thomaswilburn

thomaswilburn, 5 days ago to random

I think it's hard to describe just how deranged this whole conversation is: https://www.theverge.com/2024/6/3/24168733/zoom-ceo-ai-clones-digital-twins-videoconferencing-decoder-interview

It's disturbing that someone with power over other people's livelihoods thinks most of their job could be done by an LLM. "I could be replaced by a bot for the purposes of decision-making and nobody would notice" is not the ringing endorsement they think it is.

thomaswilburn, 15 days ago to random

The reason we used to call it web surfing was because you'd click a link from Yahoo! or whatever and end up going down a rabbit hole for a couple of hours. I think that's kind of the salt in the wound of Google's brainworm AI search results--we've lost so much of the indie publishing side of the web to social networks, but also lost the idea that the journey of discovery might be worthwhile or fun in and of itself.

thomaswilburn, 29 days ago to random

Prime example of where web components really shine: I'm working on a scrollytelling piece where moving through the document updates state linked to a map background layer. At the bottom there's a graph that appears if you click a school board district. Now I want that to appear in multiple places. Easy: just move that into a custom element, and it takes care of initializing itself wherever I put it. The ability to create reactive islands is deeply underappreciated.

thomaswilburn, 1 month ago to random

"AI is the only way I've been able to get stuff done on deadline, using tools and languages that I don't know very well."

My dude, what I'm hearing is that management is not setting realistic schedules for data journalism or providing adequate time for professional development. You don't need an LLM, you need a healthier newsroom culture and a union.

thomaswilburn, 2 months ago to random

The anti-phishing security training I had to complete for work sent me a congratulatory certificate as a link in an e-mail, and I'm choosing to believe that if I click that link they'll revoke my results.

thomaswilburn, 2 months ago to random

Re-reading again Daniel Davies' classic "one-minute MBA" post (thanks to a great MeFi collection by @brainwane) and I'm struck by how it also makes a case against LLM content (dishonestly sold, constantly fibbing, impervious to audit) from all the way back in 2004. Same snake oil, different day. https://blog.danieldavies.com/2004/05/d-squared-digest-one-minute-mba.html?m=1

thomaswilburn, 3 months ago to random

How my morning's going:

Reporter asks if I can help scrape some data out of a national political organization's site. It's in tables, but they've been rendered out to PNG.

Set up a quick download script, crop images, feed the table portion through Tesseract, write some Node to read those files and parse into CSV.

Discover that when the original tables were rendered, they truncated all the numbers to an arbitrary decimal place (for each individual cell).

Sigh, write back to the reporter...

thomaswilburn, 3 months ago to random

A) Don't ask us which security and privacy concerns a "home screen shortcut" adds.

B) Don't ask how Safari does this without any existing integration or architecture, it's clearly just magic.

C) DEFINITELY don't ask about Webkit's security record vs. the alternative browser engines.

thomaswilburn, 3 months ago to random

Reading the Tailwind discourse and my opinions will not be a surprise but it has solidified my opinion that (in web dev at least, and possibly on other platforms) the degree to which "requiring a bunch of build infrastructure to use a tool" is a warning sign.

thomaswilburn, 3 months ago to random

I'm upset to find that there's a Python 3.11 and it's not for workgroups.

thomaswilburn, 3 months ago to random

Here's the thing I see a lot: relatively knowledgeable developers saying things like "Copilot often generates weird code that you should handle with care" and junior devs saying "Copilot solved all of my problems!" or "I used an LLM to process or generate data." https://joshcollinsworth.com/blog/copilot

I think at some point it's our responsibility, as role models, to not use tools that we can (hopefully) use safely, but which in the hands of less experienced people are just automated footguns.

thomaswilburn, 4 months ago to random

American McGee's Bartleby the Scrivener

thomaswilburn, 4 months ago to random

I am really fascinated to soon learn just how petty Apple will be about geolocating who counts as an "EU" user for the purposes of running a new browser engine.

thomaswilburn, 4 months ago to random

Welp.

thomaswilburn, 5 months ago to random

The only end-of-year review that matters, here's the Chalkbeat data crime roundup for 2023: https://dataviz.chalkbeat.org/2023/12/19/data-crimes-2023.html

thomaswilburn, 5 months ago to random

I miss when "AI images" meant "dogs with eyeballs everywhere, like 50% dog and 75% eyeball, I'd still pet it though."

thomaswilburn, 6 months ago to random

Taking bets on which Substack Bro writes a contrarian, heartfelt paen to Kissinger tonight.

thomaswilburn, 6 months ago to random

Genuine question for people who work with React a lot more than I do: are there actual cases where you use Contexts to create state that is different in one part of the component tree than another, or where a sub-Context overrides its ancestor?

I ask because all the examples make it basically seem like a way to effectively set a global. Is this just used as a way to make values A) accessible more widely while still B) triggering a re-render?

thomaswilburn, 6 months ago to random

In general I like to rely on what the browser gives me but I've almost never run into a case where TreeWalker was as good as--or even sufficient compared to--manually recursing through a DOM fragment.

thomaswilburn, 6 months ago to random

Don't let the focus on shadow DOM as style/subtree isolation scare you off--it's also a neat tool for building expressive mini-languages in HTML that anyone can use.

https://www.milezero.org/index.php/tech/web/components/chiaroscuro.html

(Maybe the first of a series of posts exploring the parts of the web component API that I think are currently under-explored.)

thomaswilburn, 6 months ago to random

Naomi Alderman's The Future introduced me this morning to the MENACE, which is probably going to be my go-to metaphor for LLMs from now on. https://en.m.wikipedia.org/wiki/Matchbox_Educable_Noughts_and_Crosses_Engine

thomaswilburn, 7 months ago to random

Bullying works: https://android-developers.googleblog.com/2023/11/increasing-trust-for-embedded-media.html

(Very funny that they're announcing the end of "web integrity" as an aside in a post on the Android dev blog, though.)

kissane, 7 months ago to random

If you want to destroy your brain with frothy SV technocapitalist rhetoric as well as the news this week, this would be a fantastic starting point. I’ve been staring at it for awhile and I’m just going to have to come back to it later.

https://a16z.com/the-techno-optimist-manifesto/