@simon BrowserStack , basically it is OS and/or browser on a cloud mainly used for testing. Not sure how much you can install it though... and quite pricey.

@simon Get a Windows 11 key off eBay, then click that checkbox…

@simon I use https://shadow.tech/en-GB for nerdery and gaming

@simon 7zip is a compression software
https://letmegooglethat.com/?q=7z

@simon in theory, winget should be able to handle a .zip install and get the path right (I think it installs to user profile by default). It doesn't handle .7z, but it knows how to install 7zip and in principle converting the file format should be straightforward... I don't know if you need to write a local manifest though, or whether it can figure it all out itself.

@simon not only Python

I use other tools, and $PATH is equally annoying there too

@simon I know @wsvincent has spent a lot of time battling this for his Django for Beginners book. (“First learn some esoterica…”, it feels like.)

@simon respectfully disagree: $PATH is such a fundamental part to how computers execute programs, we should be educating here, not abstracting

@simon Nitpick: Whisper is speech-to-text. Text-to-speech is speech synthesis.

@simon Do you know how this compares to whisper.cpp (which also uses the GPU on Macs)?

@simon I wanted to know how the speed compares to `whisper.cpp`, since the openAI whisper is very slow on my mac, so I ran a test: https://notes.billmill.org/link_blog/2024/08/mlx-whisper.html

mlx ran almost 3x faster than whisper.cpp with a model of the same size, and both were using the GPU. I would love to know why it's so much faster!

@simon I find LLMs useful for assisting with a few things, but like you’re saying, any suggestion that they can do this on their own is utterly misguided

@simon I believe you invented the term “slop” right 😁 my question is, if “slop” is “unwanted AI-generated content” why is this called “slop” when it’s obviously getting a ton of engagement? Is the claim that all the likes are bots? If it’s actually hundreds of thousands of people liking AI-generated images then isn’t it … not unwanted?

@simon "...it is unlikely that GPT-4o is capable of catastrophic scheming."

So either it's bad at it or really good at it.

@simon Do they actually believe that stuff?

@simon
With all these price drops I'm waiting until they start paying me to use their models. 😜

@simon thank you this is hilarious 😂

@simon This conversation is giving me strong "there was an old lady who swallowed a fly" vibes

@simon that was hilarious, thanks!

As part of working on this I figured out (with more help from Claude) a good pattern for writing automated tests for a reusable Django app like this that can live in the same repository and spin up a minimal Django project, just enough for the tests to run. I wrote that up in detail as this TIL: https://til.simonwillison.net/django/pytest-django

@simon does this level of LLM “authorship” give you concerns about its provenance?

@simon do you use any IDE integrations to work with LLMs?

@simon I guess I don't understand how this is an attack. The malicious prompt came from the attacker, but so did everything else. So the attacker already has access to the "exfiltrated" data, right?

Or is there some missing context here?

@simon So, Walz knows Where, and Harris knows Venn.

@simon @kjhealy Do we know his opinions on the Mercator projection?

@simon

https://minnesotareformer.com/2024/08/06/former-geography-teacher-tim-walz-is-really-into-maps/

> Walz also says maps help implement the nitty-gritty details of otherwise abstract policy. “You have to have a plan” for “how we’re doing power and economic justice and environmental justice,” he said. “The tools for that plan are GIS.” Those tools “transfer a vision of a fair society into one that actually has results.”
#HarrisWalz2024

@simon I did quite a bit of testing tonight I was surprised that structured outputs also worked with GPT-4o-mini. I didn’t crunch the numbers but I was already using it to parse text to json and I noticed an immediate improvement. I assumed we had to use the new model for pydantic support but it just worked.