Blog: On the capabilities of modern AI coding models in developing genomics software
My lab develops high performance methods and tools for computational genomics (salmon, oarfish, alevin-fry, etc.). Previously, I've been a pretty outspoken skeptic on the quality and utility of AI coding models in this domain. Some recent experiences have led me to seriously update my perspective!
You can read more in my recent blog post here!
0 answers
No answers yet.
Log in to answer this question.
thanks for sharing your experience. I think posts like this are important. there absolutely was a huge shift in the capabilities of tools in 2025 and claude code + opus has been a game changer. I wrote about my experience here and i expect I will likely write more https://cmdcolin.github.io/posts/2025-12-23-claudecode/
there are still many skeptics who seem to require very clear evidence of their advanced capabilities so i think we, particularly academics who have the liberty of developing open source software, have a unique position where we can openly and freely communicate the things that they accomplish for us. like i said in my post above, it is almost a taboo and sensitive subject that triggers a lot of backlash but i think it's still worth posting about.
I think also being clear about what model was used can help because there are many that still believe that the models are bumbling idiots and this is likely true for lower quality models but opus is very capable
bonus edit: the downsides that you point out are spot on and serious issues, and are just the tip of the iceberg also
I think one aspect of brainrot that you've not touched upon is that of humans taking credit for work that other entities have done (in this case that other entities being AIs). We probably already have a lot of people using AIs secretly all the while dispararing them in public, caught between the rock of convenience and the hard place of adverse public opinion ("AIs are stealing jobs"). AIs having become effectively equivalent to humans in tasks such as programming has probably put quite a lot of people that were hitherto programmers themselves into more of a (sometimes disguised) managerial role now. And I don't know if human society -- as it is at present -- can handle this.
If AI constructs are involved in producing something, they should ideally be credited explicitly, essentially as equivalent colleagues. Viewing them as tools is, in my limited opinion, the wrong way to go.
From your blogpost:
I think, for most, the ultimate reason for resorting to automation of any kind (be it resorting to consultation with experts, or outsourcing, or using an AI) is to work less -- so think less (even in the context of learning something new) -- while maintaining/improving their standards of existence in some relevant sense. Outcomes of the kind you have alluded to here are inevitable in this context. We cannot expect students to continue to put in the effort to learn when the necessicity to learn has essentially been circumvented: why should they waste hours upon hours of their lives trying to understand something thoroughly when, instead, they could spend their time as they wish and simply consult with an AI to acquire instantaneous (but perhaps transient) understanding as and when it is required? Especially if the only reason they sought to get educated in the first place was/is to simply make ends meet?
None of this bodes well for how we all will treat each other (humans and non-humans alike) going forward, I feel.
Thanks for the thoughtful response. I'm completely with you up until this point:
Rather, I think the answer to this is precisely the point I hope to get across (but grant that I have not figured out how to explain properly). That is, one thing that I think these models really bring to the fore is the distinction between students with an _intrinsic_ motivation to learn and understand versus those with a _mostly extrinsic_ motivation. The point of putting in the "hard work" is precisely to learn. While AI models are on a completely new level by comparison, there are still analogies to prior technology. For example, why would someone learn about the integral calculus, how it works at a deep level, and even how to calculate difficult integrals, when we have software like
Mathematica? I think there are a few answers to that question. One is intellectual curiosity, and I feel that figuring out how to inspire that is part of the key to any good approach to education. Another is that, for those that truly understand the subject, they will have an idea of what to do in those rare situations where they are confronted with a problem thatMathematicacan't solve, or where they must break down a complex problem into simpler pieces that it can.This later example joins together my concern on "brainrot" with my concern about the "growing gap". One fear I have, in particular, is that rather than increase the capabilities of everyone by an approximately equal amount, these models and widespread access to them will _vastly_ increase the capabilities of the _intrinsically motivated_ and _intellectually curious_, but only moderately increase the capabilities of the _extrinsically motivated_ folks seeing education only as a temporary means to an end. I feel this is bad for multiple reasons, but a big one is that I think it will further demotivate the people in the latter category, making it even more difficult for them to overcome the deficit.
In brief, I think that from an educational perspective, these technologies may greatly magnify, and lay bare, this fundamental distinction in the attitude that different types of students bring to learning and to their education. Previously, even if greatly flawed in many ways, the "challenges" provided to the students in terms of e.g. difficult programming assignments or deep essays, provided a way to motivate the actual learning process even for primarily extrinsically motivated students (i.e. they still "needed" the decent grade to achieve their end). Now, when that can be easily offloaded to an AI, we lack even a flawed mechanism to attempt to "force" the process of actual learning / education. This is a problem to which I think we need to find a solution in short order.
I absolutely concur with you here with respect to what will happen to the divide between those with intrinsic motivation as opposed to those with extrinsic motivation. I think that "demotivating" effect you have alluded to is causative -- to an extent -- of the anti-AI sentiment being witnessed on platforms such as Reddit.
I think this will simply lead to us hopefully confronting the actual problem: is it actually necessary for those with extrinsic motivation to learn whatsoever? I mean, after all, you can't really choose nor control whether or not your motivation for learning something (or doing anything really) is extrinsic or intrinsic. It is just luck in the end. Is it fair, then, to essentially "punish" those that weren't lucky enough in this respect for that (non-)crime? Honestly speaking, a lot of this just points towards universal basic income (or something to that effect) becoming a necessity.