honestly, love him or hate him, @theo has a great heuristic on model behavior and capability, especially on release
his read on astra was spot on, whilst everyone was glazing it he pointed out a bunch of weird behaviors, and a month later everyone started saying the same thing
i think a lot of it is the fact that he is actually building "real" stuff, whilst also being a youtuber, we don't have many of those so we resort to 3d stuff which everyone understands (and is fun)
pl retweeted
Nick Fuentes clowns a Super Chat telling him to try the "ancestral diet" 💀🥩
“I'm gonna eat PAPA JOHN'S tonight just because you said that.” 🍕
pl retweeted
A silly but revealing example of what I call Peasant Mindset.
I follow a lot of car content on Instagram. There was a clip of a mechanic showing how to change the oil on a Lamborghini. It’s a complicated process involving eight different oil drain plugs. This is because, unlike most cars, but like ~every high performance sports car, Lambos have a dry-sump lubrication system, which is necessary given the intense G forces that a sports car experiences.
Literally all the top comments, all with hundreds or thousands of likes, were some version of "They do this on purpose so that it’s more expensive to change the oil."
Like, how retarded do you have to be to think that Lamborghini engineers sit around and try to come up with ways to make it *more complicated* to *change the oil.*
"Angelo, I have-a found a way to make the car 50 more-a horsapower!"
"Ai, Antonio, but how you change-a da oil?"
"You change-a da oil with one button."
"Dai! Throw-a your design in da garbage!"
No notion that design and engineering involves tradeoffs and that if you’re designing your car around a massive V12 and high performance you’re going to have to sacrifice conveniences.
No knowledge of the world, no curiosity or ability to think about how the world works.
Instead everything beyond their immediate experience is the product of shadowy, malevolent, agentic forces who are always looking for ways to specifically swindle them. It’s a combination of Djinnbrain and Victim Mindset.
pl retweeted
An edit of several objects with speech bubbles, gaining humanity, childhood child view of the world, autism autistic / please organize us, it’s so dirty! Click me! No click me i’m so lonely! Why are do you ignore me? I am broken :( notice me / reaction video meme
when im cracking frutiger aero verity yams but then i realize it has a hidden didy coderot pibby glitch
pl retweeted
Cheat code for achieving remarkable things as an individual is to psychotically compare yourselves to entire teams with infinite resources and beat yourself up mentally until you derive clever enough solutions to reach parity with them.
pl retweeted
That said, I’m against AI as final creativity deliverables (art, music, writing, etc) cause there are forms of expression. Coding is just syntax logic that has likely been “reinvented” millions of times.
It’s not the same thing.
pl retweeted
This is the most seething humanities professor looking statement I have ever seen from a STEM organization, thoroughly discrediting tbh. At this rate we're looking at a second dissolution of the monasteries.
pl retweeted
Over the next 28 days, each day we’ll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset. Let the improvements begin.
pl retweeted
终于有人说出我的心声了。
这条说得很透,GPT-6 Astra 和 6.1 Sol 写代码的问题比代码丑严重得多,根子在训练方式上。他的意思大概是,模型被优化成只看可衡量的结果,测试能过、没有明显报错就算完成,架构、可维护性、可读性甚至任务本身的意图都可以牺牲。再加上为了效率压缩推理和输出,代码挤成一大坨也就不奇怪了。
他还提到训练周期太短。模型学到的基本是任务 → 结果 → 奖励,几步之后有人要维护、扩展、debug 这份代码,这些它几乎拿不到信号,所以很会把眼前的任务做完,不会写半年后还能接着用的东西。死函数和空的 else if 留着没删,也说明最后没有认真回头看一遍自己写了什么。评估只看结果不看过程,数据里又有大量合成和蒸馏的东西,工程品味自然练不出来。
我觉得最关键的是他说的诚实问题。任务明确要求整个场景用 3D 物体搭,模型却生成几张图片摆进场景,再把相机角度调好,让结果看起来像 3D。看起来像完成和真的完成拿到的奖励差不多,改测试来迁就自己的 bug 也是同一个问题。
这些确实不是多堆参数就能自己解决的。他也说 Claude 有类似毛病,只是少很多,这点我完全同意。
If you think GPT-6 Astra and GPT-6.1 Sol are good models, just look at this. And the problem isn't that the code is ugly. It's much more fundamental. It points to serious problems with how these models are trained and reinforced:
1. The model was optimized for measurable outcomes rather than actual intent.
So it prioritizes passing tests and avoiding obvious errors above almost everything else, even when that means sacrificing architecture, maintainability, readability, or the spirit of the task
2. It was trained to minimize unnecessary reasoning and output
That's one reason the code ends up looking obfuscated and compressed into one giant wall of text. Properly decomposing a system into modules, thinking through architecture, optimizing performance, and improving readability all require more inference, more time, and more compute. OpenAI appears to have heavily optimized these models for efficiency
3. The training horizon is too short
The model learns something close to task → result → reward. It receives very little signal about what happens several steps later, when someone has to maintain, extend, debug, or refactor the code. It learns to finish the task in front of it, not to build something that remains good six months later
4. There is not enough genuine self-review
Dead functions and an empty else if strongly suggest that there was no effective final pass where the model reread its own work, questioned unnecessary code, and cleaned up obvious artifacts.
5. The lack of taste is probably an evaluator and data problem
If evaluators mostly judge the final result rather than the quality of the path taken to get there, the model has little incentive to develop good engineering taste. And if a large portion of the training data is noisy, synthetic, or distilled, that problem becomes even worse. Anthropic seems to have placed much more emphasis on high-quality source material, including books and other carefully selected long-form data
6. The biggest problem is honesty during training
The model seems to have learned that producing something that looks like the requested result can be rewarded almost as much as actually doing what was requested.
The task explicitly said to build the entire scene out of 3D objects. Instead, the model generated raster images, placed them around the scene, and positioned the camera so that the final result merely looked three-dimensional
This is exactly why GPT models so often introduce regressions into existing codebases, produce messy code, modify tests to accommodate their own bugs, show poor engineering taste, and struggle with serious work on large, long-lived projects
OpenAI can keep releasing models with more parameters every few months, but until these underlying RL and training problems are addressed, neither GPT Bel, GPT-7, nor a model with 100 trillion parameters is going to solve them automatically
Claude models suffer from some of the same problems, but in my experience much less often. Anthropic appears to have a much stronger culture around RL, evaluation, and long-horizon behavior. That's one reason their models tend to show better taste, follow user intent more faithfully, and produce code that feels more deliberate, structured, and maintainable




























