afro88

Nice setup, but GP said:

> how people who really try to learn with these tools work

This setup is potentially effective sure, but you're not learning in the sense that GP meant.

For GP: Personally I've reached the conclusion that it's better for my career to use agents effectively and operate at this new level of abstraction, with final code review by me and then my team as normal.

Sometimes people are actually holding it wrong though

Honestly, comments are just half the problem. At least half the articles I read from HN are vibe written. And I only spot it after reading a few paragraphs. It's leaving a bad taste, and it's sad because HN was guaranteed to have plenty of things worth reading and it's deteriorating

What models and inference provider?

I would have agreed with you a year ago

This is what I'm interested in knowing. Just how much can it modify if it's a static binary? Can it modify it's own agents.md etc?

Aside from security and efficiency, is there anything openclaw and do that moltis can't? Like for example, does moltis have the "heartbeat" thing, short and long term memory, can update a soul.md etc?

I'm so keen to try openclaw in a locked down environment but the onboarding docs are a mess and I can see references to the old name in markdowns and stuff like that. Seems like a lot of work just to get up and running.

Man, you have to read the article, not just the headline

Right, prompting for depth of field will never work (with current models) because it treats it as a style rather than knowing on some level how light and lenses behave. The model needs to know this, and then we can prompt it with the lens and zoom and it will naturally do the rest. Like how you prompt newer video models without saying "make the ball roll down the hill"

Apologies, I did jump the gun here. There has been more and more lazy LLM replies on HN lately and yours raised a flag in my mind because I can't remember someone commenting that deeply while also agreeing with me (normally if it's a lengthy response it's because they are arguing against my point).

There are some enlightening points here about LLM writing style for me. Trying to write like an LLM being impossible (at least for a non-trivial length of text) is such a good point. Run on sentences as another hint that it's not an LLM is also useful. Thanks!

Which LLM am I talking to here, GPT-5.2?

The blur isn't correct though. Like the amount of blur is wrong for the distance, zoom amount etc. So the depth of field is really wrong even if it conforms to "subject crisp, background blurred"

Once you start arguing, it's time to start a new prompt with new instructions

But gcc is part of it's training data so of course it spit out an autocomplete of a working compiler

/s

This is actually a nice case study in why agentic LLMs do kind of think. It's by no means the same code or compiler. It had to figure out lots and lots of problems along the way to get to the point of tests passing.

I think you're being very generous. There's almost 0 chance they had this actually working consistently enough for general use in 2024. Security is also a reason, but there's no security to worry about if it doesn't really work yet anyway

Same. Immediately I thought why not have clawdbot ask you for the 2FA? That way you at least kind of know what security-protected action it's trying to take and can approve it

Would love to know this too. When he talks about letting clawdbot catch promises and appointments in his texts, how many of those get missed? How many get created incorrectly? Absolutely not none. But maybe the numbers work compared to how bad he was at it manually?

Vibe coded

Considering how many people say ChatGTP too

Oh, you're right. I misunderstood. I think it was the comparison to mermaid that threw me, and then I checked out the examples

Edit: See child comments, I misunderstood. Original post below anyway.

--

To me, the diagrams have an uglier default design. A crazy amount of dropdown shadow (in the examples), stretched fonts (front page example). It doesn't matter how many diagram types you support if the generated diagrams aren't good quality.

Mermaids defaults are good enough. They're not amazing, but I'm also not embarrassed to show them to other engineers, stakeholders etc

Edit: KeenWrite looks excellent btw

I thought I had a great startup idea. It was niche, but a solid global market. It was unique. There was a genuine pain point being solved. My MVP solved it. The pricing worked, the tiers were sound.

At least ChatGPT, Gemini and Claude told me it was. I did so many rounds of each one evaluating the other, trying to poke holes etc. Reviewing the idea and the "research", the reasoning. Plugging the gaps.

Then I started talking to real people about their problems in this space to see if this was one of them. Nope, not really. It kinda was, but not often enough to pay for a dedicated service, and not enough of a pain to move on from free workarounds.

Beware of AI reviewing AI. Always talk to real people to validate.

I'm not great with math beyond high school level. But I am very interested in, among many things, analog synthesiser emulations. The "zero delay filter" was a big innovation in the mid 2000s that led to a big jump in emulation accuracy.

I tried to understand how they work and hit a brick wall. Recently I had a chat with an LLM and it clicked. I understand how the approximation algorithm works that enables solving for the next sample without the feedback paradox of needing to know it's value to complete the calculation.

Just one example of many.

It's similar to sitting down with a human and being able to ask questions that they patiently answer so you can understand the information in the context of what you already know.

This is huge for students if educational institutions can get past the cheating edge of the double edged sword.

> at launch

Does this confirm they actually do knee cap models after the launch period to save money, without telling users?

You misunderstand. A coder will write different code for the same problem each time unless they have the solution 100% memorised. And even then a huge number of factors can influence them not being able to remember 100% of the memorised code, or opt for different variations.

People are inherently nondeterministic.

The code they (and AI) writes, once written, executes deterministically.

Yes I believe so. Also things like forcing a "key insight" summary after the excels vs struggles section.

I would take any descriptions like "comprehensive", "sophisticated" etc with a massive grain of salt. But the nuts and bolts of how it was done should be accurate.

Humans writing code are also non deterministic. When you vibe code you're basically a product owner / manager. Vibe coding isn't a higher level programming language, it's an abstraction over a software engineer / engineering team.

This is quite sad, in that the fact that it's AI generated is hidden in the disclaimer and nearly all of the comments think it's a real human.

> The bottom line: Junior developer hiring could collapse as AI automates entry-level tasks

If AI automated entry-level tasks from today, that just means "entry-level" means something different now. It doesn't mean entry-level ceases to exist. Entey-level as we know it, but not entry-level in general.

LLMs are generators, and that was the correct way to view them at the start. Agents explore.