terhechte

you need the max plan to break free of most rate limits

Anthropic isn't slow. I'm running Claude Max and it's pretty fast. The problem is that Cursor slowed down their responses in order to optimize their costs. At least a ton of people are experiencing this.

The "open Terminal of selected folder" is the proper way, it requires a different workflow though, so I understand the issue in terms of muscle memory. An easy enough alternative (for me) is to drag and drop the current folder onto iTerm. When you have a folder open in Finder, and hover over the name of the Folder in the titlebar of the window, a blue "folder" icon appears. if you drag that to the iTerm icon in the dock and drop it, iTerm will activate and open a new terminal window in this folder.

The error library he seems looking for is „error_mancer“

I've tended to calling him Smaug. It's a fitting name in that the roots of Apple's success were software development (the "dwarves"), and for a long time developers had a great standing (internally and externally). But then Tim "Smaug" Cook took over the Iron Mountain of Apple (the hoards of Gold). And boy does Tim like money. And anyone who dares wanting any of his money are thieves. Hence software developers are thieves for wanting to maybe pay a bit less than 30%.

I find the multi monitor solution really good. I use it a lot. You can have an ultra wide screen display which is like having multiple separate monitors.

This is my biggest iPad gripe. I understand the security, but just make it a new "entitlement" that is only given to UTM, Parallels and VMWare fusion. Or make it a "developer mode" that you can only enable if you pay $99 a year. I'd be fine with that, but the whole iPad is unusable for any kind of software development and I'd love to be able to travel with just an iPad because for everything else I do while I'm traveling (watching videos, reading, browsing, writing, drawing) the iPad is great and I don't have to lug two devices around.

I had the same issue recently and used https://crates.io/crates/rusty-s3

I agree, but I think they're not in the position anymore to enforce this. Years ago this would have been a valid move, but companies that switched to cross platform will not move back. It saves costs and adds velocity to just have one team for both platforms.

But not these, afaik. RN only supports basic view types and composes them into controls. But these offer complete distinct controls. Using these in RN is possible, but usually via iOS-only plugins

The "it would have been different under Steve Jobs" trope is coming up every so often when Apple introduces a radical new feature. Oftentimes I don't agree but in this case I absolutely do. A huge amount of engineering to get these shader effects right, but the outcome will be that the platforms will look more like a fruit salad and less coherent. Companies that invested in React Native, Flutter, Tauri, or alternatives in the past years will not necessarily be willing to redo their whole app just so that the controls get the glass look. At the same time replicating this in every cross platform toolkit might be difficult, and so I think this will lead to many apps looking out of place which worsens the overall experience.

No they're not. I prefer closed source LLMs because I'm not interested in having a small hardware park, but DeepSeek 0528 or Qwen 3 235 are really good models. Not as good as Claude 4, but absolutely good enough for a lot of development use cases. You just have to spend money on the right hardware.

I sat down at a Piano yesterday and tried to play a beautiful song even though I never really used a piano before. Sounded horrible, must be the pianos fault.

One reason to use LLMs is to understand when and how to use them properly. They're like an instrument you have to learn. Once you do that, they're like a bicycle for your mind.

You could split it up in two separate entities. For vector search there's a myriad of good Rust projects. I've personally used:

- https://crates.io/crates/lancedb - https://crates.io/crates/usearch - https://crates.io/crates/simsimd

search and simsimd are fast and lightweight, but I'd advise to use lancedb if you're a bit new to Rust as the other two are a bit trickier to handle due to the C dependency (e.g. usearch needs Vec::with_capacity and will otherwise crash, etc).

And then, you take the result of this query and can combine it with a sqlite `in` query.

Or you use SQLite with a vector search extension: https://crates.io/crates/rig-sqlite

You can run the 4bit quantized version of it on a M3 Ultra 512GB. That's quite expensive though. Another alternative is a fast CPU with 500GB of DDR5 RAM. That of course, is also not cheap and slower than the M3 Ultra. Or, you buy multiple Nvidia cards to reach ~500GB of VRam. That is probably the most expensive option but also the fastest

Why did you bundle Redis and not use one of the many key value libraries available for Rust (or even sqlite)?

Would this also be possible with other LLM engines / GPUs? E.g. Llama / Apple Silicon or Radeon?

But for any coding task.

Just at note that

> If the user doesn’t know the answer, the LLM can’t help. That’s not just a minor limitation—it’s foundational.

doesn't mean the LLM is not useful. My favorite use case for LLMs is them doing something I know 100% how to solve myself. However, they do it much faster and I can 100% understand their solution and give feedback in case it is wrong. No surprises. I much prefer this as it allows me to work on multiple tasks in parallel.

Can you give some examples where it didn't work for you? I'm curious because I derive a lot of value from it and my guess is that we're trying very different things with it.

I'm also a huge fan. Started watching him after a Japan trip. I like the format so much, I'd love to have something similar (day in the life) for various other countries.

It says `Coming Soon` for the `inference.py` for the quantized version. Does anyone happen to know how to modify the non-quantized version [0] to work?

[0] https://github.com/Lightricks/LTX-Video/blob/main/configs/lt...

No currently not. It would be easy to add though. I haven't updated the tool in a while (after using it to clean up my Gmail inbox). Thanks for pointing out the certificate!

I build something to visualize huge amounts of email (such as from Gmail) some years ago:

https://github.com/terhechte/postsack

There's aider (https://aider.chat) or Claude Code (https://docs.anthropic.com/en/docs/claude-code/overview) or Codex (https://github.com/openai/codex) or plandex (https://github.com/plandex-ai/plandex) or kwaak (https://github.com/bosun-ai/kwaak)

I'd venture to say there's more of these than there are UI Editors tbh.

In the new agent panel not everything is editable. Maybe give it another try.

I'd also like to throw my own app into the mix

http://hyperdeck.io

This is a very accessible way of playing around with the topic: https://transformerlab.ai

Local models, due to their size more than big cloud models, favor popular languages rather than more niche ones. They work fantastic for JavaScript, Python, Bash but much worse at less popular things like Clojure, Nim or Haskell. Powershell is probably on the less popular side compared to Js or Bash.

If this is your main use case you can always try to fine tune a model. I maintain a small llm bench of different programming languages and the performance difference between say Python and Rust on some smaller models is up to 70%

Image input has been working with LM Studio for quite some time