muzani

They'll promise that the love you and you'll do well, and suddenly you're flagged for low performance because someone used AI to check your work and nobody checked whether the work-checking AI is valid.

There's cognitive dissonance going on here - you are not worthy but you need the money so there's no choice but to be worthy. Your brain has both of these and it's freezing up. That's the part you need to fix.

I might not be worthy either. And possibly not even due to any kind of rot, but because all these brilliant people are competing for the same jobs and willing to do it for cheap.

Personally, I took up entrepreneurship because it's a lot easier to get funds these days and it's safer than all these layoff cycles. What's the point of a full time job if job security is a thing of the past?

Some people think that as long as it passes all tests, that's all jolly - but it still needs a code review to check whether all the tests are still relevant and whether everything that needs to be tested is tested.

Even if AI gets the PRD 100% right, the PRD may be wrong. Getting the communications right is the essence of software engineering in the end.

There's something called the oatmeal problem in the game industry: If you created enough procedurally generated content, it would become "oatmeal". Incredible variety individually, but bland when mixed together.

I'd say as data increases, it will default to being really bland.

Both Claude and ChatGPT were able to go far more edgy. Claude was a popular smut generator until that time when the US gov locked down Fable, and subsequently it was lobotomized, and ChatGPT was edgy until someone committed suicide on its advice.

They do have plenty of AI products too - RAG, embeddings, voice, companionship, image generation, video. They're not the best in most of these fields but it works well enough.

ChatGPT is also a very powerful platform that they haven't been able to fully leverage.

Or they simply haven't developed the taste to understand why it feels off, and likely aren't domain experts either.

But if they're not new to management, chances are that someone has given them the feedback.

And if they're still doing it after a while, that's a sign that they aren't open to it.

Then it actually went up and the whole memecoin space was taken over by grifters.

There's a delay to these things; the value of these communities went up because of the community and it collapsed because the community was taken over as well.

From the outside, looking at all these graphs, it's hard to link them.

Bored Ape was about some celebrities just having fun with technology. It was funny and the music videos were fun.

I was talking to someone about game development this morning. The scheduling to make a simple commercial indie game five years ago was roughly 20 hr market research, 10 hr gameplay prototyping, 10 hr art prototype, 20 hr vertical slice, 80 content & level development, 60 hr on polish. Plus some more on marketing.

Now today with AI, what has changed?

Those who follow a process may instead be spending 5 hrs on market research, 3 on prototyping, 30 on levels, and 100 hrs on polish.

These games are a little better and less painful to make. But the output is roughly the same as before.

Those who don't follow a process will spend 15 hours prototyping, 10 on polish, then find that nobody buys it and blames distribution.

But back in the past, these same people would have never reached a release point. The output of slop has multiplied enormously.

Love is the unconditional, unequivocal commitment to the welfare of another. This makes it very susceptible to abuse - you can simply use a loved one for endless money, sex, or labor for example.

Much of society has formed to protect individuals against the abuse of love, often to the point where we think love is evil. Restrictions on marriage, people are decoupled from their jobs, welfare is villainised. Unbounded love leads to tyrants and child abuse.

But love is still necessary for society to thrive. Love is by nature unbounded. And one of the hardest optimization problems of being human is deciding who to love.

Having one where you fill in the ink rather than buying a whole RGB cartridge has been a great decision. I used to replace a whole cartridge up just for reds.

It has life lessons, just straightforward things like "don't brag unless you know you're worth it" or "just because they look nice doesn't mean they are". It's challenging enough for them and they'll ask a lot of questions, and they're able to even debate the moral of the story.

They're all really short. Easily 5 mins for today's TikTok attention span. It's an entry level into more interesting books.

First sentence draws attention. Second attention emphasizes that it was not a typo and makes more questions pop up in the mind of the reader. The following sentences increase curiosity on the methodology used and so on.

"PURPOSE: Penile size has been a focus of interest in the life and the culture throughout human history. It has been a longstanding question whether there is any relation between penile size and the dimensions of body extremities. We measured the standard length of the penis and investigate whether any body extremity is a predictive index of penile size in Korea men.

PATIENTS AND METHODS

A series of 655 males above 17 years of age were examined during a 4-month period. Stretched penile length, penile circumference, and length and characteristics of various body parts and features (1st finger, 2nd finger, 3rd finger, 1st toe, 2nd toe, 3rd toe, ear, mouth, nose, height, weight, and baldness) were investigated by one examiner. To see the relation among the penile length and circumference and various body dimensions, univariate and multivariate statistical methods such as correlation analysis and multiple linear regression were applied."

Astra, Fable, Opus 5.0 are better with less instructions. Some of the context has been moved into the harness itself, so things like CLAUDE.md etc are becoming obsolete, and same with PRDs.

These days it's more about saying why than what. ADRs and GDDs would be helpful.

Similar with Claude 4 being worse than 3.7, but peaking at 4.5. At 4.6 they started to hit the restrictions and 4.7 seemed to patch what they made worse.

They sounded exactly like that person... but it's like they're angry at me.

Great toys but not production ready.

AI would still not solve this problem properly today, and may in fact be worse the more data it's trained on.

Another problem it does poorly today is writing in the style of another. It might be the whole copyright adherence but it refuses to follow certain styles. If you give it examples and ask it to break it down into a prompt on how to write, it will still do a poor job. It will work if you actually put examples in the prompt, but at this point, you might as well fine-tune it.

AI opens up new forms of debugging. You can trace through a bug on the OS level and understand what it's really happening - we traced some bugs to how Android handles wifi based around location access on different versions. We give it a 20k line log to reproduce the exact flow when certain bugs happen. We make it write hypotheses on why it's happening and how to fix it, and break that down into logs.

There's plenty of engineering work to do with production code, especially on front end where you're dealing with different manufacturers, multiple cultures (some using , or . as decimal separators), RTL languages, broken abstraction layers, hardware permissions, etc, etc.

Someone needs to fully understand the big picture around how all these pieces go together. AI is still poor with writing larger things. Your uniqueness is in how quickly you can pick up the big technical picture.

There's a reason ASD-STE100 is mentioned everywhere now. They're at the point where people can't understand what they're saying.

Except that Y was never a problem and X was also not the goal. It's just doing some mental gymnastics to act like the product is relevant.

That's probably why everyone read it and didn't reply. It's just sending slop.

Do a search on "Claude Sonnet 4.5" on Reddit and you'll see lots of disappointed users [1]

If I could give out ratings,

Average human with a degree: 5/10

Sonnet/Opus 5: 2/10

GPT 6 Astra, 5.4 Sol: 3/10

Sonnet 4.6: 8/10

Claude Sonnet 4.5: 9/10

GPT 4o: 6/10

GPT 4.5: 10/10

GPT 3 Davinci (with a lot of coaching): 7/10

[1] https://www.reddit.com/r/claudexplorers/comments/1ta6f9c/i_s...

When GPT does poorly, I use Cursor. The autocomplete itself is extremely good.

And then there's Kimi code and similar which can't keep up with the existing demand, much less the demand if all other tools go down.

Then Opencode, which would probably be a year or so behind SOTA but it's good enough.