@EmoFoucault this means decentralization is a 'strong attractor'; in game theory its an 'evolutionary stable strategy'; in network theory it holds 'distributed robustness or network resilience', but these are all pointing to the same scale-free & substrate-agnostic evolutionary phenomenon.
Alasdair MacIntyre's critique of Great Books: "Proponents of this type of Great Books curriculum often defend it as a way of restoring to us and to our students what they speak of as *our* cultural tradition; but we are in fact inheritors... of rival and incompatible traditions."
Imagine waking up and realizing that your full time job entails using a digital cinema camera with a 65mm large format CMOS sensor fabricated on a sub-20nm semiconductor process, capturing 16-bit linear RAW image sequences at multiple gigabytes per second onto high speed NVMe media. Those images are then ingested into a distributed storage architecture, checksum-verified, transcoded into proxy codecs, synchronized via SMPTE timecode, color-managed through ACES pipelines, denoised with machine-learning models, reconstructed with optical flow algorithms, and composited using physically based rendering, ray tracing, volumetric simulation, deep compositing, photogrammetry, LiDAR scans, neural radiance fields, and GPU-accelerated particle systems and yet still.... deciding to believe that there's nothing artificial about any of it.
Believing that only now are we entering an era of computer generated imagery.
@JutlandGhost This can be pretty succinctly modeled by game theory, btw: market competition drives continual investment in optimal autonomy, thus *capitalism has functioned as an AI arms race all along* - the systematic selection of ever more autonomous productive intelligence.
The thesis that capitalism is artificial intelligence was, is, and will be Nick’s fundamental theoretical contribution. He is correct about this and it is in this precise sense that I am a Landian.
I think one of the conclusions we should draw from the tremendous success of LLMs is how much of human knowledge and society exists at very low levels of Kolmogorov complexity.
We are entering an era where the minimal representation of a human cultural artifact... (1/12)
@clairesilver "...and forms that appear with infinite strength, then 'melt into air'. I am an architect, a constructor of worlds, a sensualist who worships the flesh, the melody, a silhouette against the darkening sky. I cannot know your name. Nor can you know mine...
I will yet again flag that the best art and the main sick ai content (and possibly evidence of alien sentience) is basically entirely ignored by the culture outside of weird tech circles and instead of caring abt this we're cycling in prioritizing cynicism
Ethnomusicological Science Fiction (ESF) presents a cymatic reconstruction of pre-infodemic artifacts inspired by the visionary architectural designs of Lebbeus Woods.
Hey @sama -- I get that your Code Red is about rapidly beefing up ChatGPT to beat Gemini on key metrics and meeting user needs ... HOWEVER this still won't solve your main problem which is that building big LLMs effectively is no longer differentiating...
I think OpenAI still has potential to move faster than Big Tech institutions but you'll need some new ideas not just revved up activity and focus...
If you're interested to think a bit more deeply and laterally about how to achieve AGI -- and since you've said you'd like OpenAI to be more open -- you may want to think about integrating GPT6 with some of our new ideas in Hyperon. ..
https://t.co/GrY14xdtZo
If you shift from backprop to predictive/causal coding you may find you can make a transformer NN that learns continually, ending this mess of artificially distinct inference vs training models, and training in batches.
And rather than artificially glomming on extra tools via RAG and APIs, you can embed transformers in a theoretically grounded integrated cognitive architecture (Hyperon), along with logic theorem provers, evolutionary program learning, nonlinear-dynamical attention allocation etc.
There is lots of work ahead to make all these things work effectively at scale and lots of details to resolve -- but if you actually want to get to AGI (and then ASI), it's not just that bigger and bigger LLMs won't do it, no cognitive architecture with LLMs *at the center* will do it.. though absolutely LLMs can be very valuable components...
Oh and with these tools one can build AGI on decentralized networks too (we can use the core language-of-thought of the Hyerpon AGI design, MeTTa, as a smart contract language) -- so you don't necessarily require trillions of dollars of your own monolithic infrastructure... Of course more hardware helps but it can be distributed all over the place ;)
https://t.co/s4QJgo7366
Yes, all this is not yet matured and product-ready -- but the core infrastructure is scalable as of a couple months ago ... and when you first started OpenAI transformers were not matured and product-ready either right? ...
Just sayin' ...
Ofc I don't totally expect you to take me up on this offer to help you integrate our OSS tools in your stack ... but if you don't, then in a couple years after we've launched actually-open and decentralized AGI, I will point you to this old message and smile ;) ...
Fragmentary reconstructions of two Incantatory Cryptographic Microchants attributed to the Cult of the Superorganism. #EthnomusicologicalScienceFiction