Pinned Tweet
Cheers!
My Stellar MD Editor is now available on Google Workspace.
If you, like me, have a hive of MD files stored on Google Drive - you'd probably love it ;)
it's also free
Google Workspace Store: workspace.google.com/marketp…
Direct link: md.vedaispace.com
mvl retweeted
AI dadacenters just don’t use that much water, there’s a panic over almost nothing.
Here is a short animation to put AI water use in context:
Another issue worth flagging in regard to AI outputs watermarking:
It's raising the bar for new AI companies, especially those EU-based, to an almost unreachable level. What costs big AI labs a fraction of their resources to implement may simply be too much for smaller ones.
mvl retweeted
What effect does having kids have on the male-female wage gap?
It's hard to tell from cross-sectional data, but at the very least, the gap is still largely between married men and the rest, and those who don't have kids seem to have somewhat lower mid-career wage trajectories.
mvl retweeted
Adding a short prompt telling the model to stay within the scope of the task significantly improved performance on scope criteria, although it didn't result in high/xhigh fully outperforming medium effort.
See our prompting guidance:
platform.claude.com/docs/en/…
mvl retweeted
To illustrate, in the task below, both low and xhigh performed the same changes. Opus 5 at low effort made exactly a single character change, while xhigh refactored the surrounding code.
Opus 5 feels like it has undiagnosed ADHD.
-- working on a task
-- Claudeing
-- nowhere near finishing
-- "hey, noticed you changed something in a file I wasn't working on — want me to switch to that? also, unrelated, but here's 5 things I'd like to flag: [...]"
Caught myself editing an article for a website I launched this year, with an editor I built this month, while listening to the radio through a player I wired up this week.
Welcome to the new normal, I guess.
On SpaceX IPO,
Q1 results are worth noting:
$4.28B net loss on $4.69B revenue,
add +$3.75B/q from Q3 (+$2.5B in Q2, Anthropic)
and +$2.76B/q from Google starting Q4,
it likely will be profitable before CAPEX by year-end,
and has its x2.4 Rev YoY for Q1 2027 already signed.
I may be wrong, but Opus 4.8 seems practically unusable for coding atm: it assumes variables, ignores instructions and offers random suggestions instead.
If the intent was to force users into using other models for their respective tasks - a clear success. Otherwise - not sure.
mvl retweeted
It has never been more obvious that socialists don't actually care about working people, they just hate the rich.
