iiinova — where your digital self keeps evolving

it's a great idea if it's in addition to, not in place of, printed references

040

i hate it, but part of me is glad that those distinguishing features still exist. like yeah this stuff is polluting our common spaces, but also it's nice to be able to tell when someone is actually putting in a bit of effort.

110

it's nuts. have you listened to the prior seasons of in the dark?

000

Patrick Wyman @patrickwyman.bsky.social · 12/10/2026

i burst out laughing when ryan was struggling to describe what a billionaire's bunker might be like and landed on "... i've got all these herbs and stuff growing"

130

Reposted by abir

There is absolutely no viable path forward for this country that doesn’t hinge on extreme accountability for everyone involved in this administration. They’re going to whine about politicization and credulous media types will echo them, but the thing is, they’ve committed tons of obvious crimes!

8570641790

i think terry tao's thing is silly but i am begging people not to overindex on the idea that "we don't know how drugs work"

240

yeah for sure! i'm sure there are efforts underway to reanalyze the entire corpus of data because an LLM doesn't "care" if the data point is a number in a 3M row table or a throwaway observation in the discussion section of. doubling the 1% success rate would still be a big deal!

but for a drug to go to trial nowadays you have to have put in a good faith effort to figure out how it works. i think it is definitely fair to say that we don't know why a lot of these candidates fail in trials (or work in rats but not humans!)

020

don't get me wrong, if you look through my publication history, there are a couple of papers where i've written something to the effect of "we don't fully know what these things are hitting in the body, here are some results demonstrating that fact"

220

The sorta glib but direct answer to your question is this:

i don't think "... generally there isn't an expert who really understands how [cancer drugs] work" is a fair statement.

abir @majumdar.bsky.social · 10/10/2026

whole bio is grating, but halfway through the first sentence i had to check the url to make sure i wasn't reading an entry in some watchlist

160

abir @majumdar.bsky.social · 09/10/2026

sam altman: guys i just found out we have to test the molecules. how many beagles can we fit in the conference room?

abir @majumdar.bsky.social · 08/10/2026

my devout hindu mom was deeply confused when i went vegetarian for non-religious reasons. she still frets about what to feed me every time i go back to visit.

2361

abir @majumdar.bsky.social · 08/10/2026

sam: i wish to be embedded in the model weights (his eyes dart around the semiconductor clean room) sam: jensen melt me into that b300

050

abir @majumdar.bsky.social · 07/10/2026

it seems like foster is just saying they need to prioritize in the short term? which seems correct because congress has other jobs. just kind of a boneheaded way to express it, i expect nothing less from a scientist who spent decades in academia.

110

abir @majumdar.bsky.social · 07/10/2026

did all of them have to call him "daddy" (this is a cool project)

abir @majumdar.bsky.social · 07/10/2026

do we think he'd be an absolute sicko if he put his mind to something else? or is he exclusively a baseball machine who is lucky he found the sport at an early age?

130

abir @majumdar.bsky.social · 06/10/2026

That's right! Symphony No. 8 with the Minnesota Orchestra in 2016.

abir @majumdar.bsky.social · 06/10/2026

Ok but the author is also a crank

000

abir @majumdar.bsky.social · 06/10/2026

This reminded me that next week will be a decade since I watched Skrowaczewski conduct a full symphony, 90 minutes, no intermission, from memory(!!!) just after his 93rd birthday and it was one of the greatest things I've ever experienced.

abir @majumdar.bsky.social · 05/10/2026

I remember reading "The $25,000,000,000 eigenvector" and being like wtf that's all it is????

160

abir @majumdar.bsky.social · 05/10/2026

i guess i ought to find this tweet unsettling because the spelling bee thing all but guarantees it's about people like me, but i'm laughing my ass off at "you don't take little league seriously unless they show remarkable early talent" this doofus has never been to a little league game

1100

abir @majumdar.bsky.social · 05/10/2026

put a sandwich just outside the schwarzschild radius and i'll make it out no problem

020

abir @majumdar.bsky.social · 03/10/2026

yeah seemed unlikely. i didn't realize how much of a social aspect there is to the terminal, but it makes sense.

000

abir @majumdar.bsky.social · 03/10/2026

george have you bought anything on posh

121

abir @majumdar.bsky.social · 03/10/2026

no, in this case i meant it more generally, as in "i have also noticed that they're really good at making connections across huge datasets in my own work". but i can understand how you got there from the way i posted this. fwiw i thought the RT "paper" was stupid lol

abir @majumdar.bsky.social · 02/10/2026

it's funny because so many of the replies and qts are like "ok i don't need to keep on top of particle physics either, why do i need to understand this?" you don't! will's ask is that you stop listening to the people who insist particle physics doesn't exist.

170

abir @majumdar.bsky.social · 02/10/2026

yup this was the thing that convinced me that they're not useless

250

abir @majumdar.bsky.social · 02/10/2026

aw hell yeah tim chidders is back

abir @majumdar.bsky.social · 29/09/2026

this is a very good joke

000

mr. shinyhunter if you're listening all you have to do is take him to an islanders game and drink 17 beers with him. water under the bridge and he'll even give you a punisher challenge coin.

030

*although if your bet is that this is a net negative for society, i feel like that's also a good reason to be up on the technical details? LLMs are going to facilitate deeply stupid and/or evil things, probably good to keep an eye out for that.

yes. but also if it's not their beat, i'm totally fine with them not caring about the details* so long as they don't blast out laughably inaccurate things to their tens of thousands of followers.

220

lmao the secret to parties and dates is just to ask non-work* related questions and to not interrupt when the other person answers them *this includes history!!!

100

i am deeply sorry for putting this into the ether

000

it doesn't take bobby three sticks to see / guys it's time for some game theory / drinking vodka by myself / it's kompromat i can tell

100

the first one. that's not to say that the internal chat log thing absolutely didn't happen (nothing is off the table with these companies), but i'm confident that they wouldn't have needed to do it that way given who was prompting it and how many tokens they were allowed to burn.

000

right. it was also being driven by people who are themselves domain experts.

frankly, i think he's just wrong about this. claude can generate plausible hypotheses that would impress me if they were presented in a ph.d. qualifying exam. it also does stupid shit all the time and i think anthropic 100% deserves the scrutiny lol

160

abir @majumdar.bsky.social · 26/09/2026

what type of bug is that

100

kills me that the first bengali government official anyone in this country has heard of is this doofus

030

abir @majumdar.bsky.social · 26/09/2026

they can be very useful! as a reformed AI hater who still despises the industry, i support you in this.

yeah i also got really annoyed at him for last month's "this AI created new viruses" story about the synthetic phage. he's a good reporter. i just think the "intersection of AI and ______" is a beat that most people haven't figured out how to write about responsibly yet.

020

*Zimmer, but yes, might as well just work for anthropic, at least they'd pay more for the PR copy (for now)

good news is by firing lafleur, you might be able to confuse flores into dropping the blitz rate to 70% in november

090