Scientist. ♥s epidemiology, running, prob+stats, 🦀/🐫, fly fishing, greyhounds, backpacking. Author of book Bioinformatics Data Skills. All views my own.
Happy to share the alpha release of camdl, a new compartmental modeling framework I've been cooking up.
With coding agents writing more of our scientific code, I think it's more important than ever to shift to architectures that engineer in robustness...
https://t.co/RSJjZFYa5c
The AI revolution tests whether we as scientists are actually engaged in a search for truth or not (however imperfectly). Preserving the human role is crucial *b/c it's essential to understanding findings and advancing our research* but it would be foolish to not use these tools.
Nature News covered a recent preprint of mine where I showed hundreds of papers made impossible claims of protein expression in cell lines with deleted or mutated genes: https://t.co/HXhwlZBdur
@SponceyM No, I disagree. The nature of technology derives from its inventors’ visions and designs. Many different products can have the same technological capabilities but have vastly different feels and aesthetics, which determines their adoption and impact on humanity.
I think a huge problem with our current AI era is it lacks the type of visionary that Steve Jobs was.
He saw that great technology fundamentally uplifted and amplified human capabilities, not replaced them. Great tech expands human agency and creativity, not crushes it.
I think this attitude also makes current trajectory of AI products self-limiting — in part because it’s so unimaginative of how humans and AI could interact.
Where is the vision for AI that enables us to build mastery of new subjects, and be more creative and collaborative?
So few AI products are built this way though. So much is predicated on an almost fatalistic attitude of “better models will replace us, so what can we do about it”.
Ironically, this resigns us to inferior, less-imaginative technology that AI is supposed to make easier to build.
Someone should build a system that has AI generate many different proofs of the same theorems, from elementary to extremely difficult. Have mathematicians review each one and identify the first genuinely new, reusable idea.
What’s the first thing AI can teach us we don’t know?
This has to be the most absurd case of image fraud in academia I have ever seen. Now I have found three different papers on chicken nuggets with problematic figures. Systematic chicken nugget fraud 🐔🍗
Say what you want about LLMs - and how smart you are without them. I can say without hesitation that they’ve made me smarter. Being able to interrogate an idea, ask anything, and go as deep or sideways as I want has changed how I think. Bottom line: I’m learning more, and better, than at any other point in my professional life.
@jmschreiber91 This is usually what I do too, it works well. I spend most of my hours iterating on big proposals, then having it implement them. I think my mistake is trying to avoid this for small to medium sized tasks and then they just run haywire.
Increasingly I want Opus 5.5 to be more of a thought partner rather than charging forward full steam ahead. It’s impressive how far it can get alone, but its choices in scientific projects and code are always a bit off.
I'm maybe overly obsessed with code beauty, but even the best models leave a lot of code ugliness around. It's faster/better than humans on average, but with high variance.
@jeremyvancleve@SashaGusevPosts The only reason why people use preprints is because they are pre-journals. Journals should cease to be how we judge and share good science, and probably will with AI.
@jeremyvancleve@SashaGusevPosts I can blog immediately new results, invite collaboration and scrutiny, ask people for methods input, I don’t have to pass some preprint screener that is in my experience rather arbitrary and slow.
@jeremyvancleve@SashaGusevPosts I think you should invert this: why preprint and not blog though? What does it offer other than just a veneer of authority over a blog post? I disagree on formal style but aesthetics are harder to debate. Personally I prefer original, good human writing over a AI explainers.
@jeremyvancleve@SashaGusevPosts It’s not the same speed. It needs approval from someone. I can write a blog and post it immediately. Preprints are often written in standard scientific prose which is needlessly dry IMO. Why not blog then preprint?
Fantastic read. I use AI a ton, and it’s clear that offloading tasks is great, but offloading thinking is likely a long-term losing strategy for intellectual labor.
I’m surprised how little innovation there is in human-AI interfaces designed to augment thinking, not replace it.