Профиль

en.wikipedia.org/wiki/Garbage...

IMO it should be avoided. there are other more specific terms that describe all the phenomena it encompasses and muddles.

Mark's been quiet for a while. Someone needs to do it!

It's cautionary tale for LLM use though. Data aren't democratic; some are better. Reminds me of something a biologist recently said to me of an AI colleague: "His models are fantastic. It's just a shame he doesn't realize the output's garbage because half the papers fed into it are wrong.." 5/n

So all models define epigenetics as gene regulation but without pointing to the major mechanism of gene regulation (transcription factors) about which we arguably know most and have most evidence for. This isn't a problem of the models per se (Wikipedia is little better) but the training data. 4/n

ChatGPT: Epigenetics is the study of changes in how genes are turned on or off without changing the underlying DNA sequence…The main mechanisms include DNA methylation; histone modification; non-coding RNAs” 3/n

Gemini: “Epigenetics is the study of how environment and behaviors change how genes work without changing your underlying DNA sequence. Your body uses...DNA methylation; histone modification; non-coding RNA.” 2/n

I guess ultimately it’s all just models…but this reminds me of some schemes/animations where people point out the distinction between structures that are experimentally supported and artists impressions is not always clear.

Paul ends his Baric lecture with the admonition that “scapegoating is one of the most reprehensible activities humans have devised”…

"participation in formal scientific communication is negligible...Most leading developers of AI tech do not engage with the scientific literature, raising concerns for the transparency, reproducibility, and accountability of this rapidly moving innovation frontier" www.biorxiv.org/content/10.6...

Negligible participation in the scientific literature of AI unicorn startupsArtificial intelligence (AI) development is concentrated within private firms, yet their participation in scientific publishing remains poorly understood. We conducted a bibliometric analysis of all 3...www.biorxiv.org

I was waiting for that one 😆

I remember little about it beyond on the intriguing notion of breeding people for "luck"

logorrhea… 🤣

Meanwhile most AI-generated slides/schemes just look awful frankly. That may change but currently I can spot them a mile off. Plus of course they increase the tendency to create metaphorical images on slides, which I can't stand and make everything feel like a commercial for IBS meds... 2/2

I see the logic in prettifying a figure (if one's not an Illustrator whizz) and obviously a use for creating images you could not draw yourself (e.g. a heart). But I've always been skeptical of visual abstracts unless done by professionals. Do they inform or just decorate? 1/2

"HHMI’s guiding principle is to support ‘people, not projects'" www.hhmi.org/about/policies

Policies | HHMIView HHMI's policies, including research policies and operations and conduct policies.www.hhmi.org

The panda is courtesy of our hosting service. Moves to replace it with something (not my image!) are underway...

Yeah this is another issue. One ML expert told me “all LLMs are broken and need to be rebuilt” for that reason. The other strategy of course is using convos of LLMs to check each other. Sadly many will not do that and just trust outputs without verifying.

Meanwhile, many people would much rather read less/not at all or at least fewer papers. So I can't help wondering if the data : human narrative ratio changes. Maybe we get more science conveyed/consumed by machines with less frequent but more read narrative snapshots aimed at people 2/2

Hard for me to be objective as I value writing and it's one of the very few things I think I'm reasonably good. But I'm aware that a lot of people kinda hate it and feel it's just a tool/medium to communicate their science and they'd rather that be less effort 1/2