Greg Brandt

[Greg Brandt @greg_brandt]

@sbilstein The 5090 has already paid for itself!

[Greg Brandt @greg_brandt]

https://t.co/KdcWGGj3nc

[Joseph Viviano @josephdviviano]

me: "can you use whatever resources you like, and python, to generate a short 'youtube poop' video and render it using ffmpeg ? can you put more of a personal spin on it? it should express what it's like to be a LLM"

claude opus 4.6:

[Greg Brandt @greg_brandt]

@thdxr Yeah

[Igal Tabachnik @hmemcpy]

damn, the economy is brutal

[Greg Brandt @greg_brandt]

@ThePrimeagen Java 1.6

[Arnav Gupta @championswimmer]

So I have already started discovering a pattern about how "amazed" people are with LLMs generating code.

1. PMs, non-tech founders, wordcels and yappers are the most amazed. They can sit on their couch and write "make me an app" on their iPhone, and it makes them yet another landing page. They cannot stop waxing eloquent about how AI is changing the world.

2. Indie developers, script kiddies, and more junior developers also love "vibe coding". They keep saying "Fix it" to Cursor till it works, and mostly they like that instead of searching StackOverflow and figuring out what to copy from there, this is a much faster approach to coding

3. The senior engineers, neckbeards, old-school nerds wrangling with complexes codebases have mostly been saying "eh... it is good, in places. I don't like the code it writes, but I like how I can ask it to explain API docs and new pieces of code I have not seen before". They are using it to write some pieces of code in languages they are not well versed with, and they prefer 'copilot' mode more than 'chat and generate' mode.

4. The absolute elites (like Linus below), basically just say "it is all hype", and are not impressed at all with it, and are mostly ignoring it.

I think I am somewhere between 2 and 3 myself, and it was quite a bit of cognitive dissonance initially to find my friends from both 2 and 3 buckets give such contradicting signals about it. Then I started seeing the pattern. Those who are objectively worse programmers than me, were objectively more impressed by it than I am. Conversely those who I look up to as much better engineers than me, seem to be impressed even less than me with it.

So yeah, basically LLMs write code like a ~p75 programmer (probably ~p50 when codebase is large), and anyone below that mark is impressed by it, and those above it find it overwhelming, and mostly an irritating pair programmer, whose code they need to keep nitpicking.

[Greg Brandt @greg_brandt]

@n0w00j Code creating and managing rows in the DB still counts as frontend FYI

[Nathan Cofnas @nathancofnas]

Anyone under the impression that ChatGPT "thinks" is either delusional or unfamiliar with actual thinking (in most cases the latter).

LLMs are useful for a handful of activities like coding, doing basic calculations, writing doggerel, generating cartoonish images, churning out fake news, and conducting searches. They have zero creativity, ability to deal with genuinely novel problems, or understanding of what they pontificate about, as can be confirmed by anyone with a ChatGPT or Grok account. (See the screenshot of my conversation with ChatGPT where it fails to understand the concept of longer vs. shorter lines.) Training LLMs on more data (which increasingly consists of bot-generated bullshit, though that's immaterial) isn't going to usher in some history-altering singularity.

AI enthusiasts—including a few people who I respect intellectually—will say that in the past LLMs performed feats that skeptics said would be impossible, and there will be egg on my face when ChatGPT replaces me as a philosopher before turning us all into paperclips. But I don't think so. After being trained on 40 gazillion words, ChatGPT hasn't produced a single joke that's funny, or an original philosophical idea, or one sentence that doesn't sound like it was written by a glorified autocomplete (what ChatGPT is).

Many people are spooked by ChatGPT's ability to write an essay of the form "X is an interesting topic. Some people say A about X, but others say B. In conclusion, a good case can be made for A." The chatbot writes this kind of thing better than most people can, so it might feel like the machine has achieved human-level intelligence. But there is a qualitative difference between writing summaries of text in the style of a college term paper and writing something that's new and interesting. Most people write by following the same principles of ChatGPT: they respond to a prompt by shuffling around clichés and phrases (data) that they picked up from their environment. It's not surprising that computers can beat us at that game. But they're not beating us at what matters, namely, adding intellectual value.

Science tells us nothing about what what happens in the magical moment when the mind grasps an idea or comes up with something new. Maybe one day we will figure out how to instantiate that process in silicon. That day is not now.

@gregisenberg shared a story about his lunch with a 22-year-old Stanford grad who was struggling to finish sentences, grasping for basic words. The zoomer explained: "Sometimes I forget words now. I'm so used to having ChatGPT complete my thoughts that when it's not there, my brain feels...slower" (https://t.co/zW6EVt6kxy). Most people would balk at the idea of outsourcing their thinking to this extent to a human guru—even one who is wise and cares about their well-being. But for some reason we are okay with merging our minds with a trippy chatbot that can't and never will understand what it means for a line to be long or short.

[Greg Brandt @greg_brandt]

“I’m better than Wifejak, right babe?”

[Greg Brandt @greg_brandt]

@ryanflorence I knew you were my favorite frontend guy for a reason

[Sam Lambert @samlambert]

A principal engineer showing you around their home made framework.

[Greg Brandt @greg_brandt]

@bigblackjacobin @LevineJonathan @Slate I’m not sure Keynes was thinking of it in an active sense…

[Bennett's Phylactery @extradeadjcb]

here are two identical bell curves, separated by one standard deviation

we talk a lot around here about "failure to distinguish the general from the particular", but it goes both ways

[Aaron Levie @levie]

Somewhere out there the founders of Ask Jeeves are saying we told you so

[Steve Everley @saeverley]

With all the talk about #gasstoves and alleged health threats, it's worth providing some background on where the research currently is on this issue, and how the feds suddenly decided these appliances are a health risk.

🧵

1/x

[Greg Brandt @greg_brandt]

@Brotsu Lol unfortunately I don't think this is new...

[Martin Bauer @martinmbauer]

A thread on one of the most important sums in all of mathematics.

This result will surprise you! 🧵

1/n

[yammakazoo 🇺🇸 @yammakazoo]

@TBops42 @TheSmarmyBum Are these the guys from myth busters?

[Brian Chesky @bchesky]

Today, we’re announcing that Airbnb employees can live and work anywhere.

Our design for working at Airbnb has 5 key features: