Opinion: Cutting off the Lizard's Tail: A Flawed Perspective

I guess I should start by saying that I’m in a much better place now than I was two years ago. Even compared to a year ago, things have been changing and moving so fast that it all feels a little surreal. I’ve had more opportunities that I wouldn’t have believed possible not long ago and I find myself to be one of the only bottlenecks left to my own success. And that makes this perhaps a little strange to write, maybe even self-indulgent. But that’s also why I think it’s important for me to write this right now, while I still remember what it was like to struggle to even get a foot in the door, to not even know what I know or what I don’t know, and how it feels to have people actually give you a chance.

I’ve done academic research for less than two years at this point, and there’s obviously a lot I don’t understand, especially what happens in the room where it happens. I don’t claim to have some sort of magic bullet for the problems I talk about below, and maybe there isn’t even a solution that would satisfy everyone. And yet I write this in a moment of clarity in hopes that this will reach people much more capable and powerful than I am, so that it can add another perspective to the ensemble.

So take this as a terribly flawed perspective from someone who probably isn’t seeing the full picture. This is how I see it.

As an undergrad at UIUC, I wanted to do research. After all, my main motivation coming into a computer science and linguistics degree was to create something like Baymax, and I wasn’t going to get there by learning how to do LeetCode. But having just “entered the field”, by which I mean I’d just finished reading a couple papers from Percy Liang’s group about LLM agents going off and organizing a birthday party together (or at least that’s how I remember it), my interests were all over the place. I emailed practically every professor that seemed to be working on something even remotely related. At this point I had built personal projects, was the president of a decently sized coding student organization, competed in hackathons and datathons, and even presented smaller projects at student research symposiums. Unlike others around me, I didn’t have anyone to tell me what I had to do to get into research, and I didn’t even know about what I didn’t know, so I guess in my mind I was doing everything I could think of.

But even after all that, I still couldn’t find a way in. I would have been happy to shadow someone or help run experiments, whatever was needed of me. I wasn’t asking for a salary or a recommendation letter. I just wanted my foot in the door.

I say this, but I guess it’s not like those professors owed me anything. I understand this. I’m sure they had much more on their plate than I could imagine, writing grants and teaching and whatnot. I’m sure some of my cold emails could have been better, and I probably wasn’t as ready to contribute as I thought. But at the same time, if not at one of the top research institutes in the US, how was I supposed to get my start?

Eventually a professor gave me a chance around 3 years into my search (FOCAL Lab shoutout!), mostly because I happened to be close to one of his favorite students. I’m very grateful for that. I learned much from this experience that I carry with me today, and of course it changed my path a great deal. But it’s not like I suddenly became a better researcher because I knew the right person. I was still the same student who had been sending cold emails for years. And I wonder what would have happened if I hadn’t known that student, and how many people are in the same spot but never meet the right person.

UIUC is a well-regarded research university with a lot of resources. I think overall I had a good education and had people around me with similar interests, and I had room to learn on my own. And if despite all this, getting started was that hard for me, I can only imagine what it’s like for someone at a smaller university or without an academic network. The same goes for people trying to enter the world of research on their own.

It’s strange how early we expect people to have research experience nowadays. Competitive Master’s and PhD applications put a lot of weight on prior research experience, and prestigious undergraduate research positions often explicitly favor students who have already worked in a lab or published. I’ve even been rejected from a student organization because I didn’t have any publications. You end up needing research experience to get research experience.

I suppose things get better when people are able to get past that first barrier, as they then often get access to opportunities that make the next barriers easier to overcome. This is known as the Matthew effect: a 2018 study found that early-career researchers who won a grant went on to get more than twice the funding over the next eight years compared to those that didn’t, even when the two groups had almost identical review scores. Once you’ve had that first opportunity, people have something to go off of. You have people who can vouch for you and, more importantly, a better understanding of how things work. Before that, you’re kinda just asking someone to take a chance on you.

And when getting those initial opportunities becomes increasingly competitive, I think it’s almost inevitable that people start looking for other ways in.

As research experience and publications become increasingly important for getting into graduate school, I’ve also noticed that there’s basically an entire industry around charging high school students thousands of dollars for publications.

One such example is Algoverse, which apparently costs $4,325 to participate in. On one hand, I’m sure that running something like this costs a considerable amount of money, and I’m sure it’s not impossible for good work to come out of a program like this. After all, if someone is capable of making a meaningful scientific contribution at sixteen, more power to them. My concern is more with the idea itself: the incentive structure building up around research.

I may be jumping to conclusions, but considering that this is something that most Master’s students I’ve talked to struggle with, I wonder how many students are participating because they actually have a question they’re curious about, and how many are doing it because they think having a paper will help them get into a better university. And considering that this is happening because admissions committees reward research experience, and research experience is increasingly difficult to get without already having some, it’s not entirely right to blame the students for this either.

Needless to say, those who can afford to pay thousands of dollars for papers are probably not the people who would have struggled much to find opportunities to begin with. It’s another way that existing advantages widen the gap over time.

Let’s say someone finds a question they’re excited about. Arguably, this is already a pretty high bar when starting out. But on top of this, let’s say they have no official advisor, or their institution has little funding. In machine learning, the research itself can be quite expensive due to compute and potential API access to closed-source models. But I suppose at this stage, it is not impossible to get a small compute grant or something along those lines. But even if they somehow end up with something worth sharing despite these obstacles, they then run into a second set of costs. Conference registration regularly costs hundreds of dollars. Add flights, a hotel, food, and maybe even a visa, and a single international conference can easily rack up a huge bill. For independent researchers or people at underfunded institutions, especially in lower-income countries, the cost can make attendance effectively impossible.

It’s not like nothing is done about this; NeurIPS and ICML both run financial assistance programs, but the programs leave something to be desired. NeurIPS 2026 financial assistance can cover registration and hotel, but not travel, and ICML 2026 also offered help with registration and accommodation but not airfare. Even with an accepted paper and financial assistance, it’s quite realistic that you might not be able to afford the trip.

Much of a conference’s value in my opinion comes not from the event itself but from the conversations, where you can get feedback and meet possible collaborators. If you can go, your paper can lead to new opportunities. If you can’t, you get a line on your CV and end up back at square one. It’s not just missing a conference. It’s missing an opportunity to meet the people who might make the next opportunity possible.

Of course, I’m sure running a conference is expensive. There’s a lot of staff and logistics involved, often working full time to work out venue contracts and whatnot, and all of that costs money. And I’m sure this is what the registration fees go towards funding. At the same time, according to its public nonprofit tax filings, the NeurIPS Foundation had about $22.3 million in net assets at the end of its fiscal year in May 2025, after about $15.2 million in revenue and $10 million in expenses. That’s a lot of money to be sitting on when there are countless people at every conference who end up not being able to attend due to financial restrictions and there are talented individuals who have no way to break into the field. Since I haven’t come across anyone who knew about this while writing this post, maybe it’s at least worth discussing.

Academia is struggling, especially in machine learning, and from there especially in LLM/agent related fields. I think most people in the know would agree with that.

NeurIPS went from 9,467 submissions in 2020 to 21,575 in 2025, more than double in five years. The reviewer pool has grown somewhat too, thanks to reciprocal reviewing, but it seems almost inevitable that review quality slips.

Reviewers are usually overworked PhD students with a conflict of interest where the game-theoretic way to “play the game” seems to be to just reject your entire pile to marginally increase the chances that your own paper gets in. It’s like some nerdy version of the prisoner’s dilemma. There is a world where everyone reviews thoroughly for each other and contributes to progress and the accumulation of knowledge, but alas, it’s not the one we’re living in. Even if we don’t go that far, with LLMs making it easier than ever to produce something that is “paper-shaped”, it calls for changes to how we look at academia.

I guess for one, the goal of research shouldn’t be to get your paper accepted at a conference. Research should be done to find out something new, contributing to the pool of knowledge. Publications are just one way to share this. Many senior researchers have thus been calling for a shift away from conferences, towards things like blog posts and Twitter threads to be spread around the community so that we can avoid going through one overloaded submission cycle.

And I think this is all well and good; it would likely make things a lot easier to “filter out” “slop” and focus on things that are likely to be more impactful and stronger works. But this also comes from a somewhat convenient moral high ground, in the sense that it overly simplifies the effect that this would have on academia across different levels of experience and background. Yes, I have plenty of complaints about how conferences work now. I’ve already gotten my share of sloppy papers to review and even sloppier reviews to my own papers. I’m just worried about what would replace them, and who that replacement would work for.

On the surface, the idea of a crowdsourced filter sounds great. Instead of spending months only to get rejected by some AI agent, you could just put your work out there and let people judge it for themselves. The problem is that I don’t think this is how it would actually work in practice.

There is already an overwhelming amount of research being published every day on arXiv. People have thus turned to somewhat superficial indicators to decide what to spend their time on. Who wrote the paper? Where are they from? Do I recognize any of the names?

If someone like Yann LeCun puts his name on a paper, or even just posts something on Twitter, people will read it. But what if you’re an undergraduate at some random university? Nobody knows who you are and you don’t have a professor with thousands of followers to share your work. You can try to publish it on arXiv, maybe post a thread on Twitter, and hope that someone looks your way.

Maybe someone will. But I don’t think we should build a system around the assumption that they will.

I find it personally difficult to separate the value of publications from the opportunities that come with them. Sure, research shouldn’t be done for the sake of getting papers accepted. But telling someone who is just trying to get their foot in the door that they shouldn’t care about publications seems a little strange when the system that evaluates them hasn’t changed to match that.

It’s easy to say that publications don’t matter once you’ve reached a point where they don’t matter anymore.

And I don’t think it’s enough to tell people to “stop caring about publications” and to focus on doing “good work” when practically everyone else, from graduate admissions to hiring decisions, still rewards them. Until those incentives change too, we’re basically just making it harder for people without connections to compete.

I don’t think any of these problems are easy to solve. Academia is already struggling to keep up with the amount of research being produced, and I agree that something probably has to change. But I worry that when we talk about making academia “more efficient”, we’re mostly hearing from people who have already made it through the system. The people who haven’t even gotten their foot in the door don’t really have much of a say in what happens to it.

I guess this is where the title of this post comes in.

A lizard can detach its tail when threatened, sacrificing a part of itself to escape. And in some sense, it feels like that’s what’s happening with academia right now. The system is becoming difficult to sustain, and so we’re looking for parts that we can “afford” to get rid of.

For an established researcher, getting a paper accepted at a conference might just be like a formality, an opportunity to meet up and hang out with old friends on someone else’s money. They already know who to talk to and how to get people interested in their work. If they stopped submitting papers tomorrow and instead published everything on their personal website, I imagine they’d have no trouble getting people to read what they write.

But for someone trying to enter the field, getting accepted at conferences might be one of the only ways to get their work taken seriously. I, for one, didn’t get into research because someone saw that I was “good” and deserved an opportunity. I got in because I happened to know someone who knew someone. And I’m sure there are plenty of people who would have been just as capable as me, if not much more capable, who never got the same chance.

I think accessibility should be treated as something we actively try to preserve when reforming academia, rather than something we hope will improve as a consequence of fixing other problems.

The lizard might be able to survive without its tail. I’m not really convinced that everyone depending on the current academic system can afford to lose the same things.