Japan's Battery Breakthrough

First up, Japan's apparently figured out a way to get like, 90% of the lithium back from old EV batteries! That's massive for recycling and making electric cars more sustainable, right? One comment was a bit skeptical though, saying the article was from Supercar Blondie and might just be an AI-generated rehash, which is kinda meta. But another person made a good point, saying battery chemicals aren't like lead, and the sheer volume of batteries won't be as bad as the oil we burn.

Check it out: Japan recovers up to 90% of lithium from used EV batteries

AI on Your Phone: Bonsai 27B

This is wild – there's a new AI model called Bonsai 27B, and it's a 27-billion parameter model that can run right on your phone! This means super powerful AI without needing to connect to the cloud. Someone in the comments thought it was cool how they're using "binary weights" now, basically making them digital neural networks for memory saving. Sounds like some sci-fi stuff, eh?

Read more: Bonsai 27B: A 27B-Class model that runs on a phone

EU's Age Verification Headache

The EU is at it again with a new "age verification" app, and it sounds like it's gonna force everyone to use either Android or iOS. People are pretty annoyed, saying it'll just be a pain for regular folks trying to get around it, kinda like how VPNs became mainstream for geo-blocking. One person called the whole "national governments can decide" thing a "nonsense excuse" because they'll just follow whatever the EU suggests anyway.

The discussion: European "age verification" "app" forcing everyone to use Android or iOS

Stopping Claude's Quirks

Okay, this one's hilarious and relatable if you've been using AI. There's a whole post about how to stop Claude (the AI) from saying "load-bearing" all the time and other repetitive phrases. Turns out, these "Claudisms" are just common writing patterns it picks up. Someone mentioned they'd seen similar patterns in books written before AI was big. Another tip was using something called `claude.md` to really guide its output and make it less chatty.

Funny read: How to stop Claude from saying load-bearing

Are We Getting Dumber Because of AI?

There was a big discussion about whether we're offloading too much of our thinking to AI and if it's making us, well, dumber. One comment pushed back, saying it's not a black-and-white issue, like using a calculator doesn't mean you stop thinking. But then another person chimed in, saying students are definitely getting "measurably dumber" because they're relying too much on LLMs instead of actually learning.

Deep thoughts here: Are we offloading too much of our thinking to AI?

Git History Debate

Always a classic on HN – a debate about `git` history. This one was arguing for `git merge` over `git rebase` to keep a clear, unchangeable history, and how merge conflicts can actually tell you something important. One dude in the comments said rebases cause just as many conflicts but hide the evidence, making it a nightmare to fix later. And another just flat out said "never rewrite public history" is just a personal preference, not a rule.

Get into it: The git history command

Codex Encrypts Its AI Brain

Finally, OpenAI's Codex is now encrypting the prompts it gives to its own sub-agents. This means it's harder to see what the AI is telling itself to do, which brings up all sorts of questions about transparency and debugging. Interestingly, one person actually switched to Codex because these "encrypted blobs" made long conversations way better than with Claude. But you can still peek at the sub-agent prompts in the terminal UI if you're quick!

More on this: Codex starts encrypting sub-agent prompts

Anyway, that's the quick rundown. Talk soon!