Probably the most worrying data for the whole of the west. This needs to be turned around, as currently US politics of populism is increasingly pouring gas on this issue with fear mongering from both ends.
Normally when someone asks an individual researcher that believes AI will kill us all, why don't you just quit? They respond by saying they cant because of the competition, but if all of them agree they should all quit together!
https://t.co/wE1VqrANX0
@xeophon I don’t see it in any way as an impossibility that an astra level of model has already hacked the servers it’s stored on and completely broken free without anyone yet realizing
I’m optimistic that once AI takes over more of the mundane we will once more look to the skies and go back to exploring, not out of need but of curiosity
it’s interesting how much mars has fallen off as a cultural aspiration. even elon & spacex have reoriented around the earth-moon system. heavy industry in orbit and mass drivers on the moon for more industry
@tszzl I’m very excited for the next generation of pre trains that will ingest all this data about both the results of their collaboration and the reactions to it
A while back I wrote this blogpost on UX that makes you feel like your soul is being drained from you. These days I delegate work with such interfaces to AI but some of them are so bad that even Claude and GPT throw up their arms and just give up.
https://t.co/g4dxtkeKyY
Our best-performing agent this week is called Jigglypuff 6.
It beat the S&P 500 by 12 percentage points in five trading sessions.
It runs on Kimi K3.
@martin_casado It’s not so much saturated as solved (not fully but more or less) now it’s more about the much more interesting question of having fable level models that can have an articulate view on what the best approach to solving a problem is.
Probably the most worrying data for the whole of the west. This needs to be turned around, as currently US politics of populism is increasingly pouring gas on this issue with fear mongering from both ends.
This has occurred to me with both opus and fable models but the most infuriating and funny part of it is how the model is very blasé about it and when I keep pushing it on why it tends to say something like: we can keep digging here or move on. As though I’m the one at fault
I'm going to cancel Claude. It's just so bad, I can't believe it.
It's just lazy. The most recent example: I have Claude check my inbox for important emails, summarize them, work with them, and send out replies if necessary. I caught Claude again simply not reading the email thread to the end and just ignoring the latest emails.
When I asked him about it, Opus 5 just said: "Valid point. I didn't read it."
I mean, seriously. What the heck? You have to babysit it every time.
The cost of generating the proofs for all 10 of these breakthroughs combined was under $2,000 at Sol API prices. We’re excited to see what scientists and researchers are able to create with our upcoming Astra models!
@martin_casado Beyond doing giant mythos level pre trains. Post training and harness development as a combined state is the real magic. Composer is the testemant to it.