Kyle S. Johnston

@ksj.social

Uses ink for crossword puzzles. Views are my own; posts are never advice; links/reposts/likes are not endorsements.

how do you make workflows not suck as a team gets bigger it's just easier when people have access to things. but then there's risk with people having access to things i hate when people get blocked because they have to ask someone to do something, this is 10x worse now that we have agents

If anyone can suggest a good "how I used Prime Agent" article, that might help me to understand whether to try it. I agree that an IPython REPL has better properties than a stream of markdown. I want to see how that compounds into a different user experience.

right now i’ve got it setting up SQL Server on Mac, and it’s written a whole set of helper functions & classes to rerun various tasks “code is liquid” Prime Agent is the full recognition of that statement

i think Prime Agent is likely a watershed moment, there’s a ton of overhang captured just by a better harness RLMs aren’t new, but no one’s made a coding agent based on RLMs for big problems, it seems likely that the ReACT pattern has been replaced by RLM

There’s something like a “responsibilities treadmill” in middle-age where if the quantity of your responsibilities increases by 75% and you maintain a 85% completion ratio you have the same feeling of falling short even though you got done 75% more.

Some initial thoughts, and a complicated mix of feelings. Wow. I mean, Erdos problems are cool (I genuinely mean that), I didn't know about the Jacobian conjecture before it got disproved. But this newest batch from OpenAI hits home in a way the previous announcements did not.

the Dinitz-Garg-Goemans conjecture, a graph theory problem that’s gone unsolved for 30 years, was solved with this prompt

Construct a counterexample to general (non-planar) case of Dinitz Garg Goemans conjecture. You should do a breakthrough and find a structured counterexample.

I always liked Ben Evans. this is an especially good example of what a good version of Zitron's analysis would look like: axes of variation, how they could play out, this or that is more likely. as opposed to "I'm certain what will happen next despite having been wrong every day for three years"

Ways to think about token pricing — Benedict Evans

AI is in a supply crunch today, but what happens when we come out of it? How and where will supply, demand, price, capacity and capex get back into equilibrium? Today, model labs can name their price,...

ben-evans.com

I've returned from my writing hiatus to bring you some research I did on determining an explosive play threshold. Using scoring-probability added, I found the yardage at which a gain starts acting differently in terms of its impact on scoring. Read it FREE on my Substack 🔗⤵️

Bild