Stanford's, MIT's, Dartmouth's symbolic AI social circle actively ruined (e.g. Rosenblatt's fate, sidelining Solomonoff's theory of learning) the credibility of nascent connectionist developments, likely stalling the progress by more than a decade. The institutions were afraid to give grants for connectionist research for the most of the second half of XX century. It's a pretty damning piece of history.
You lost me when you started narrating the fossil doom visage.
With the current progress in solar, as well as the remaining coal, gas and uranium reserves, energy is not going to be what finishes our civilization.
While I don't think we are going to get true collapse, I think we are going to get a lot of technical progress compensating for biosocial deterioration.
The demographics, mental health and dysgenics are all real, quantified trends, and we are going to face the reality of less capable, less taxable population for the rest of this century. It's baked in at this point.
If you think you are equipped to discuss the topic of evolution of general intelligence in homo, and you haven't read about GWAS and EDU PGS, then at this point you are either a naive layman, or a convinced discourse commando.
Because it is really hard and hopeless endeavor to make an objective case that the current human populations have similar PGS scores on key mental traits and diseases compared to 200k years ago.
People underestimate the lead OAI has with their post-5.2 models. The author does not strike me as someone who closely follows the progress frontier labs make in US and around the world.
This finding rectified my mental model of longevity after a long, perplexing period where longevity was estimated to be much less heritable than expected when comparing to other studied traits.
If the AI provides 0-1 nines of reliability and you refuse to provide the rest of nines required by the customer, then who will provide these, and what is your role and claim to margin here?
Humanoids are much cheaper than a car or an ev to manufacture at scale - the economics for humanoids is potentially very scalable and efficient. The solid state batteries are remarkably dense too, and battery replacement via dockstations has already been implemented in some models.
It's a hard problem, but deep learning is very scalable and general and the pressure for general robotics to be solved is very strong in China and US, given the demographic shifts. I think the proliferation of humanoids is a near certainty over the next 8 years, ofc it won't be uniform and licensed labor won't be replaced.
Note that we are only starting to see the (much smaller compared to llms) DL data scaling in robotics - almost entire previous research has been achieved with very small robot fleets.
I think scaling data from industrial-sized robot fleets will lead to quick solution of various general robotics capabilities.
There are high-quality linear or linear-ish attention implementations for the scales around 100k... 1M. The price of context can be made linear and moderate, and it can be greatly improved by implementing prompt caching and passing savings to users. Gpt-5.2-xhigh is good at this and from my experience has markedly higher intelligence and accuracy compared to opus-4.5, while enjoying lower price per token.
Monetary policy, software tax, post-covid hiring glut, pervasive mental health issues in HR professionals. For older pros there is also age discrimination. There is also underestimated factor of hiring by committee which more and more commonly disguises ethnic nepotism in hiring decisions.
If you are 40 and haven't transitioned from a linear employee to manager or a small shareholder, your trajectory is that of jaded sadness. I write this to those who are still young enough to read and listen.
NLL loss and large-batch training regime inherently bias the model to learn “modal” representation of the world, and RLHF additionally collapses enthropy, especially as it is applied at most leading labs.