I think I'm warming up to using local LLMs for helping me code. I don't trust them for large scale code generation but it's started saving me some time in looking up syntax and docs via DuckDuckGo. I still need to look up docs for most of the work I do but not so much for common use cases you'd find on stack overflow.

@villares Neat. For sandboxes I like to use JupyterLite in browsers to skip having to install anything natively.

github.com/jupyterlite/jupyter

The second last layer was set to relu which messed everything up.

Show thread

Lmao, re-running an neural network training thing I did a month ago and it's no longer able to train a working model. Not even sure what changed 🤷

nazi techbro avoidance 

Over the past month, I've been doing my best to de-mega-corp my tech ecosystem.

My conclusion: unless you are a skilled tech nerd with money and are willing and able to do it, it's basically impossible to not be exploited by your tech.

Tech's current state is abusive in all the worst ways, and frankly it should be illegal.

Tech professionals should be ashamed by what we have wrought.

tips for folks getting started.

Don't use the system-level python as that can update and break dependencies. E.g. Arch uses 3.13 if you update and Tensorflow can't go above 3.12.

Use pyenv to do version management per project. github.com/pyenv/pyenv

Don't install pip dependencies globally. Use virtualenv to set up dependencies per project. virtualenv.pypa.io/

Python versioning is rough out of the box and these tips can save you some pain.

@delicta @hhardy01 Big mood. I love only knowing about what folks are up to by meeting up for dinner or a walk. I find DMs can also be useful for async catch ups too. It's not as personal as meatspace but it's closer than not getting to talk at all.

@lucid00.com Another group you might like is this VrChat Biosensing discord which uses EEG signals to control avatars. I think even though the corpos overhyped there's still some hope in us actually having access to interoperable spatial computing tools on the future

@lucid00.com There's still pockets of folks working on standards outside of the NFT/Corpo co-opting of the idea. omigroup.org/

Lots of neat stuff happening in the Godot and Blender space for example.

The appeal of machine learning to me is how clean the actual structure of models can be once you split it up in the right layers.

A lot of "application" code is a tangled web of dependencies and small independent or coupled bits of state that you need to coordinate in bizarre ways to account for the difficulty of state management in distributed systems.

ML on the other hand needs maths knowledge but the blocks fit together in a nice clean linear way of input to output.

youtube.com/watch?v=0VLAoVGf_7

Friendly reminder for those looking for a new , Agregore has built in UBlock origin ad blocking and doesn't track you or sell your data.

I'm seeing a lot of folks making the same suggestion, and now I'll say it too:

Use Amazon like a catalog you browse.

Find a thing you want.

Go to the manufacturer's site (NOT their Amazon "store!").

Buy it there. Often you'll find coupons or discount codes NOT present on Amazon.

#Boycott

In manual memory management, the garbage collects YOU.

Looks like the emails used to register for fediforum have either been leaked or scraped or something because I've been getting spam on the one I created just for the event. 😿

Hard to find time to become transhuman while I gotta do my tasks. Luckily humans on average have a decent lifespan to get stuff done in.

@tinker Kinda like how mRNA vaccines can be used to get the body to attack arbitrary cells by provoking the immune system

Show older
Mauvestodon

Escape ship from centralized social media run by Mauve.