Lightnews — Scholar-powered news

Reilly Wood

Reilly Wood

@reillywood.com

200 followers 310 following 350 posts

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

Posts Media Videos Starter Packs

Reposted by Reilly Wood

Why @why.bsky.team · 9d

Yeah basically, software is boring once its in a container not called “my terminal”

Reilly Wood @reillywood.com · 9d

Polyglot Conference was quite good today. Nice to meet people from a wide variety of backgrounds doing cool stuff in Vancouver www.polyglotsoftware.com

Polyglot Vancouver Conference

www.polyglotsoftware.com

Reilly Wood @reillywood.com · 9d

LinkedIn, man. I do not think it is "poor form" to try to recruit software developers from your company www.linkedin.com/posts/dennis...

Reilly Wood @reillywood.com · 25d

Yeah, the borrow checker and strings are probably the biggest sources of complexity in the name of performance.

It gets easier (I promise!) and it's still my favourite language but I sometimes fantasize about a Rust with garbage collection that was a little more chill about implicit allocations...

Reilly Wood @reillywood.com · 26d

Not that I’m aware of (but my info may be out of date, I haven’t contributed to Nushell in the last year or so)

Reilly Wood @reillywood.com · 26d

I think Nushell users would generally not enable that; parsing JSON is trivial in Nu (pipe something to `from json`) but doesn't happen automatically when CLI tools emit JSON.

Maybe it should, will give that some thought :)

Reilly Wood @reillywood.com · 26d

Yes I think that’s a great approach.

Reilly Wood @reillywood.com · 26d

I’m glad you’re asking! Personally I think the best way is to have an easily discoverable way to emit a Nushell-supported format (usually JSON or YAML, maybe CSV for flat tabular data) without any autodiscovery.

(I used to be on the Nushell core team)

Reilly Wood @reillywood.com · 26d

uv is the cheat code for getting past the worst of the package management stuff. but I don't use Python enough to know the good ways to do type checking well, and then I try to use a popular, well-documented library and its examples are doing weird type checking with string comparisons

Reilly Wood @reillywood.com · 26d

every single time I do Python stuff

Reilly Wood @reillywood.com · 27d

That’s fair; I think it’s a fine option to wait it out while early adopters figure out the best ways to use these things

Reilly Wood @reillywood.com · 27d

I have found them to be crazy useful for many tasks but it took and still is taking a lot of experimentation to figure out what they’re good at and where they fall over

Reilly Wood @reillywood.com · 27d

oh no

Reilly Wood @reillywood.com · 27d

I did this at the age of 37; went from zero to 3-4h of Crossfit every week. And yeah, the results were very noticeable after just 6 months or so reillywood.com/blog/strengt...

Strength Training

I started strength training for the first time in January 2024, in my late 30s. I wish I’d started earlier, but it’s been a hugely positive change in my life. Some observations: Standing up straight i...

Reilly Wood @reillywood.com · 27d

Figma hired Krazam to promote their MCP server, lol (it's pretty good) www.instagram.com/reel/DO86iR4...

Login • Instagram

Welcome back to Instagram. Sign in to check out what your friends, family & interests have been capturing & sharing around the world.

www.instagram.com

Reilly Wood @reillywood.com · 27d

That's interesting and surprising to me; I imagined that the tools would end up looking the same to the underlying LLM either way. Maybe something's amiss in popular MCP client libraries?

Reilly Wood @reillywood.com · 28d

I met Remy at an MCP event in SF recently; super sharp guy, you’re building some neat stuff!

Reposted by Reilly Wood

Gergely Orosz @gergely.pragmaticengineer.com · 28d

Ironic: Gartner had my post ridiculing Gartner removed. The one that shared Gartner's Magic Quadrant about AI Code Assistants, which does not include Claude Code, OpenAI Codex, and ranks Amazon above Cursor.

This one: github.blog/ai-and-ml/gi...

Are they embarrassed to share their Magic Quadrant?

Reilly Wood @reillywood.com · 28d

can't find it but Dave Plummer from Dave's Garage has talked about doing something similar to this in the 90's for Windows stress testing

(he's a weird dude who frequently exaggerates but this story seemed legit)

Reilly Wood @reillywood.com · 28d

as a Canadian who was recently subjected to this: can you not?

Reposted by Reilly Wood

Hilleloween @hillelwayne.com · 28d

I was doing some software history research and stumbled on this absolutely FASCINATING letter from 1964: dl.acm.org/doi/10.1145/...

Some random defense contractor writes in to say "You should deliver a minimal prototype as fast as possible to get feedback and involve users at every stage of labor"

Some observations concerning large programming efforts | Proceedings of the April 21-23, 1964, spring joint computer conference

Reilly Wood @reillywood.com · 28d

This is a good read and I appreciate that it's honest about the reasons for hiding CoT! Thanks for posting it here

Reilly Wood @reillywood.com · 29d

galaxy brain (positive valence)

Reposted by Reilly Wood

Grace @gracekind.net · 29d

Worse news: there are uncountably many undecidable problems. Computers have been a scam this whole time

Prof Mike Yearworth @mikeyearworth.bsky.social · Sep 21

"In a landmark study, OpenAI researchers reveal that large language models will always produce plausible but false outputs, even with perfect data, due to fundamental statistical and computational limits."

www.computerworld.com/article/4059...

OpenAI admits AI hallucinations are mathematically inevitable, not just engineering flaws

In a landmark study, OpenAI researchers reveal that large language models will always produce plausible but false outputs, even with perfect data, due to fundamental statistical and computational limi...

www.computerworld.com

Reposted by Reilly Wood

hailey @hailey.at · Sep 21