i asked claude opus 4.5 to write some prompts that could work for both ai and humans.... such little respect for humans!

new prompt in 3 pieces. me: what is the right "role" that you should assume to help me think about this? scientific communicator? ui engineer? marketer? statistician? scientist? something else? claude: <answers intelligently> me: great. what are this person's priorities in our setting? 1/2

always a good day when you can get claude to swear

One thing about PCA/embeddings/political-leaning that should get more attention is the role of zero or “the origin”. It’s often special in a way that depends upon how you do the embedding. This post is a good example of that. Once you accept that the origin is special, then….

claude code just rejected a request i made due to safety reasons (request below). i was reminded of this conversation wife and i had driving in michigan UP

Google in Claude???

“[because of LLMs] it’s become so hard to tell what’s slop” - @alexpghayes.com

claude, what the heck

A conversation I had with Claude about extrapolating enmeshment to cultural and political scenarios, whether it applies to Trump supporters, and paths forward. claude.ai/share/92446e...

ABANDONED THE TODDLER IN THE BACKSEAT Arrest these motherfuckers for child endangerment

i got claude code to swear!

Sometimes we don’t yet know the question. That doesn’t mean our curiosity isn’t part of science!

Causality is a model

But they can be much heavier.

Me: Eli, what do you want in your lunch this week Eli (age 3): ordinary demigods

I believe science is much more like art than many scientists are willing to consider. I have a feeling that it would free us to do better science if we embraced this view. I may be mistaken but that's where I am mentally.

If their response is worse than Sonnets, then they should lose their jobs. Already it is too late. This should take 20 minutes.

(bluesky user bursts into Waffle House) OH SO YOU HATE PANCAKES??

What if, bear with me… fevers in pregnancy were bad. Then, we might only identify these cases by who took acetaminophen. If that were true (just my hypothetical)… but it strikes me that this would be confusing for bad statisticians

advice from LLM: "You won't regret this at all - it's the textbook solution for development databases with schema issues!" should i believe it? lol

to do good science, you need to fall in love with your dreams. everyone always emphasizes the flip side of rationality, which is also kinda true. but the reviewers help you with that. your reviewers can't fall in love for you. and if you aren't falling in love, then stop.

“Duty to shareholders” becomes exceptionally weird when that duty requires sucking up to Trump. Like all those CEOs fawning over him. It’s arguably their job to do that? But it plays very differently in the press and for public opinion.

Half baked thought: communication is always a weird crazy strong bottleneck. So, us sending a little text off to a data center for massive calculations and the text center sending a little text back is like way way off the charts on some metric of ratio between total computation and communication.

“Non-linear” is the first word people say before they go off the methodological deep end

My quote of the day What is objectionable, what is dangerous about extremists, is not that they are extreme, but that they are intolerant. The evil is not what they say about their cause, but what they say about their opponents. Robert F. Kennedy, Sr.

Is 9/11 now liberal coded? Or is it just me?

When I was a kid, I had a hard time believing that The Emperor’s New Clothes spoke to a real thing that could ever happen, like ever anywhere in history.

half finished thought... an advantage that engineers have is that no one questions the "constructivist" nature of their work... like clearly they create it. very deliberately engineered.

“Karl, why do you have such a bad impression of Causal Inference™?”

Somebody once said that part of the pleasure in buying a book is the mistaken belief that you're also buying the free time in which to read it... Just sayin'

It was and remains a very curious feature of the US presidential election that it was fairly transparently a "sufficiently many voters are mad about inflation" election but it was treated as a huge Zeitgeist Shift by both proponents and opponents of Trump. This led to some over-confidence from Trump

Is it a confounder or a proxy?

Seems like Epstein died, again

thinking "critically" or "objectively" is actually just a vibe. Lots of folks, when they try to vibe that way, they turn into cranks (anti-vax etc).

i think "ai makes me less smart" is analogous to the sentiment that engineers feel when promoted to management... both ai and management remove you from the more concrete things. there is something lost, but I don't think it is ~smarts~

Happy to announce I just defended my dissertation at Harvard and I’m now a PhD.

Proposal to replace "John Doe" with this

One of the best parts of my job is encouraging undergraduates to take bigger professional/academic risks. These risks are scary because no matter what, they will come out the other side different. They need to know that is ok. then, the risk is simply exciting

I now regularly ask the computer to explain jokes to me. Three years ago, this would have been so far beyond unfathomable to me

There are 12 = 4*3 paths through a standard four way intersection. Isn’t that weird?

Happy 4th of July, Batman.

DIY OTC GLP-1s OMG LOL.

abundance vs scarcity is the new PC2

we are in a weird period of acceleration where publishing does not make sense for lots of researcher. publishing is comically slow. saw someone post a picture of a book that was criticizing LLMs and the critique just... didn't fit anymore. this is a very exciting time to be alive

When "trial" is used to refer to an observational study it's actually describing the emotional experience of seeing people call an observational study a trial.

Authors, editors, reviewers, and other potential users should use CONSORT 2025 when writing and evaluating manuscripts of randomized trials to ensure that trial reports are clear and transparent. Learn more in this Special Communication. https://ja.ma/3TlnKQe

imagine if computers didn't have windows. just an entire screen of the terminal window. This is what computers used to be! And this is what LLMs are now.

This battle is old. Folks been trying to convert universities into trade schools since the beginning. However, every year the market speaks! Folks literally make their kids go into debt to do it. Off the charts user satisfaction because universities are a great place to grow up.