Zach Hensel

I don’t know if this response indicates self awareness or just awareness of what answer I wanted and hallucinated some likely errors

Same prompt for whatever free chat gtp is doing these days for comparison.

I banished the image and chat from existence and here’s another with a different set of errors from the same prompt as best I can recollect.

The notation hinting that this is difficult because it’s a dumb prompt wasn’t there the first time 😂

You do not want to see the hallucinated structures up close.

Go big or go home.

My n=1 experiment says Gemini has no qualms about defining a hydrophobic/polar split which seems like a bad idea in the first place.

What's supposed to be disingenuous about it? Blog post says there wasn't a front page story and there wasn't.

>It’s like the domino theory, but in reverse.
>Yeah, yeah. Exactly.

There's a cost to maintaining privacy; the information is valuable. I bet most think it's worth it, though.

Yup; I figure there will be pushback and a demand for European-type privacy regulations. Just need to figure out how to enforce them, which rarely happens in Europe.

Selfie age verification without an ID already a thing; depending only on benevolence of companies and/or existence and enforcement of privacy laws to not link to your identity.

Until pretty recently this was still eerily accurate to me; running text through a bunch of random translations and eventually back to English with near unrecognizable text and still getting the same results was interesting; wonder if a modern approach would perform similarly...

Here's the mid-2000s version of a not very different task (predict authors from scientific paper text): jane.biosemantics.org/index.php

And the paper: academic.oup.com/bioinformati...

I guess, on net, democratizing fast online identifiability is perhaps better than having people mostly be unaware that that capability is available to many companies (and presumably governments that don't sell the data graphs).

Someone should let Kevin Durant's PR team know that this means it's also possible to fabricate someone's pseudonymous language

Ditto for improvising a plan to get ships moving again. Would’ve been announced on day 1 had they planned this out.

Predicting protein-protein interactions (PPIs) at proteome scale can take months with co-folding models due to the massive all-vs-all comparisons required.

We are excited to announce FlashPPI, a contrastive learning framework that predicts proteome wide physical interfaces in minutes. 1/🧵

Hopefully people figuring out how to regulate AI don't agree with Google/DeepMind people to build a regulatory moat hamstringing competitors.

Everything announced since AF-multimer was a sales pitch for the next set of multibillion dollar exclusivity deals. Openness of AF2/AF-multimer quickly proved how valuable the state-of-the-art model can be; makes sense to pivot at that point.

actual prompt shown 🤷 but don't worry I fixed it

no problem here's a perfectly appropriate one

Post civil war, Nixon pardon, and Oliver North the celebrity, etc says we we’ll always decide to move on

Since this crew is all about transparency, let’s see the peer review history for previous submissions…

Authors claim they added a sensitivity test with a lag, presumably in response to peer review elsewhere since it wasn’t in the preprint. But weirdly they only consider a 7-day lag.

At 8 days and beyond, the effect is significant 👀

Take for instance this article in the first issue of the journal he founded — publichealth.realclearjournals.org/research-art... — on top of being unrandomized, they simply measured the wrong thing (didn’t account for lag between exposure and eventually reporting a case).

What, N=30 to determine specificity when the unadjusted positivity was 1.5% wasn't enough for you? 🤯

This came out of nowhere. What a lame reflex.

Don't need a control group to see that JB's baseless assumption about your proposed research proves there's bias in medicine.

Very weird format to have people go back and forth about this without just asking you for clarification.

Looks like there's an audience that's now bigger for Netflix and the rest of WB abandons it.

So I am happy to hear that @frontiersin.bsky.social is taking issues with the paper seriously: methods falsification and plagiarism. I hope @uoregon.bsky.social takes it seriously as well.

More details @pubpeer.com

pubpeer.com/publications...

This paper was what Harvey Risch @yalesph.bsky.social used to advise Sen. Johnson that Moderna engineered SARS-CoV-2 to cause “turbo cancer” — here’s what Risch is up to now: www.wired.com/story/risch-...

They also covered this for some reason. A conspiracy theorist that the lab leak crowd mostly wants to go away because too much of his conspiracy theories take place in the USA.