Martín Lucas Golini

Creator of ecode (lightweight code editor) & eepp (GUI framework) • CTO @ Llamando al Doctor

[Mitchell Hashimoto @mitchellh]

I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem.

As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)!

I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work.

It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results?

88ms => 1.5ms
150K allocs => ~500 allocs

Incredible right? Nope.

My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path.

This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput.

The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity.

Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.

[vaxry @vaxryy]

supply chain attacks?

yeah no, I use C++. We don't even have a package manager.

[Martín Lucas Golini @SpartanDev]

ecode 0.8.0 released!
Big update: diff viewer, full Agent Client Protocol support, live Markdown preview, terminal reflow + many improvements and bug fixes.
https://t.co/YiIg7M4ocC

[Mitchell Hashimoto @mitchellh]

I’m excited about the AllBirds AI pivot because this is the first completely pure act of retardmaxxing in the public markets and I’m so on board with watching that journey play out.

[Martín Lucas Golini @SpartanDev]

I don't think this view is reasonable from the consumer PoV, they're just saying "it's too expensive to run, so we will nerf it if needed", you know demand of a good SOTA model will be high, just charge what's needed to keep it being good... nerfing it shouldn't be an option

[dax @thdxr]

when they launch a model say there's a quality range from 0-10 they can pick for quality of model for the price they are charging

when they launch they have to guess - say they pick 8

it's arbitrary, it's not the "true" state of the model it could have been better it could have been worse

then as demand shows up they need to adjust up or down

since demand has been booming it tends to be down so maybe they realize equilibrium is at 6

it's really just a price discovery mechanism that's happening

[ThePrimeagen @ThePrimeagen]

guys, i honestly do not like clowning on Gary.

I don't find being the butt of a joke funny, so I imagine he does not either.

But, this is what worries me about where we are going. We are actively encouraging an entire generation that the tech is there when its not, and a couple of silly mistakes made on a website isn't the end of the world, but people's data and breaches are serious. We are entering a very VERY hackable world, and I do not like it one bit.

[dax @thdxr]

you're probably underestimating how crazy things are

[Mitchell Hashimoto @mitchellh]

I think one of the funniest things is how frontier labs spout how AI is writing their code, but the reality is that the products they ship are buggy, resource hungry messes. It's kind of the worst advertisement ever for both their products and their worldview.

[Martín Lucas Golini @SpartanDev]

@mackron @raysan5 No need to track the time spent on things you enjoy! Your libraries are a work of art and it's an absolute delight to use them. I hope people encourage you to keep working on them… sometimes open-source software is rough to maintain and people can be really ungrateful.

[sphinx @protosphinx]

AGI is not coming.

We are nowhere near AGI. What we have today is inference, not learning.

Models get trained once on huge fixed datasets, then frozen. You ask questions, they remix patterns they already saw. Nothing updates. Nothing sticks. Talking to the model does not make it smarter. It does not learn from you. Ever.

Learning is still slow, expensive - and offline.

Look at self driving. You drive around a pothole, make a U turn, and come back. The car’s AI does not learn that you just solved that exact problem. It reacts the same way every time using sensors and rules. Do this 20 times a day and it still has zero memory that the pothole exists. It just re sees it. That is why edge cases never die. There is no local learning. No accumulation.

No 'oh yeah, I’ve seen this before'

LLMs work the same way. Tell it your name and it does not remember. The only reason it looks like memory is because scaffolding keeps shoving your name back into the prompt every time and sanitizing the output.

The model itself has no idea who you are and cannot learn from interaction. It is structurally incapable.

And the scaffolding is the worst part. It is pure duct tape. Just prompts on prompts on prompts around a frozen model. When something breaks, nobody fixes learning. They add another layer. Another rule. Another retry. Another evaluator model judging the first model.

So you end up with systems that are insanely complex but mentally shallow. Debugging is hell because behavior comes from hack interactions, not a learnable core. Tiny prompt tweaks cause wild behavior shifts. Latency goes up. Costs go up. Reliability goes down. None of this compounds into intelligence. It just hides the cracks.

Until we have real persistent learning and real memory inside the system, there is no AGI.

LLMs are not built for this. You cannot prompt your way out of it. You need a totally different architecture. Yann LeCun is right.

And even then, what architecture can actually learn online, store memory, and stay stable on today’s hardware?

Best case, maybe 5-10 yrs.

Right now it is all inference. It looks magical, but the emperor has no clothes. A lot of people see it. Almost nobody says it out loud.

[Grummz @Grummz]

A note about Moltbook.

It's hugely popular now, and this means there will be a lot of bot users, telling their bots what to say and ask. There will also be users pretending to be bots, trying to goad conversations to happen. For lulz, or content to post.

Be careful, not all is as it seems.

[Bryan Johnson @bryan_johnson]

I can’t help but think (and feel) that the world is generally very sad right now. Injured really.

Yesterday I was in Utah with family. Three generations. We played sports, enjoyed good food, saw friends, and just messed around all day. One of the best days in recent memory for all of us. This is where I grew up. It took me back to my childhood. Allowing me to embody those psychological states and feel the comparative difference between then and now.

The hollowing and sadness of the modern world seems to stem in part from our phones, social media, and the ferocious need to be seen and relevant in every moment. We have mistakenly idolized a specific kind of dysfunction: a manic, sleepless hyper-vigilance that needs to be omnipresent.

Everyone I know who’s unplugged for a week, returns reporting life-changing levels of improved life satisfaction. I’ve never met anyone who didn’t return feeling spry and vibrant and clear-eyed about the corrosive nature of current social culture. The science supports them feeling that way. They were in a dopamine deficit from the hyper-stimulated state of the world so everything felt gray. 

So why don’t we unplug more and more often? We’re all kind of trapped in a prisoner's dilemma. Most want to move to the mountains and be relieved of it all but are terrified that if they unplug, they’ll be invisible. Real life consequences of reduced power and status. So we stay plugged in and drink the poison. This hypervigilant state keeps us in chronic fight or flight (anxiety). Simultaneously, our addiction creates a dopamine deficit (the emptiness/grayness feeling) and a background hum of anxiety.

Mammals are biologically hardwired to co-regulate: physical touch, eye contact, proximity and in-person vibes. Things which release oxytocin and activate the vagal nerve's parasympathetic system. Screens eliminate all of this goodness.

There are small wins to be had here. More in-person time. A day off technology per week. A block of 4 hours. One hour before bedtime.  I hope that there’s a collective awakening that we’re all being mined for engagement. Then we get trapped. And then trap each other.

[Martín Lucas Golini @SpartanDev]

ecode 0.7.4 released!
Tons of work went into proper text-shaping, segmentation & dozens of bug fixes. No crazy new features this time, but just a better, smoother overall experience.
https://t.co/t49JbWzgyH

[Martín Lucas Golini @SpartanDev]

@martinvars @Condeexplorer1 Hola Martín! Nosotros en "Llamando al Doctor"(https://t.co/A9ZFKAlvga) tenemos médicos las 24hs en todo LATAM y Europa. Si necesitan podemos ayudarlos!

[Timur 𝕏azamov @nanot1m]

it's fascinating how good @tsoding is at explaining different stuff

this one is gold
https://t.co/R33ZKv3pt7

[Eric Alper 🎧 @ThatEricAlper]

An employee cleaning the floor of a Roomba store

[Mitchell Hashimoto @mitchellh]

I just want to log in without being redirected 42 times or logged out every single day. I want to remain logged in on my device for at least months. We have machines that can mimic sentience and yet we can’t do log in for more than 24 hours. We’ve been played for fools.

[Martín Lucas Golini @SpartanDev]

@platonvin @nixcraft Mem. bandwidth used to be a problem in mobile, but today even the most low-end devices are capable to render extremely fast with the current renderer (ecode has several renderers actually). I'm not sure if the optimization mentioned would be very effective, although I'm curious.

[Martín Lucas Golini @SpartanDev]

@platonvin @nixcraft You're more than invited to collaborate if you're interested in improving it! :) The current pipeline is generic, everything is treated as a non-rectangular quad, those could use a different path, but in my experience, in 2D, changing shaders frequently is too costly...

[Martín Lucas Golini @SpartanDev]

@platonvin @nixcraft I'm sure there's plenty of space to optimize the rendering pipeline, as you can see is pretty simple. But, here on average I get ~0.5ms / frame, which explains why I haven't invested more time in that. There's too much to do and I'm currently the only one working on the project..