A screenshot of the Claude status dashboard, displaying a distressing number of outages in the past several weeks.
This is why I still have a job
A screenshot of the Claude status dashboard, displaying a distressing number of outages in the past several weeks.
This is why I still have a job
One thing Iβm thinking about: when using Claude for software, a significant factor in success is βautomated verificationβ to steer the agentic loop (tests, type checking, linting, etc). Do other forms of knowledge work have similar systems that can be applied as rapidly/iteratively?
roadhouseendgame i was talking to my mom about cardassians and said something along the lines of what must it be like to be a cardassian civilian. like abstractly knowing the rest of the galaxy thinks of your species as 'the evil ones' and kinda sorta knowing your government runs labor camps and torture prisons and is a military empire, and it's a huge deal for everyone else in the galaxy, but you just like run a restaurant so it's not a big part of your daily life, so you just like, go about your day and make soup and don't think about it' and then halfway through my sentence i actually heard myself and then had to go sit down for like ten minutes. like sat down in a chair and stared at the floor for ten minutes
whoops
Itβs truly awe inspiring how completely broken large cross-platform group chats are on iOS.
I did an impromptu lightning talk last night at Austin Systems, it was really fun! Itβs a great meetup, @adamchalmers.com has done an amazing job of building a community.
Attached are a representative sample of my slides.
This is also what human babies are like.
To be clear: fuck Newsom, I wouldnβt vote for him.
(But like, I probably would, though. Or would I? Hmmmm, probably not. But maybe? Do you really want to take that chance?)
Three obviously true things:
- the two party system is bad.
- threatening not to vote for a candidate you donβt like is an entirely reasonable form of political discourse and persuasion.
- it is highly unlikely even the worst dem nominee would be remotely comparable to Vance. Fascism sucks!
If she runs, I would strongly consider quitting my job to volunteer full time for the campaign.
To elaborate further here:
In from the cold is the thesis statement. If you donβt like it, go find a different author.
You need to see the circus be effective before the Karla trilogy, and definitely before the looking glass war.
You need some imperfect spies before you see the perfect one.
Favorites:
- a perfect spy
- the looking glass war
- smileyβs people
Suggested order:
- the spy who came in from the cold
- call for the dead
- tinker tailor soldier spy
Oddly, this is a case where the order of βhow good I think they areβ and βthe order I think you should read them inβ could not be more different.
Iβve read a whole stack of Le CarrΓ© novels, and 200 pages in, A Perfect Spy is _by far_ my favorite
Vegas really is the America of America
Itβs a good bridge! And a good trail!
Yummmm!
Say more about the marinade, what recipe are you following?
ohhh βdemocracy dies in darknessβ was aspirational
JUST IN: Democrats flip a red seat in the Texas Senate.
Taylor Rehmet wins against Leigh Wambsganss, a conservative activist who helped lead right-wing efforts to take over TX school boards.
With all early vote, and 95% of precincts, in, Dem up 57/43.
Trump won this seat by 17% in Tarrant County.
Steve is getting to the good stuff now.
Immigration is good. It is good for immigrants, and good for the country. We should have more of it, and our government should do everything in its power to encourage it.
The current legal immigration system is an absurd, Byzantine, kafkaesque nightmare. Abolish ICE is the moderate position.
So⦠general strike?
or, to quote another reference that has been bouncing around in my head more and more over the past days, weeks and months:
βthere are four lightsβ
This is a _really_ excellent overview of the deranged, visionary(?) madness of Gas Town.
Pretty amused by Star Trek: Flirty Teens so far.
- Holly Hunter canβt sit in a chair normal!
- Unexpected basketball sequence!
- βKittenβ!
Yeah, that resonates for sure.
Iβve always been on team tiny files, but I think the argument against large files is much stronger when working with LLMs.
I donβt think this take is even particularly warm, especially once you dig into the attributes that make LLMs more effective.
This is essentially:
- well tested, easy to verify
- clear boundaries and interfaces
- well organized and well named files/modules
Give Claude a way to verify its work Include tests, screenshots, or expected outputs so Claude can check itself. This is the single highest-leverage thing you can do. Claude performs dramatically better when it can verify its own work, like run tests, compare screenshots, and validate outputs. Without clear success criteria, it might produce something that looks right but actually doesnβt work. You become the only feedback loop, and every mistake requires your attention.
the new best practices for Claude Code doc is really good. they chose their #1 rec very well. every CC user should be typing "Give Claude a way to verify its work" over and over on a typewriter in an empty hotel while their wife thinks they're working on a book
code.claude.com/docs/en/best...
Money Stuff on BAGS:
> It is, from first principles, stupid.
@matt-levine.bsky.social is a national treasure.
www.bloomberg.com/opinion/news...
Pro tip: listen to a bunch of Yo La Tengo records