these are the two fundamental constraints: space & time.
to truly support scalable recombination, you'd need to account for every incorrect model as well, storing all those false paths. which would require more capacity than the system itself.
And that makes it hard exercise.
true. simple primitives take care of complexity, but the hard part is finding them, because
1. The search space is vast. Without memory of failed primitives, you keep walking into the same walls.
2. The search gets slower over time. Reason alone can’t finish.
so luck, may be?
The most complex phenomena arise from scalable recombination of very simple rules. Whether it's galaxies, chips, or neural networks, if you find the right primitive building blocks, the complexity takes care of itself.
The most complex phenomena arise from scalable recombination of very simple rules. Whether it's galaxies, chips, or neural networks, if you find the right primitive building blocks, the complexity takes care of itself.
In a stunning convergence of science and nature, our satellite-tagged turtles are helping decode an ancient oceanic map ! As the Southwest Monsoon sets in, several Olive Ridleys have moved towards the southern Bay of Bengal and the waters around Sri Lanka, to seasonal hotspots where warm Bay waters meet cooler Indian Ocean currents. This oceanic mixing fuels a burst of marine productivity, creating rich feeding grounds that turtles have navigated to for millions of years. Long before satellites, they knew exactly where to go. Today, Tamil Nadu Forest Department’s pioneering satellite-tagging project is helping us understand how these remarkable navigators read the ocean and connect distant marine ecosystems Credits- TN Turtle Telemetry Project in partnership with Dr @sureshwii , @wii_india@Aiwcrteofficial & @tnforestdept #TNForest
AI in 2040 will not be built on the stack we are using today. It will be much closer to optimal. The current stack has 3-4 orders of magnitude of data inefficiency and 4-5 orders of magnitude of compute inefficiency.
Near-optimal AI is what symbolic learning will deliver.
I have had a different experience, Vaibhav. If the aim is to change someone's behaviour, kindness is a must. It is anyway hard to change anyone's mind on anything - without genuine kindness, it is near-impossible. "Brutal" feedback is often a power play, designed to steamroll, not convince.
@drfeifei The deeper question is whether reality is fundamentally mathematical, or whether mathematics is merely one representation among many. world models implicitly assume the former.
@drfeifei The most interesting signal in the article is the industry's return to representation itself.
world models feel new, but the question is ancient: how should reality be represented?
capability advances do not eliminate this problem. we are only rediscovering it again.
@intacel I am a technocrat because of the strength of our ancient civilization.
Technology and business are both downstream of culture and culture is shaped by civilizational values. Without those values, all we have are shiny gadgets and a soulless existence.