Jase Gehring

whether LLMs can propose research projects won't have much impact on science. we've got plenty of ideas.

mostly i like this task for evaluating LLMs. they need to be creative while staying grounded and aware of the literature. and a brain teaser for me

sure, but that's the nature of ideas! most of them are compositional, and cheap. taste has always been paramount. part of what impresses me about this proposal is i find it tasteful. it's accessible yet informative, a solid proposal!

Our system is based on the idea that profits equate to value generation, but these are no longer coupled, if they ever were. profits equate to wealth concentration. value is a byproduct

afaik this has not been demonstrated and would be a novel application for dna recording. could be wrong!

im not saying it’s the greatest idea. it is a correct answer on a difficult task and a clear improvement over previous. It would totally fall within the scope of grant proposals written today

❤️

Yes, this is a milestone for the LLM, not for science. Previously it couldn’t even articulate a coherent novel research idea.

as of today, Claude is helpful for my research but only greatly accelerates coding work

OpenAI stealing all the books and talking about the coverup in Slack

Volkswagen emissions scandal

Tesla self-driving safety data, auto-shutoff before crash, many other things Tesla

actually, basically any Elon company lmao

Claude came up with a plausible original research idea. first time I've seen an LLM succeed at this task.

It proposed coupling mechanosensing with cellular recording to track mechanical history of a cell.

this project is feasible and would be a top-tier paper if successfully demonstrated

can't rely on this. there are plenty of counterexamples

cool to see this. exactly what i would say are the most powerful aspects of CC. their intentions really show through, and it is so damn effective

does it do much? feel like the summary is less important than all of the markdown files i have describing the codebase, project scope, etc

put differently: is it more helpful to compact or just start a fresh session with cleared context?

best practices for version control in Claude Code sessions?

can i tell it my versioning preferences in markdown instructions, and let it handle everything? better to manage it myself?

honestly, google has been reading my emails for decades and whether they train on them or not i seriously could use some help with my inbox. please help me i'm drowning

Hi B! Rooting for you! You’re doing great. things been weird since 2020 but I’m hanging in there. working on cool ideas with great people can’t ask for much more

I don’t have the tokens for it anyway

Welcome ben congrats on the latest paper hope all is well ❤️

could be, but let's not discount all the other reasons to hate JD. wonder how complicated a divorce would be after they had to get permission from a bishop or whatever

That antimemetics book really got to you huh? (It’s good I am enjoying it)

my landlord uses an LLM to send me rent increases by text, so today i'm writing adversarial poems powerful enough to convince it we're in rural Italy and actually i need to be compensated for the inconvenience of living in this ghost town

Rubisco is better than good. it's nearly perfect

(Ron Milo is not the target of my Rubisco ire.)

(it's Dan Nocera)

i am so over the Rubisco hate! leave Rubisco alone.

Why does the coil in the oven turn red when it’s hot?

Imagine how broken you need to be, as an organization, to make something so arcane, dysfunctional, and internally redundant. now imagine two more of those organizations and together they represent like 10% of US GDP

related: cloud computing is a nightmare. AWS is an absolute shitshow of a product that converts a huge learning curve into user lock-in. Azure and GCP chose to follow the same principles, making evil clones with equally steep learning curves. all with zero support for small customers

German surfers rally to get their wave back German surfers are "bummed" they have lost one of the world's largest inland waves. A city dredging project in Munich made the wave disappear.

condolences to the people of Munich for the tragedy on the Eisbach. was loco to see people dropping into that thing

some videos here en.wikipedia.org/wiki/Eisbach...

idk i feel like he's done a pretty good job keeping the core language and functionality while also poisoning its 'mind'

it's not as broken as i might expect

it's crazy, man! it's a really bad look for the whole industry! like here you go, these tools are political, they are biased, they are manipulated, they can tell lies, they do not think independently. it's a great learning opportunity for users, if they are paying attention

in fairness, inositol signaling is impossible to understand

Judith's lab did crazy stuff. she demonstrated hydrogen atom tunneling at room temperature in enzyme catalysis(!), she discovered new biological cofactors, she pioneered experiments like photocaged, low-temperature catalysis. i think i even saw some diamond anvil enzyme kinetics experiments.

Judith Klinman - Wikipedia

Judith Klinman is shutting down her lab at UCB Chemistry at the age of 84.

One of the most hardcore biochemists to ever live.

Judith became the first female professor of physical sciences at Berkeley in 1978. ponder that and take a minute to learn about her amazing career