For those in the reliability community, www.revelara.ai is highly, highly worth checking out. I ran it on a codebase and it immediately found a ton of embarrassing things.
"Like Mythos, but for reliability"
Average of 100 +\- 5 is often way way way better than 100 +\- 50
one of the biggest companies in the AIOps sector right now is just doing automation of runbooks printed on Confluence or Quip and translating them into fuzzy automation and selling it as "AI SRE".
I understand their minimum pricetag is a 7 figure yearly deal. They have many clients.
Prince in unusually reflective mood, I always thought.
I strongly suspect that Zvi doesn’t need more inbound, but his latest model card assessment has a couple of notable quotes about Mythos Preview escaping containment ("if you caught a human doing [this] even once [...] you would obviously have to fire them.")
thezvi.substack.com/p/opus-47-pa...
As an author, I strongly dislike O’ Reilly’s Early Release model, since my stuff gets poked at before it’s ready. As a reader, I strongly like O’ Reilly’s Early Release model, since I can poke at other people’s stuff!
O’ Reilly’s platform is hosting the latest - www.oreilly.com/library/view...
Reflecting suddenly that I can't think of a modern writer whose vision of the future has been so thoroughly, utterly vindicated as @greatdismal.bsky.social
What? It was right there!
These bots are made for walking
(elsecord)
i keep returning to this sorta spooky question of how much various tech cos are effectively already being run by models. i presume it would be hard for outsiders to know. but you gotta guess that "this is the least AI-run that large 'orgs' will ever be"
samuel beckett was born on this day 120 years ago. astonishing to think that if he had survived, and could run 100m in 9.57 seconds, he would be not only the oldest but also the fastest man in the world - together with his nobel prize for literature, an astonishing trifecta
Part of the musical background of my teens, and a talent seemingly wholly original, but actually deeply rooted in its locality.
My cherished and long-held belief that actually we’re just not very good at knowing what good looks like (or let other considerations override, etc) particularly in leadership remains undefeated.
H/T @yvonnezlam.bsky.social
I guess if your goal is to try and make something fake appearing human, picking Zuckerberg for an early effort has some advantages.
We’re all trying to find the guy who did this
and quantization turns abstention into votes for the guy who's gonna win anyway?
"The pope. How many divisions has he?" - apocryphally attributed to Stalin
Project Glass _wing_, right? Not project glassing? Which would imply Anthropic got drunk and did some unwise conversions of things into weapons… no, wait, checks out.
That’s astonishingly better than
Everything is a magic quadrant!
shipping it to production ten times a day
Can I just make the argument that it's different shit, _same_ day
1/ The Russian IT sector faces being crippled by new, harsh penalties for using VPNs. The Russian public also faces an imminent ban on the use of foreign AI systems, which developers say will wreck Russia's development of its own AIs. ⬇️
The revolution devours its children?
Cc @yvonnezlam.bsky.social
Not now, fungus that has evolved to feed on radioactivity - we’re busy www.sciencealert.com/chernobyl-fu...
if they follow the trend established by 1st -> 2nd systems, then they all get more theoretically capable/powerful and also simultaneously more unfinished until in the limit, you have have a system that could do everything but in fact can do nothing at all
