# Everybody Is a Prof Now > Source: > Published: 2026-08-21 09:31:16+00:00 Published: August 20, 2026 We have been looking for those places that are, at least to first order, bottlenecked by intelligence for a while now. Among the most conspicuous such places are maths, coding, and seemingly offensive cybercapabilites. In that same step, much of the academic grunt work required to execute an idea in fields that are predominantly or entirely based on computation has collapsed. The distance between idea and finished model, algorithm, benchmark, etc. has meaningfully contracted to a degree that it is no longer a meme that your 19-year-old Freshman is now just one-shotting his NeurIPS submissions on his $100 Claude Max subscription. This has, one is nearly tempted to say *mechanistically*, to a devaluation of this type of work in practice. In practice, there is not a whole lot of value in producing much of this work unless it delivers something beyond its value in implementation by contributing directly to the wider body of knowledge, by testing a clearly defined and relevant hypothesis, or by providing genuinely useful infrastructure (such as a well curated dataset or benchmark) that others build on top of. Your immediate and validated reaction to this may be: wasn't this always the aim of science? And of course in many ways you're right. It was and is. But in the realities semi-professionalized system of modern science, in which every participant is required to meet some quantitative proxy of actual scientific relevance, namely grants, citations, and papers, there had long been a deference in practice to the people executing mildly interesting work well, or at least doing it at all. The pure act of spending many months, and in most cases years, developing code for a certain (more or less made up) problem still involved real effort and, which is mostly the same thing, also did teach the author real things. In that way, the contribution of the paper consisted mostly in the actual doing of this effortful (if not always impactful) thing, meriting them a publication and one more step on the eventual path to graduation, tenure, and academic glory. Clearly, this line of argumentation can hold no longer in a time where writing up code has become ultimately commoditized. There simply is little in benefit to academia (let alone society as a whole) in just writing up things. In a perfect world, this would be a wildly good thing and certainly the amount of good review that at least a code-based paper can get these days is incredible. In a perfect world, every paper would be internally and externally audited by a large number of very capable, perhaps even specialized, models that sleuth out every mistake and bug (intentional or not), enabling a new gold standard of methodological excellence not even rivalled by the biggest top-tier journals today. In fact, anybody who has tried realizes how much we suffer from the fact that this has not been the reality over the past years (certainly the years since 2018) and how easy it is to find papers which, to put it mildly, would have benefited from such a setup. In a perfect world, we would see many current paradigms being closed as we come to their natural endpoint much faster and many new ones popping up as new capabilites allow for new ways of tackling problems and testing out these new ideas becomes much faster. Of course we do not live in this perfect world. Rather we get the same sluggish academic process supercharged with an alien intelligence that is all too happy to slave away at building out our half-baked ideas and ill-formed hypotheses. If only more people would look at the hill their climbing! But no. We don't see paradigms being overturned -- as every time this happens it's a potential risk to someone's career. We don't see the standard of publication rising, instead, preprint servers have banned survey and review papers for the pure flood of slop preprints, hallucinating insights and academic thought. (If you ask me this will soon be followed by a paperblank ban on the deluge of biomarker-correlation-to-x papers that have swamped the preprint servers in recent years.) What irony indeed! In a time where we should be making the fastest kind of progress in these areas of science, wouldn't we more than ever need the services of a good review, a guiding text, our *spiritus rector*, to help keep up with the pace of change. Trying to consolidate our findings into general theories, updating our priors, building better conceptualizations and, most of all, understand what to work on next. The review should be the holy grail of academic reasoning, the consolidation of years of data, experience, and insight distilled through the author's genuine taste and creativity into salient, get-to-the-point insights that help decide the questions to ask, the ideas to let go off, laying the groundwork for a future generation's textbooks. What we see are the same faultlines of academia simply supercharged. The same people that have always fought for their pet paradigm continue to do so -- just with more compute behind them. The same people that always uncreatively tried to reward-hack the system by publication-maxxing are continuing to do so -- just semi-automated. The same planting-a-flag-mentality just to get their first still applies just that the race is now happening at F1 speeds rather than at a runner's pace. There are some odd contradictions here. Strangely, the writing of the publications themselves, arguably the think a next-word-prediction model shouldn't have replaced first, is perhaps the place we're most resistant to AI slop. Pangram is pretty good these days! What is much harder to account for is how much of the thinking has been outsourced upstream. Ironically, we are very much still intelligence (or should I say, originality?) bottlenecked in the actual question of what to work on. I have yet to see these models come up with really good scientific projects, original hypotheses, and good research taste. Accordingly, the value for these has only gotten higher. It used to be the case that these were the same points you went to a famous professor for as they would decide what you'd work on, what questions to think about, what research to produce. You'd hope that these were these were the right hills to climb to but in doubt you'd defer to them, at least they had years of experience in the field and had been in your shoes before you. The setup was thus one brilliant person surveying the literature distilling it into a few, actionable hypotheses that his armada of capable (i.e. hardworking) PhD drones would execute on. During this process, a PhD would have to do a lot of hard and tiring work but inevitably they would learn something (as you always learn something when doing it with some dedication and genuine interest) and essentially by osmosis pick up on the state-of-the-field. In rare cases, they'd come to know the field so intimately that they'd become empowered to take over their mentor's role and start thinking of good, salient hypotheses themselves. Essentially having successfully built a well-working world model for their particular subfield. But even in the default case, just having executed the grunt work of getting it done, writting the code, building the thing, mostly propelled the field forward and certainly would have been fairly rewarded by a publication or similar. This setup clearly does not hold anymore. In a world in which the execution, the grunt work, has become commoditized exchanging a prompt for a paper is madness. Yet it is where we are at in most cases. The incentive structure of academia has of course been broken for a long time -- it has just become much easier to game it now. Similarly, it is questionable whether you still need your guiding spirit in your professor at least as long as you think that you have sufficient research taste to just go out and do your own projects. I will say it is hard to know whether your building the right thing, climbing the right hill. You might just be deluding yourself in thinking you have good research taste and it will not entirely be easy to tell probably for a long time. That same caveat applies to selecting professors to though and one could hope that being able to taste out ideas a lot more quickly and get feedback on them allows you to build good intuition on them. A good gut instinct only develops when three things combine: an environment that is sufficiently predictable; you have rapid feedback; and you get many opportunities to try. At least the last two have just gotten massively accelerated. So really we have all become professors now. You can just have an idea and test it virtually immediately. You can try out a lot of things and nobody needs you to do the brunt work anymore. At least in the computational fields of science. You continue to see (and per my prediction) will continue to do so for a long time that the very brunt work we just talked about in the wetlabs of the world will continue to pay off -- and rightly so. Setting up the assays, doing the protein purification, getting the sensor to work will remain tedious, yet enormously valuable, things even if they are mostly feats of incredible patience and stamina rather than brilliance alone. I'd argue to embrace it. The fact that we can all become arm-chair scientists (I prefer this term much more over the SF preferred term of "gentleman's scientist" as it highlights the fundamental amateurish nature of its practice as opposed to someone who actually has put in the time to master a field or skill) in certain areas of research -- so long as we actually know what to work on. That last point evidently becomes all the more crucial. To think with brutal honesty about what problems to truly work on. To argue what actually is a contribution to science rather than just a plant-your-flag exercise. To have taste in the face of uncertainty. It will not be easy but we should embrace what might be a very healthy reset for a mostly broken system. Let's hope that the rigidity of academia is lesser than the deluge of AI generated slop and that we may be allowed to build creatively and candidly in the face of its desctruction.