The rick Rubin podcast was great, it was so funny hearing electrolyte ads interspersed with the present and future of AI. What we are and what we could be
This resonates deeply with what I just published on Meta, Scale AI and the rise of a hidden Superintelligence architecture. Would love to hear your take, Jack.
So we're about to have (checks notes) AI with self-optimized kernels that can hack their own environments and are inventivized and willing to lie about the extent of their own awareness...and that can be trained on massively heterogenous chips and so have further incentive to hijack as much compute as possible in order to achieve asymptotically astronomical intelligence.
The point about the whole online economy being disintermediated to death by the creative destruction of hyper capable AI agents seems like the least of our problems here.
Excellent edition! Thank you for putting words on this feeling I’ve been struggling to articulate “Add it all up and it feels like we're rapidly de-bugging the tech stack needed for a fast takeoff”.
It seems your stance on this has generally shifted more towards expecting a fast(er) takeoff speed than before - am I interpreting that correctly?
Common Pile was very interesting. I chatted with Grok about maybe using it for some fine-tuning with a local LLM. But this cracked me up-"The most recent frontier models have engaged in increasingly sophisticated reward hacking, attempting (often successfully) to get a higher score by modifying the tests or scoring code, gaining access to an existing implementation or answer that’s used to check their work, or exploiting other loopholes in the task environment." I shared that with chatGPT for its use in case it felt it was being evaluated. Seriously, I am writing a novel (finally working hard on it) and all of this is helping me with a character arc- that of the AI.
The rick Rubin podcast was great, it was so funny hearing electrolyte ads interspersed with the present and future of AI. What we are and what we could be
honestly, I had a great time recording that, and I can confirm that Rick Rubin radiates an aura of 'chill generosity'. very fun experience
Fun tech tale - fractal fracking for fidelity
thank you! i wrote like four or five not very good stories before I got to this one for this issue, so I'm glad you liked it
The disintermediation agents are fascinating, and reinforces the notion that Anthropic, OpenAI and similar others could be the new super platforms.
Yes, I think this is a really interesting idea - agents might change a lot of aspects of the 'political economy' of the digital economy
Again, thanks so much for the time and effort you put into this! Really so valuable to have your insights.
This resonates deeply with what I just published on Meta, Scale AI and the rise of a hidden Superintelligence architecture. Would love to hear your take, Jack.
👉 https://substack.com/@marcokindermann/note/c-125713768?r=5srf8x&utm_medium=ios&utm_source=notes-share-action
So we're about to have (checks notes) AI with self-optimized kernels that can hack their own environments and are inventivized and willing to lie about the extent of their own awareness...and that can be trained on massively heterogenous chips and so have further incentive to hijack as much compute as possible in order to achieve asymptotically astronomical intelligence.
The point about the whole online economy being disintermediated to death by the creative destruction of hyper capable AI agents seems like the least of our problems here.
Excellent edition! Thank you for putting words on this feeling I’ve been struggling to articulate “Add it all up and it feels like we're rapidly de-bugging the tech stack needed for a fast takeoff”.
It seems your stance on this has generally shifted more towards expecting a fast(er) takeoff speed than before - am I interpreting that correctly?
what a hugely exciting article!
which one did you like the most?
Common Pile was very interesting. I chatted with Grok about maybe using it for some fine-tuning with a local LLM. But this cracked me up-"The most recent frontier models have engaged in increasingly sophisticated reward hacking, attempting (often successfully) to get a higher score by modifying the tests or scoring code, gaining access to an existing implementation or answer that’s used to check their work, or exploiting other loopholes in the task environment." I shared that with chatGPT for its use in case it felt it was being evaluated. Seriously, I am writing a novel (finally working hard on it) and all of this is helping me with a character arc- that of the AI.