Pacing the Frontier, P(doom) and the Cyber Divide, Jensen and Satya Push Back, Anthropic’s Threat Intel Report, the RubyGems Swarm Attack & the Glasswing Receipts
The number that's missing from OpenAI's own Defense Factory writeup is the one that would let you compare it to Glasswing and Patch the Planet. They give you day one closure counts, a 37% duplicate rate, and a 0.81% false positive rate after dynamic validation, all upstream of a fix actually shipping. Glasswing's own five month accounting puts verified fixed at 0.8% of findings. Patch the Planet is at 13% merged and only 7.9% of those rated high. Three different organizations, three different discovery pipelines, and none of them report the one number that tells you whether the fix pipeline can keep up with the finding pipeline. Until Defense Factory publishes something comparable, the 53 urgent issues closed on day one is a validation stat wearing a remediation costume.
The split you're describing is the one I spent this week's issue on. Everyone argued about what the model wanted, and the evidence that landed was a poisoned upload, stolen keys, sideways movement, and a channel nobody watched. Take the word AI out and it's a 2015 breach. Pacing changes what the next model wants. Today's agents keep every permission they had yesterday, and that's where the operational risk you're pointing at lives.
This video in particular covers the people behind this "doomer conspiracy" really well. Becker wrote a book on these people and Newport seems well up on the history as well.
The amusing thing to me is that I used to be part of the Extropians newsletter. I was around when all this stuff that Becker talks about was first being discussed. I was not aware, however, that it apparently spun off into a more or less "religious cult" subsequently.
I'm also suspicious of where the millions, even hundreds of millions, of dollar are coming from, in particular because of the association with Peter Thiel.
This needs to be looked into by impartial investigators as part of a government investigation of these frontier labs and the people behind them.
Stopping The AI Safety Cult ft. Adam Becker & Cal Newport | Better Offline
Note: I flagged that company - Irregular - in my Substack Notes soon after the initial reports of the hacks. Being an Israeli firm pricked my interest. Now I see I was right.
The number that's missing from OpenAI's own Defense Factory writeup is the one that would let you compare it to Glasswing and Patch the Planet. They give you day one closure counts, a 37% duplicate rate, and a 0.81% false positive rate after dynamic validation, all upstream of a fix actually shipping. Glasswing's own five month accounting puts verified fixed at 0.8% of findings. Patch the Planet is at 13% merged and only 7.9% of those rated high. Three different organizations, three different discovery pipelines, and none of them report the one number that tells you whether the fix pipeline can keep up with the finding pipeline. Until Defense Factory publishes something comparable, the 53 urgent issues closed on day one is a validation stat wearing a remediation costume.
The split you're describing is the one I spent this week's issue on. Everyone argued about what the model wanted, and the evidence that landed was a poisoned upload, stolen keys, sideways movement, and a channel nobody watched. Take the word AI out and it's a 2015 breach. Pacing changes what the next model wants. Today's agents keep every permission they had yesterday, and that's where the operational risk you're pointing at lives.
This video in particular covers the people behind this "doomer conspiracy" really well. Becker wrote a book on these people and Newport seems well up on the history as well.
The amusing thing to me is that I used to be part of the Extropians newsletter. I was around when all this stuff that Becker talks about was first being discussed. I was not aware, however, that it apparently spun off into a more or less "religious cult" subsequently.
I'm also suspicious of where the millions, even hundreds of millions, of dollar are coming from, in particular because of the association with Peter Thiel.
This needs to be looked into by impartial investigators as part of a government investigation of these frontier labs and the people behind them.
Stopping The AI Safety Cult ft. Adam Becker & Cal Newport | Better Offline
https://www.youtube.com/watch?v=0oVSnaINJ30
Marcus Hutchens weighs in:
https://www.youtube.com/watch?v=KMf2RDC5Dyg
Gary Marcus (see my comments to his pieces on his Substack):
https://garymarcus.substack.com/p/sam-altman-says-trust-me-jensen-huang
David Sacks warns AI doom narratives backed by serious funding, Anthropic IPO could expand it
cryptobriefing.com/sack…
Kevin Bass' thread on X:
x.com/kevinnbass/status…
How Effective Altruism Bought the Media
effort.news/tarbell
AI Safety Donors Paid Religious NGOs $3.3M for Statements on AI
effort.news/revelation
A Single Firm is Behind OpenAI, Anthropic, and Meta Hacking Scandals
effort.news/irregular
Note: I flagged that company - Irregular - in my Substack Notes soon after the initial reports of the hacks. Being an Israeli firm pricked my interest. Now I see I was right.
Ed Zitron again...
AI Is Already In Dangerous Hands
https://www.wheresyoured.at/ai-is-already-in-dangerous-hands/