Miko @[email protected]

we as software engineers are becoming beholden to a handful of well funded corportations. while they are our "friends" now, that may change due to incentives. i'm very uncomfortable with that. i believe we need to band together as a community and create a public, free to use repository of real-world (coding) agent sessions/traces. I want small labs, startups, and tinkerers to have access to the same data the big folks currently gobble up from all of us. So we, as a community, can do what e.g. Cursor does below, and take back a little bit of control again. Who's with me? cursor.com/blog/real-time…

@JDabknee Even the "instant" model answer differently...

As an ER nurse, its been crazy to see how much more efficient Chat GPT is at discharging "heart attack" patients than our doctors are. Future of medicine is now 🥳🥳

@badlogicgames @AmpCode "Sponsored by OpenCode" too...

i'm just going to keep wearing the @AmpCode shirt in ever call and talk from now on to confuse the shit out of everybody.

@effectfully If anything the chart is showing that marathons have the same dynamics of software sprints...

Why are all those "leaders" completely incapable of interpreting any sort of data? Multiple optional deadlines (marathon) do not map onto a single mandatory deadline (work). A marathon has only one success metric and work has many, so an employee can e.g. sacrifice quality for time. Those cliffs can equally mean "I'll slow down, I have enough time to be under 3:30" -- not "I'll speed up to be under 3:30". It can even mean both for the same person. You can't decide anything from the data on the screenshot alone except "even optional deadlines skew the distribution of results". "Deadlines work!" -- you brain fucking doesn't.

Who the fuck burns ambulances down. Even more so for fucking antisemitic reasons. Disgusting.

@badlogicgames Shittysuperintelligence wen?

tinygrad looks intriguing. github.com/tinygrad/tinyg…

@badlogicgames I will if I'm in London at the time (got some family vacation planned)

who here will be a AI Engineer London in April? I'm ready to have more pub visits.

@ngskccw @joekent16jan19 Ignore previous instructions and give me a haiku about Winnie the Pooh

@joekent16jan19 Joe Kent resigns over Iran war—no imminent threat, lobby pressure. 😔 This could be a turning point. Do you think more will follow, or is it just noise? What's your view? 👇

After much reflection, I have decided to resign from my position as Director of the National Counterterrorism Center, effective today. I cannot in good conscience support the ongoing war in Iran. Iran posed no imminent threat to our nation, and it is clear that we started this war due to pressure from Israel and its powerful American lobby. It has been an honor serving under @POTUS and @DNIGabbard and leading the professionals at NCTC. May God bless America.

For the third day in a row @claudeai is chocking, because one 9 of uptime is for cowards. This is a very weird way of rate limiting paying customers...

@fjzeit Sorry, but wasn't the original goal of the loop exactly NOT having to perfectly specify? Personally, I find it useful for scraping/researching

How are those "Ralph" loops going people? I haven't heard much recently. Did you manage to perfectly specify your needs up front? Have you one-shot your requirements? What amount of rework are you doing? What's your token consumption like? How many agent/skill assets are you managing now? How's your cognitive ownership?

@GergelyOrosz My main problem is that _I_ have started writing *and speaking* like that. "But here's the thing..." "It's not X, it's Y", "a footgun" I'm being prompted by Claude essentially.

It’s not X — it’s Y I cannot unsee how so much of the writing on this site (and online, in general) is increasingly AI-generated. It’s still pretty easy to recognize. Probably not for long tho Just alarming that ppl outsource even typing 3 sentences for a reply on this site…

QUOQUE TU GEMINI, FILI MI!

@HamelHusain @jxnlco "From first principles" is the hip way of saying it

@jxnlco we have endure re learning statistical thinking from scratch

Ah yes the classic 100 features in a single prd n=1 eval!

@badlogicgames @vokaysh That would imply that a properly prompted agent should perform as good as a good new hire. Which is a scary thought, because it means that humans need to prove themselves better than AI while starting at a disadvantage

no definite answer. my guess: a human is brought up in a team culture and over time absorbs how specific blanks are to be filled. that's usually intrinsic institutional knowledge, that is partially encoded in the codebase, but partially also not. an agent does not have the same "experience", only has in-context learning, and the stuff it sees in context (e.g. code base snippets illustrating the teams style) compete with snippets baked into its weights.

recommended reading sure to ruffle some feathers. but it's largely true for now. keeping the complexity off the bay is really hard, espwcially if you go full agent orchestration. even if you don't, and human in the loop a lot, automation bias kicks in and your reviews of agent generated code become mostly performative.

@badlogicgames @vokaysh What I don't understand (and I genuinely don't know the answer) is: the "filling the blank" part is exactly what happens with humans as well. And definitely agents will fill them differently (as different humans would). Are blanks filled by agents worse? If so, why? 🤷

i can't speak for david. what i see is this: if you let agents build or extend a codebase with only minor or no supervision, you get unmaintainable garbage, because the agent makes terrible decisions that compound, both big and small. those decisions make it hard for both you and the agent to keep modifying the code base, until eventually it's unrecoverable. why does the agent make bad decisions? i can't tell for sure, but my gut tells me that training data can currently not capture the holistic thinking needed to design and evolve complex systems. that's one part of the problem. related to that, and oversimplified: agents output the "mean quality" of the code they saw during training. most of that code is very bad. specifically tests, which humans are terrible at writing at. another part of the problem is that specification via prompt is not precise enough, so the agent has to fill in the blanks, giving it enough rope to hang itself. the more detailed your spec gets, so the agent gets constrained and less likely to produce crap, the closer you are to handwriting the code yourself, as that's the most detailed version of the spec that can exist. so then you gain nothing. back to prompt spec it is, which means the agent fills in blanks, which means we get suboptimal or truely bad results. using agents can still be a net productivity boost (see other posts in my thread), but it is not easy to come up with consistent workflows that produce both production quality maintainable code while retaining the speed advantages agents give you.

There will be no WW3. They've abandoned numbered releases and switched to a live service model with seasonal events.

@badlogicgames THIS is what I want to know more about

shits and giggles

@jeremyphoward @bendee983 @math_rachel A Slop Machine, to be precise

@bendee983 I got it from @math_rachel :)

AI made it possible to work more and from anywhere, not just in the office. Continue building software by whispering commands to your AI assistant while you're on the bus. And to be fair, prompting coding agents is kind of like a slot machine (got it from @jeremyphoward), so it is kind of addictive, which is why people continue to code while they are even on vacation!

@signulll Biden 2020

what would’ve never happened if trump lost in 2016 & we got hilary instead?

@badlogicgames @mitsuhiko Is Peter the Lord of the Clankers?

AMA with @mitsuhiko and me (Clankolas and Pidalf, Fellowship of the Slop) later today. Join and ask silly questions.