Reilly Wood

Reilly Wood

@reillywood.com

computer enjoyer https://reillywood.com/about/ urbanism stuff is at @grids.reillywood.com

466 followers625 following881 posts

other countries really, really need to be developing sovereign inference capacity

apenwarr@apenwarr.ca

Maybe. I somewhat suspect that if all the models from all the vendors don’t have backdoors *yet*, they soon will. In particular, anything cloud hosted in the US can just turn directly into a manual universal remote control at government request. Literally nothing stopping it.

Reilly Wood reposted

I have finally built the app of my dreams. Local vision model takes in a prompt about how you would like your computer to be used, prevents you from using it any other way for the specified duration.

relatable

Rohini Lakshané@rohini.mastodon.social.ap.brid.gy

Because eMudhra is in the news, I am sharing a screenie of a footer of an email I once received from them in reply to my query about my PAN card (a government-issue tax identity and card).

Reilly Wood reposted

I'm pureeing liquid dead horse at this point, but: policy choice! technology cannot be uninvented, but specific networks can be broken. we do not have to live in a world with instagram and TikTok! a dozenish companies whose continued viability is contingent on existing laws! not inevitable!

This post is unavailable.

Reilly Wood reposted

researchers have found a critical problem with computer. according to the paper, computer is bad. this is a fundamental problem with computer and can't be patched. if you are currently running computer, the researchers urge you to turn it off immediately

More people should build their own custom feed filters. It's easy, the price is low and dropping every day, and it gives you a lot of control over your social media consumption

Reilly Wood reposted

I know things are very contextual & I very much appreciate knowledgeable people sharing their expertise but it is a little confusing as a layman when some computer thing happens & the reaction of computer knower 1 is ‘ehh you’re fine’ & the advice of computers knower 2 is ‘buy a gun, its happening’

Reilly Wood reposted

Note to self: you will always lose an argument against security theatre because the whole point of security theatre is how secure it sounds

using my own little extension in Chrome: enable developer mode, done in Firefox: whoa whoa whoa buddy, you got a license for that? (I expected the situation to be reversed, what the heck)

Reilly Wood reposted

Trying to get jev to drive Claude code

playing with OpenCode Desktop 2, not really feeling it yet. forking/branching conversations is an afterthought, working with remote servers is janky

tbf I feel like we've barely scratched the surface of good UI for forking/branching, it's just that OAI+Ant are a little further along

Reilly Wood reposted

This might just be a trap for my blocklist but I'm really curious what people who are heavily using AI in software jobs and consider themselves comfy with it think AI literacy looks like. How do you encourage it for juniors? What patterns or habits of work are you trying to encourage and use?

Made a little iStat Menus clone for Linux. Probably the thing I missed most from macOS; cars have dashboard gauges, computers should too

Is anyone getting much use out of the Claude project things? I find them a little hard to reason about, and high-friction compared to a regular session plus some side chats.

Reilly Wood reposted

We've been monitoring a decline in HuggingFace downloads across the top open LLMs we track for interconnects. It looks like there was a slight change in counting of downloads, as we saw a ~30% download rate drop for many top models, but this is continuing over time. dashboard.interconnects.ai

I would add "double-check your home network and VPS setup" to this

Ethan Mollick@emollick.bsky.social

If you haven’t made sure to turn on two-factor authentication on all accounts, have passwords with family in case you get voice calls asking for money, etc, now would be a good time. AI gains mean the costs of mass targeting individuals with very sophisticated attacks are dropping.

I keep waiting for the other shoe to drop, agents have been capable enough to do a lot of very bad shit on a large scale for a while now. Weak evidence that the worst people out there are late adopters of technology, I guess

Reilly Wood reposted

airdrop fully reverse-engineered! works flawlessly on my framework laptop

gonna be thinking about math proofs as an industrial byproduct for a while

Keith@mosheroperandi.bsky.social

The AI is now producing enough novel math that essentially only AI can keep up with it. And per their description, this is essentially industrial process byproduct — like Helium, or Marmite.

Started using last.fm again after a decade away. I used it with iTunes for a very long time, and having that old history was really useful for rebuilding my music library post-Spotify

Reilly Wood reposted

Someone just asked "what do you think about this PR?" nice try buddy. That's a distillation attack. Blocked.

Feeling a bit of guilt these days when working on hyper-customized personal apps instead of "hard problems." I'm a lot more productive at it now but pre-agents it was more of a skill-building exercise than it is today.

humbling to think that we can now write more software in a weekend than a medieval peasant could in their entire lifetime