. Effectively, I'd have to ask for a brand new grant and hope the grant rules allow for large software payments and I won't encounter an anti-AI reviewer; such a thing would take one year at least.
As a nail to the coffin, I was "denied" all Claude Opus recently as part of Microsoft's clampdown on individual (and academic) use of Copilot.
(Chagpt 5.5 Plus does not seem sufficient for any deeper investigations into new research topics, I've tried.)
> Is the lock structuring here really deadlock safe? The model will tell you with complete confidence its code is perfect
Fully agree, in fact, this has literally happened to me a week ago -- ChatGPT was confidently incorrect about its simple lock structure for my multithreaded C++ program, and wrote paragraphs upon paragraphs about how it works, until I pressed it twice about a (real) possibility of some operations deadlocking, and then it folded.
> Every time a major announcement comes out saying so-and-so model is now a triple Ph.D programming triathlon winner, I try using it. Every time it’s the same - super fast code generation, until suddenly staggering hallucinations.
As an university assistant professor trying to keep up with AI while doing research/teaching as before, this also happens to me and I am dismayed by that. I am certain there are models out there that can solve IMO and generate research-grade papers, but the ones I can get easy access to as a customer routinely mess up stuff, including:
* Adding extra simplifications to a given combinatorial optimization problem, so that its dynamic programming approach works.
* Claiming some inequality is true but upon reflection it derived A >= B from A <= C and C <= B.
(This is all ChatGPT 5, thinking mode.)
You could fairly counterclaim that I need to get more funding (tough) or invest much more of my time and energy to get access to models closer to what Terrence Tao and other top people trying to apply AI in CS theory are currently using. But at least the models cheap enough for me to get access as a private person are not on par with what the same companies claim to achieve.
Hello, TCS assistant professor here: he is legitimately respected among his peers.
Of course, because I am a selfish person, I'd say I appreciate most his work on convex body chasing (see "Competitively chasing convex bodies" on the Wikipedia link), because it follows up on some of my work.
Objectively, you should check his conference submission record, it will be a huge number of A*/A CORE rank conferences, which means the best possible in TCS. Or the prizes section on Wikipedia.
> Realistically there's no reason government can't use open source software and open formats especially.
> Last time I had to fill out a government form in Canada (...)
Without any evidence, let me argue why maybe it shouldn't. In the past, a common opinion that I have heard is that open source is more secure because all the code is out in the open.
The recent xzutils backdoor attempt [1] kind of led me to believe it's not really true, it's only true if many good-actor eyeballs, which are willing to donate their time for public benefit, are on the code.
Almost all of the government's code that I interact with are web apps that are potential targets of foreign adversaries -- tax filing web apps, prescription + vaccination scheduling web apps, family benefit applications, and more. (This is not in Czechia, but close.)
Now, would I want to read that web app code? Not at all, I couldn't care less about it. However, foreign adversaries would love to immediately start analyzing it. Extracting the entire country's health data or tax data would be a goldmine.
And even though there probably are several people actively paid to maintain security of these systems, I feel that the foreign adversarial agents would be much more motivated (and better paid) than government employees/software developers.
You could make a opt-out for national-security purposes for the code, but I feel almost all the code a government works on would have such an impact when compromised.
(Disclaimer: I am a huge supporter of open source in general, contributed to the Linux ecosystem in the past and in my current job as an academic, almost everything I do is available out in the open in some way or another.)
> I've worked in academia - and yes, that has been in multiple countries in the EU - I've never had to utter a single word in something other than English in the workplace.
I have the same trajectory as you -- multiple countries in the EU, working in academia -- but different experiences for sure. Or at least a mixed bag.
Let me list them in order of how much English sufficed:
1. The Netherlands -- common knowledge is that their English is top notch and anecdotally it was the case as well, I also got by purely with English.
2. Germany -- their English is also good but I needed German in edge cases. One edge case was finding an apartment (not speaking German simply pushed you down the list of candidates, even with a full time job in academia). Another one were university rules and announcements; not every email was in English, but arguably easy to get by with modern translation tools.
3. Czechia & Poland -- English is good among the professors but the percentage of locals at the university level is so high that most internal meetings, announcements, local seminars take place in the local language. In my experience, non-faculty university staff (department secretaries, payroll, entrance security) usually strongly dislike speaking English or outright do not speak it at all.
---
I've omitted some more cases where local languages are required. If you live in a country, you will eventually interact with the healthcare sector, where the language experience will likely mimic the experience at the workplace (for the countries above, it would be in the same order for the healthcare sector).
Another case is government bureaucracy. For most of the EU countries I've been to, the official language of the country is their local language and only their local language. This means that government employees are not required to speak any other language other than the official one to you, plus you might be required to fill in forms and communicate in the official language if you want to talk to them.
In my experience, the helpful/good ones may try to communicate with you in English but if you need something from them or if the bureaucrat had a bad day, you better start talking in the official language.
It reminds me of the fact that the "Fake Mr Beast giveaway" ads that even raised some attention here on Hacker News [1] a while ago are still around. In fact, I have seen one yesterday. Those must have been flagged as impersonation and scam thousands of times by many people, including me personally, and Youtube finds them perfectly fine.
After that episode, where I tried myself to get rid of them, I am much more convinced that Youtube is fine with all but the worst scammers, and don't buy any of the "they're just low on manpower" arguments anymore.
> Hi! I'm a PhD student studying computer science at Rice University.
This means that we are on the same career path (I am currently an assistant professor in theoretical CS in Europe). I wish you of course best of luck!
Here is the harshest truth about teaching I learned during my PhD:
If you are focusing on teaching too much, you are setting yourself up for failure.
This sounds cruel, and in fact I am much like you, I love teaching and I love self-improvement and it is quite easy for me to invest time into my teaching prep, presentation, and more and see measurable results in class quality and usually also student feedback.
However, at least in my neck of the woods (i.e. Europe), almost all gates and gatekeepers for you as a PhD student, and later postdoc, are checking your research. At some places they really do expect you to have K publications in the top 3 CS conferences or you will not be considered at all -- and it seems these thresholds are only getting higher. Here I mean for example invitation-only workshops, postdoc positions with top advisors, and later also permanent positions.
On the other hand, if you are a talented scientist, they usually only care that your teaching skills are at the bare minimum -- have you taught something? Yes? Great.
Now orator/presentation skills are critical and presenting a coherent lecture plan might be useful for a final presentation at an interview for a permanent position. But even there, it is more about you knowing what you want to teach and how it complements the department than about your past achievements (i.e., how much you have put in a course previously).
My PhD advisor usually said that he likes to dig into teaching when research is not going well. I agree with that -- teaching really is fulfilling to me and I love to improve my class and see people happy with it, and research is all about global ranking (which is tough on anyone's psyche) and generating progress which is the fun part but sometimes takes a long time. However, at your stage of your career, the research really can't go slow.
---
PS: If the author reads this, since it is a self-post, your class sounds really nice and it is actually one I would have loved to attend. My research is in online algorithms -- a field which you can rephrase as seeing some theoretical problems as two player games between a solver and an adversary -- and among other things I would like to consider utilizing all the techniques of chess solvers (which cannot evaluate the game fully, but "almost") and transfer it to other areas of online algorithms.
> Isn't the problem de facto solved by matchmaking? The player that aims better will quickly win more and be elevated to the level of opponents on par with them.
Matchmaking decreases the odds you meet a cheater for low rank players, and significantly increases it for higher rank players -- and since there's fewer of them due to the Bell curve, they are going to feel the cheaters that much more.
If you just rely on rank and not on anti-cheat efforts, you'd be just destroying one of the loyal cores of the playerbase, one which is also quite vocal online.
From my personal experience of thousands of hours in competitive FPS shooters on PC, there is no point in ranking where playing against a cheater becomes fair or fun.
> Too often, grad school applicants are just kids that have overachieved in academic settings and think to themselves “I’ve been good at school my whole life, why not just do school forever?
Anecdotally, I did not observe this during my PhD studies (theorerical CS) in Central Europe. I think this might be due to the separate 3 year Bachelor track, then a 2 year Master track, and only then 4 year PhD studies.
Sure enough, a lot of applicants faced tough career decisions after graduating, but whoever started the PhD usually knew what research is about and that it's going to be work first and foremost, not just "more school".
Quickly skimming it, I found no evidence of what the future actually held, from Wikipedia [1]:
> In 1981, Pizza Time Theatre went public; they lost $15 million in 1983. By early 1984, Bushnell's debts were insurmountable, resulting in the filing of Chapter 11 bankruptcy for Pizza Time Theatre Inc. on March 28, 1984.
I was not very clear about it, but I was discussing regular semester work, as opposed to final/midterm exams. Think courses that are strongly grounded in theory but need the students to experience the coursework, like Discrete Mathematics or Linear Programming.
"Oral presentations" in my case meant presenting a homework solution to the TA in person, in front of the class, and the TA accepting this solution live (or not).
At least at my university, the responsibility for homework structure and homework sheets lies fully on the lecturer, and the TAs are tasked with grading the homework/projects and leading the exercise sessions.
Oral exams are great if they can be done at scale, and I do use them. Some other teachers (as well as the administration) prefer written exams, as there is a clear proof of work that can be analyzed if grades are disputed.
Speaking as an (assistant) professor in theoretical CS, I see there are many bad approaches in the original post and mentioned in this discussion, but I strongly disagree with:
> There are very simple and effective ways to teach your class that can't be cheated with AI. These professors are simply lazy and uncreative.
I can attest to the following problems to good homework creation, from my own experiences playing with ChatGPT and teaching:
1. If you want to give a very illustrative yet easy theoretical exercise in algorithm design, one that computer scientists have solved over and over in the last few decades and which furthers your understanding, there are very likely solutions online and ChatGPT will give you the solution with very high probability.
2. If you create your own dataset and want the students to implement some algorithm and create a simple plot/discussion from the results, it will be very hard to distinguish a "student solved it on their own, but they did not invest too much time into it" submissions from ChatGPT submissions produced by a couple of queries.
3. Switching to oral presentations is hard to scale (as others attest) and also does not resolve much, because some students are perfectly okay with being handed a solution from somewhere (colleague, ChatGPT), not understanding it very well, and yet presenting it. Failing these students likely leads to overly difficult classes.
4. In-classroom exams without a computer work best, but they also do not scale very well (a lot of prep/correction needs to go into them) and some students with bad anxiety management skills, which includes me as a former student, dislike them passionately.
---
As you can see, this topic is quite critical for my profession. The ugly truth is that university professors have only a very limited time allocated in their busy workweeks for teaching, and hence they have to take many shortcuts, including suboptimal homework sheets and limited innovation year-over-year. I also do not allocate as much time for philosophy of teaching/improving teaching skills as I would have liked.
If anyone here has novel ideas how to actually implement "a class that can't be cheated with AI", specifically university CS classes, I am all ears.
But it is a moot point anyway. Cook and Levin are very well known already in TCS, and credit is not directly enumerable like money, so "more than a lot of credit" doesn't make too much sense.
For this problem in particular, asking the right kind of question was really important for the field and led to a lot of discoveries even before it will be answered.