Simple trick for those outside of academia. Google the impact factor of the journal you are reading. If it's below 5 it's fake. If it's 5 or above, it's also fake.
Latent Labs (a $200M company) launched an "autonomous AI agent for drug design" that appears to wrap open-source models behind a paywall. Tristan Farmer built an alternative with Claude Code in a day and has interesting things to say about it: https://www.linkedin.com/posts/tristan-farmer-973b7a17a_anti...
Underlying Foster & Rahmstorf paper is also on the front page (https://news.ycombinator.com/item?id=47275088). This Nature piece adds context and estimates from other experts in the field.
To me, model card makes sense for something like this https://x.com/OpenAI/status/2029620619743219811. For "sheet"/"brief"/"primer" it is indeed a bit annoying. I like to see the compiled results front and center before digging into a dossier.
In the article they make that 2025 is a tipping point where open source frameworks and libraries "just work", making speedy, fun development possible without needing to fight the clunkiness of heavyweight engines.
And any new stuff regarding Celeste or from their devs will forever be relevant to me! Highly recommend to any who haven't played it.
I like to think that all these pelican riding a bicycle comments are unwittingly iteratively creating the optimal cyclist pelican as these comment threads are inevitably incorporated in every training set.
The truth about antidepressants is that the majority of people with depression that respond to an antidepressant would also have responded to a placebo. This doesn't mean that their depression isn't real or that antidepressants "don't work". It just means that placebo has a relatively high response rate in trials for depression. The hate is (among other points) because they are only arguably, marginally, better than placebo, and antidepressants also have real side effects (activation syndrome, increased suicidality, sexual side effects, withdrawals, etc.) over placebo.