In an ideal world, sure. In reality, a PhD[1] is about publishing papers in the hope that you learn something along the way.

[1] My experience in ML in 2026.

> In an ideal world, sure. In reality, a PhD[1] is about publishing papers in the hope that you learn something along the way.

You're being needlessly dismissive of what "publishing papers" mean. Writing down what you learn is what papers are for, and the only papers that count for a PhD are those which are accepted by reference journals, which impose a relatively high bar on what content they accept.

So you still need to explore uncharted areas, and your main output is still writing down what you found for others to build upon. Those are the deliverables. For the person doing the PhD, the primary benefit is developing research skills, such as independente and self-reliance.

> which impose a relatively high bar on what content they accept.

Nope. In fact they accept utter garbage most of the time. I can't count the numbers of times I've caught pure errors (literal logic errors, code errors, mathematical errors) even in highly respected journals. Journals have a high bar of entry because they've gated their communities, only allowing insiders to publish (ie tenured professors, people associated/employed with university labs or their colleagues). If you don't have an industry connection the odds of you self publishing in these journals is zero. And if you _do_ find a connection, say a professor whose willing to help you out, they'll demand to place their name on your paper (without actually contributing anything). This is how you get papers with 50 authors, because every politicker wanted a piece of the pie.

This is also why you see frontier AI labs self-publishing on their websites rather than in journals.

> This is also why you see frontier AI labs self-publishing on their websites rather than in journals.

What they self-publish on their own websites should be regarded as marketing.

> Nope. In fact they accept utter garbage most of the time.

Unless you enrolled in a fraudulent, make-believe University of Phoenix style program, no. Any real PhD program only counts publications in a closed allowlist of reputable journals. Those journals are peer reviewed and enforce quality and relevance standards.

> I can't count the numbers of times I've caught pure errors (literal logic errors, code errors, mathematical errors) even in highly respected journals.

That's fine and all, specially if what you actually want to do is take a shot at humble bragging, but back in the subject papers are supposed to be snapshots of your understanding of a topic that can be used by others to inspire or advance their exploration. This means they are not infallible or free from criticism. In fact, it is rather ok brand to publish papers improving upon prior publications by correcting errors, improving methodologies, and even presenting contradictory evidence. That is what it means to be at the bleeding edge.

The commenter is stating that respectable journals accept a lot of bad papers. This is actually very true.

Which is ironic, because self-publishing then should automatically qualify you for a PhD. In fact, I believe more people should take this approach. As far as I'm aware, there's no law or regulation from declaring yourself a "PhD" backed by autoresearch.

I've got a PhD in Physics and I have to say that the quality of most "self published" physics is very low, which probably explains why people would not accept a self-declared PhD.

> Which is ironic, because self-publishing then should automatically qualify you for a PhD.

No. What qualifies you as a researcher is being able to output work that is accepted by your peers. Hence the peer review process is critical. You fail to meet this bar if you publish something in blogspot or even arxiv for that matter.

A key problem is upper-level academia is rife with the same sorts of human-scale power dynamics. Petty squabbles, ideological (and ideological-flavoured) views can and do hold entire fields from time to time. It is only a particular type of tenacity that breaks up the status quo from time to time. (consider the Marvin Minsky era of AI research. A big personality can stifle an entire field). Peers aren't usually the appropriate measuring stick of merit.

> A key problem is upper-level academia is rife with the same sorts of human-scale power dynamics.

You are trying too hard to come up with excuses. The paper review process is double-blind, and nothing prevents you from publishing the same work in multiple journals. If your goal is to do meaningful contributions to science then "power dynamics" are irrelevant.

You do need to develop a meaningful network in the field if your goal is to collaborate with other researchers, but that's basically what everyone needs to do in any professional field.

Entrenched ideas have momentum. Your lone-warrior attitude is admirable if not lacking in substance. Most fields have had occasional hangups on dead ends, such as the aforementioned Minsky "symbolic > connectivist" debacle. Behaviourism in psychology. String theory. The list goes on.

You can't generalize from your personal experience to the rest of academia. There is far too much variation between fields, institutions, and countries as well as over time. ML in particular is an extreme outlier due to amount of money and business interests involved. Your experience would already have been quite different in a less fashionable subfield of CS.

Your role is to publish papers so your PI is able to get grants, and to network, so the lab profile gets better known in the research community, and then as a more senior grad student you can help write grant applications, and you can build a track record of mentoring (e.g. master students). At the end of the day, you want to demonstrate that you can come up with projects and pursue the research up until publication in a top-tier venue, and the ultimate goal (from the point of view of hiring committees) is obtaining prestige and funding for the university.

If you plan to go to industry, your incentives are somewhat misaligned with your PI. In that case you want to pursue multiple internships and getting networked with top companies in your field, and demonstrating practical output, well-released code, demonstrated downstream value for followup projects etc.

Yes, take away all the pretense and a Ph.D. is just an apprenticeship to be a researcher.

There's no pretense. That's always been the whole point.

That's not my experience -- there's a whole pretense about being experts in a field, which is exactly what the poster above bought into. Having a PhD doesn't make or prove you're an expert about anything. It just proves you know how to do research in a thing, but that doesn't stop us from presenting PhDs in front of juries as if they're the only ones who know what they're talking about.

I don't see how these are incompatible. The PhD prepares you to be an independent researcher. The way you prove this is by studying the same topics for a few years, and by the time you reach your viva, you'd be expected to know a lot about the topic you chose.

It's not incompatible, many people in order to do research have to gain a great expertise in a thing. Others can get a Ph.D. having really no expertise at all, but just the ability to get through a research program. Others still can be experts and unable to navigate a research program due to how capricious and political they are. So the pretense is that the Ph.D. is a marker of expertise. It's correlated with expertise, but the actual thing it signals is you've gotten through a research program.

Self made experts with no PhD or formal degrees are testifying in front of juries. They dont even need any science to back them.