Phenomenal

Furthermore, by not obscuring the truth from your team members, you invite them into the work of fixing the problems. You have to RTO bc you are struggling with training juniors, productivity or whatever? Why not *tell folks*, and unleash all that creativity towards solving it?

That's the theory. However the practice of 300% increase in productivity could mean fire 50% of the staff, if you want to keep your job then you work the same for less, company takes the 50% increase in productivity plus execs / shareholders get the extra profits ->

We fall into a habit of describing systems as complex when the cause of their unpredictability could be the perspective we are using. If you replay the same chess game, it should give you the same outcome ... unless you don't record co-ordinates but simply type of piece moved.

New: a project analyzing human language usage by scraping the web is shutting down because "generative AI has polluted the data." It's going to become much harder to analyze human use of language with the rise of AI-generated stuff being everywhere 404media.co/project-analyz…

Had quite a similar map going with some SREs Red Hat recently

The scientific method isn't only for scientists. It can help us all learn faster. Study of 759 startups: teaching scientific thinking boosts revenue. Founders become quicker to abandon bad ideas. A key to good judgment is treating plans as hypotheses and choices as experiments.

A few days ago I posted about "Hope for cynics", the new book by the neuroscientist Jamil Zaki. Here's another article he's written: news.stanford.edu/stories/2024/0…. People don't typically have "cynical personalities". Rather, the social environment significantly shapes our willingness to

really sharp people state the obvious. Only it wasn't obvious until they said it At a time and place where everyone can nod and run with it Which they helped create.

I don't ordinarily try to publish in journals anymore, but since I have a coauthor this is an exception, based on the trust work from last year. royalsocietypublishing.org/doi/10.1098/rs…

new paper from our work at Meta! **GPT-style language models memorize 3.6 bits per param** we compute capacity by measuring total bits memorized, using some theory from Shannon (1953) shockingly, the memorization-datasize curves look like this: ___________ / / (🧵)

Brian Wilson of Beach Boys passed away yesterday, sadly. Please remember you wouldn’t be “vibecoding” without him.

Context engineering

This is sort of mind-blowing. Claude estimated 8 hours to redesign and reimplement a major component that must run in the browser. The original component required polyfills for a faux Node.js environment. Honestly, I'm fairly certain it would have required 32-40 hours to do

I'll return to my broader point: evidence suggests AI, at its current level of capability, will have a big impact on jobs, education, and our society. We only get to shape those outcomes if we realize that they are going to happen. Giving people an excuse to dismiss AI hurts them

And by "prototyping in 1-2 days" it's often "build 10-15 variations of this thing and show it to people over 1-2 days, then choose the one that feels the best" THIS kind of stuff used to be extremely rare, and it's becoming more common Not sure if limited to very few places btw

Intelligence is like electricity. If you give companies 10X more, most of them won’t know what to do with it.