aka de-anonymising over 16s

Nope, working fine here. - presuming you mean claude.ai

update - it's it's sporadically not working in the app

Maybe if you'd actually read the agreement

> In connection with using or accessing our Services you agree to comply with this User Agreement, our policies, our terms, and all applicable laws, rules, and regulations, and you will not...

> use any robot, spider, scraper, data mining tools, data gathering and extraction tools, or other automated means (including, without limitation buy-for-me agents, LLM-driven bots, or any end-to-end flow that *attempts to place orders without human review*) to access our Services for any purpose, except with the prior express permission of eBay;

You do if you don't want to get banned.

using them. there really is no other way.

It's like asking "how do I use my text editor".

Almost everything you read online will be out of date, and the person working on it won't work the same way you do.

Some people love it, some people hate it.

And unless you've got your own experience, it's hard applying other people's experience to your experience.

Especially for something like Claude Code where you're just prompting and getting results back.

I mean, half the time I use VS Code, half the time I use a terminal window.

You're going to get a lot of conflicting advice because everyone's environment is different and they work on different sorts of code bases.

But I'll give you mine.

I'm a Claude Max Pro 200 subscriber.

I sit with the Opus chat-bot have a discussion and come up with a spec. We turn this into a zip full of documents and upload it to a GitHub Repo.

e.g. https://github.com/lawless-m/BattleForMoscow

I'll then get Claude Code for the Web to go to that repo unzip the zip and read the documents. It will make a first pass at the entire codebase.

I'll merge that into main and create another Claude Code for the Web Opus session with any ideas I've had in the meantime - which will usually be a few.

Then I clone it to a local machine and get Claude Code Opus to try and get it to work. And I'll prompt it from there until it works. If it's a Linux program, that'll be in a terminal window. If it's Windows, I'll use VS Code because it's a better terminal in VS Code than it is in a terminal window on Windows.

That's a general workflow. Sometimes I won't use GitHub at all. Sometimes a PXE boot an entire Linux machine and give it that with admin privs.

And sometimes I just tell it to use sudo as my own account. On my router for instance, if we want to do things with the firewall.

Why do you go through all the trouble of uploading to Claude code for web only to download it back and run it in Claude code terminal? Are there different rate limits for each endpoint so that’s why you do it? Why not just work entirely using Claude code terminal locally?

I do most of my initial work in Claude Chat, on my phone. I have my best ideas when I'm away from my desk. Originally, Claude for the Web was only on the iPhone not Android - but it is now.

My iPhone 11 won't let you download certain files, and file handling on the iPhone is awful.

I guess I've just got used to the flow. I don't always do it like that, it was just one example.

> imagine “100 million people smarter than any Nobel Prize winner,”

But imagine if were true

Seriously, I wonder if it would be possible in the future to incorporate generative AI in everyday thinking.

Something like a watch or a ring that would help people make better decisions not on code but on everyday decisions.

Well, honestly, that's already what I do every day. Claude on my phone is rather handy, especially with voice questions.

I get a lot of recipes from it, I've built some tools that I give a list of ingredients I have already and it'll suggest recipes.

And it's just helped me fix my washing machine.

These are things I couldn't do before easily by just asking a search engine.

I get it to make product recommendations.

But that does carry some risk, of course, on the questions where pushing me in one direction or another is favourable to the people answering the question - Where should I get my mortgage? Which products are best that kind of thing.

Right now it is quite neutral, it scrapes existing reviews and gives pros and cons to various decisions.

And maybe Struggle Sessions.

I run it with sudo enabled - true story

just give it its own machine and let it check out any code

I PXE boot it from a known image when I feel the need

Running it remotely on a VM seems like a very sensible option. Just don't give it permission to nuke the remote repository hah (EG don't allow force-push, use protected branches, only allow write access to branches it created)

Same solution here - keep a base diskless image on the server, copy it to the diskless area, pxeboot the machine. Works for Windows too (iscsi).

Could do the same thing on EC2 of course.

I remember when Google's ads were separate and clearly labelled.

This is expected behaviour, not a bug.

DuckDB (like most database systems and many applications using memory allocators like jemalloc or mimalloc) doesn't immediately release memory back to the OS after freeing it internally.

Memory allocator strategy - DuckDB uses an allocator that keeps freed memory in a pool for reuse. Returning memory to the OS is expensive (system calls, page table updates), so allocators hold onto it anticipating future allocations.

Also the docs say that dropping tables should free their memory:

"Running DROP TABLE should free the memory used by the table"

Source: https://duckdb.org/docs/stable/sql/statements/drop

Thanks for explaining this! I suspected there was some additional context and was digging into it now. Problem is that the memory never seems to be freed, and I could update my issue to show this.

Maintainers have acknowledged problems like this on other issues too.

But I'll nevertheless read more into this!

tbh i might dig a bit deeper as it reports zero memory use itself.