brumar

6 months ago I experimented what people now call Ralph Wiggum loops with claude code.

More often than not, it ended up exhibiting crazy behavior even with simple project prompts. Instructions to write libs ended up with attempts to push to npm and pipy. Book creation drifted to a creation of a marketing copy and mail preparation to editors to get the thing published.

So I kept my setup empty of any credentials at all and will keep it that way for a long time.

Writing this, I am wondering if what I describe as crazy, some (or most?) openclaw operators would describe it as normal or expected.

Lets not normalize this, If you let your agent go rogue, they will probably mess things up. It was an interesting experiment for sure. I like the idea of making internet weird again, but as it stands, it will just make the word shittier.

Don't let your dog run errand and use a good leash.

Great list, thank you!

My favorite book.

Best read I had in months. That,or maybe cognitive dissonance because I spent 1h of my life on it (there is a Dilbert joke just on that, mind you).

Thank you Scott A.

Sampling seemed so promising, but do we know if some MCPs managed to leverage this feature successfully?

Interesting. Skills on MCP makes a lot of sense in some contexts.

Lazy from me to not check if I remember well or not, but the dev that got productivity gains was a regular user of cursor.

Yes. Rather a lack of control of the subject and intensity of focus. Which piss everyone off because they can't steer it from the outside, hence the "lack of focus" perspective.

Yes. I did not look but most probably the non interactive mode flag is used (-p)

To find "How many moves in a game does it take to reach this position". Which was a wrong interpretation.

This is very different to my experience and I am wondering why. Maybe because I come from chess and can't help myself to compare it with this frame of reference. Anyway I felt that my progress up to 5k was largely driven by a better understanding of principles of plays than tactical training. As a thought experiment, I feel that its possible to adopt a very risk averse style that negates tactical complexities to the expense of many points on the board and still largely win against weaker players. It's not my experience with chess. If you suck at tactics, your elo sucks too.

Same. I was astonished it was remotely possible to do this.

Are dependencies easier to install or does it work only for packages that have pure wheel support?

Correct me if I am wrong, but Docling can do both. It has also, among other strategies, a non-AI pipeline to determine the layout (based on qpdf I believe). So these projects are not that different.

Of course I had to generate the legendary "shit on a stick". But for chefs. https://anycrap.shop/product/shit-on-a-stick-for-chefs

These days, I spend time training people using this kind of tools. I am glad it's called as such. It's much comfortable to explain to a tech person that it's "badly named" and that it should have been named "Code Interpreter" instead than explaining to a non tech that the "Code Interpreter" feature is a new cool way to generate documents. Most people are not that comfortable with technology, so avoiding big words is a nice to have.

1 day later: it's a total mess.

Very good. I was wondering why nobody did something like that before. At least this was my conclusion from my google searches few months ago. From a design perspective, I don't like storing state in the DOM itself, I would have find much more flexible to have the state in a json as a single source of truth and use reactive patterns such as state change => ui change and not hiting the DOM directly. That sounds like big framework things, but it saved me from acute headaches in a personal similar framework I did for self-contained apps. It was not HTML but I thought I would apply the same logic if I had to do it for html apps too.

Thanks. I'll add the answer to my own question too: pdfLib is the dependency at work to make it work in the browser runtime.

I like the idea but that seems complex to put in place and would risk degrading the perfs.

You can test this prompt yourself elsewhere, you will notice that you get sensibly the same experience.

I got this one which seems to confirm yours : https://gist.github.com/brumar/5888324c296a8730c55e8ee24cca9...

20 years ago I was addicted to a game like that named Globalcombat.com. Good times. Few years ago it was still alive. Now it seems gone.

Correction: extract monstreous profits. When I read about the revenues associated with Reddit AI deals, I can't even imagine what could possibly be deals that cover half of the internet. Cynically speaking, it's a genious level move.

I am doing pdf-as-software so that I can get a free ticket for hell or a nearby mental asylum.

So LLMs have their alpha go zero moment where training on human data is has-been? Sounds exciting? Terrifying?

Very important comment. My workflow changed dramatically with the increased capabilities of these tools.

Sampling is to my eyes a very promising aspect of the protocol. Maybe its implementation is lagging behind because it's too far from the previous mental model of tool use. I am also fine if the burden is on the client side if it enables a good DX on server side. In practice, there would be much more servers than clients.

> that plaintiff shall not file any other documents with the cartoon dragon or other inappropriate content

Formal answers to goofiness (voluntary or not) will always amuse me.

Hello fellow context owner. I like my modules with their context.sh at their root level. If crafted with care, magic happens. Reciprocally, when AI derails, it's most often due to bad context management and fixed by improving it.

True. I learned this the hard way.