Matt “Friend of the pod” Rogish 🇺🇸

[Matt “Friend of the pod” Rogish 🇺🇸 @MattRogish]

@staysaasy Yeah, it 10000% has the right architecture and answers compared to e.g. GPT5.5 but completely hand waves away the actual "draw rest of the owl" parts

[Matt “Friend of the pod” Rogish 🇺🇸 @MattRogish]

This is related, but I'm not sure if I'm right. Generally (this will be hand-wavy throughout), programmers split up into two camps:
1) The people that want to know (and do know) how everything works and
2) The people that just want to get things done

When confronted with a performance problem, the first will go back to first principles:
1. What's the algorithm we're using? What's the complexity/BigO?
2. What's the data structure look like? What's the machine architecture.

And they will deduce "Ah, it's probably slow because on 386 CPUs the way the registers are loaded conflicts with how we structure them on disk. I'll re-order the linker." and they run it and now it's 1000x faster. They will profile before and after to be sure, but the profile is a component of a bigger picture, not the place to start.

The second camp will profile each piece and iteratively try to winnow it all down. "Ah, this section takes 20% of the time. I'll poke around and see if there's anything obvious. Oh, we're looping twice. Remove it."

And they'll get incrementally better, but miss the 1000x improvement they could've gotten.

I think people trying to predict the future of AI and job losses fall into those two camps.

In the first camp, these people dive into how LLMs work. How are they trained? How do transformers work? What are those GPUs really doing, anyway? What are the constraints? How would we overcome them?

And they come to some conclusion, probably sounding something like: "LLMs are improving but the recent gains have been primarily through more and bigger GPUs, we're hitting diminishing marginal returns, and sans any sort of breakthrough, diminishing marginal returns to general models will continue. The orders of magnitude improvements will come either in specific scenarios or custom domain models." and think "Ah, my job is safe."

The other people only see the difference between point A and point B. They think that because line go up, it must keep go up, following the same trajectory. They get excited - or scared - about where line go up, because to them, there's no reason to think that line might not go up. It might flatten, slow down, etc.

I think - this is just a hypothesis - most people are in the second camp, and so are extrapolating far past the fundamentals of the whole system.

I am not an expert here, so I can't claim to be in the former camp, but I can say the ones I know that are in it, don't seem to be too worried.

[Joe Weisenthal @TheStalwart]

I like having a job. So consider this take to be drenched in cope.

But as of right now, I think that: coding being a relatively “easy” thing for AI to learn + the existence of many currently employed coders, implies that we’re a long way off from mass while collar disruption.

[Matt “Friend of the pod” Rogish 🇺🇸 @MattRogish]

I was an Ops 4.6 stan, but GPT5.5 is amazing

[Alexander @alexanderrX_]

forward deployed engineers were basically what every engineer was in the 90s and 2000s.

normal people who could code, speak to customers, understand the business and ship the thing.

we had to recoin the term because the average dev became a cave goblin who hates meetings, users and sunlight.

happy to see fdes making a comeback

[Arpit Gupta @arpitrage]

SpaceX is simultaneously three things

- An extremely profitable and promising space launch and telecom business
- A highly speculative and capital intensive AI firm
- A massive grift using exchanges and retail for exit liquidity

[Matt “Friend of the pod” Rogish 🇺🇸 @MattRogish]

I have no idea why "datacenters in space" can be a legit thing. I mean, clearly that's insane, right? It's a joke? There's almost no reason to go to those lengths to build shitty infra in the sky. Yeah, no, this is not true: "so no government can block it from making nudie pictures". You think USA won't stop you? Think again

[Matt “Friend of the pod” Rogish 🇺🇸 @MattRogish]

@james_clark Even 5.5 medium is pretty great

[Vashi Nedomansky, ACE @vashikoo]

They don't make opening credits like this anymore

[Odyssey @odysseyml]

Lost track of time again.

[Matt “Friend of the pod” Rogish 🇺🇸 @MattRogish]

Seems kinda funny how every task, no matter how big, ends up juuuuuust fitting into the context window. huh

[Luiza Jarovsky, PhD @LuizaJarovsky]

Humans in 2040:

[Matt “Friend of the pod” Rogish 🇺🇸 @MattRogish]

It's 12:27am and crypto is still stupid

[Siddharth @siddharthvader_]

https://t.co/1bTGqbhbP2

[Todd Spence @Todd_Spence]

Someone replaced the dinosaurs in JURASSIC PARK with PEE-WEE HERMAN and I love it even more now 😂🔥 (not created with AI btw)

[Matt “Friend of the pod” Rogish 🇺🇸 @MattRogish]

What is wild is I can never, ever spell hallelujah right, even though it's basically phonetic (hall-e-lu-[and that jah at the end thing]). I always type hallejuilah hallelughjah etc. bonkers

[Wes Bos @wesbos]

the four horsemen of the apocalypse

[Glenn Greenwald @ggreenwald]

The most darkly funny aspect of this video is that Ivanka Trump speaks as if she went on some profound, prolonged spiritual and philosophical journey to ascertain the meaning of life, and concluded that the most elevated human state is to live on a gigantic private island in the Mediterranean, in a sprawling mansion paid for with the billions of dollars given to her husband by the Saudis and Emiratis to ensure favorable treatment by him and by her dad in their exercise of the powers of the American Presidency. Real uplifting stuff. 🙏

[Thariq @trq212]

been asking others at Anthropic how they stay in the loop with Claude and fully understand the work being done

this is one of my favorites from Suzanne: