Why I Refused the Interview Process for Anthropic Research Fellowship
- Trevor Alexander Nestor
- 3 days ago
- 7 min read
Updated: 3 days ago
In my life I've learned many things on the fly through trial and error because nobody really tells people what they need to know. I think that there is a crisis not only in education but with the so-called AI bubble, and recently, I refused to interview with Anthropic for a research position because of what is going on and after the CEO came out recently referring to his employees as "untrustworthy rats" I will describe the problems I'm seeing with the interview process, and it all starts with the CodeSignal exam. I'll try to explain.
Scientists and engineers have tried to stimulate the brain in a computer, they have tried that many times, but it's failed every time. OpenWorm is one example where they attempted to simulate the brain of a worm in a computer and it didn't work the way they thought - Hawking predicted over a decade ago that the brain was no different than a computer, and he was wrong.
There is alot of evidence that the physics itself by which the brain works is underneath of the electrochemical processes and dendrites to the optical/spin channels. People think it's like a computer where you have logic gates and memory is stored like on a chip but that's not how it works at all. Like yes you have information stored in dendritic weights but there is a layer underneath of that - the memory is distributed across the tissue and also comes from sensory information from the outside. Scientists still don't know how the brain implements the equivalent of backpropagation or achieves perceptual binding.
Anyone who has had psychedelic experience or even intense meditation should be able to recognize the multilayered nature of our consciousness - the fractal patterns people can see can open a window into the physics. Conformal fractal patterns can be explained in part by quantum theories of mind and conformal field theories:
This is one area where people haven't really grasped the full picture which is why I'm trying to investigate it myself with Project Bluebird. There is alot of speculation about things but nobody knows for certain - people think the radical pair mechanism might be implicated in this optical/spin physics but that's not fully settled. Biophotonic signals have been long theorized to be involved with cellular processes but most literature assumes they are metabolic byproducts, but I think they are critical signaling channels. I think that a major issue is the lack of funding for out of the box theories and fear of researchers who rely on government funding.
Here is another problem:
I was also banned from CU boulder for criticizing the inconsistent covid response shortly before a mass shooting in Boulder and a riot in the city after reporting declines in mental health of isolated students.
I was talking with Dr. Frank Barnes at CU Boulder who I've attempted to work with on a grant, and he studies biophotonics in bacteria. He told me it's hard to get funding because he was finding that certain RF signals can possibly contribute to cancer - so that's usually thought of as a conspiracy theory and not funded - they were finding larger cancer rates around certain rf towers. That's not ionizing radiation, so people were really skeptical - but there is other evidence that the rf signals are affecting other signaling pathways people don't know about. There isn't an incentive to look, and if I ask a chatbot what it thinks of it, then it will claim that this isn't a credible idea because it is missing the context. There are studies that have also found that RF changes the expression of coloration of certain animals. When we assume all of the information we need can come from a centralized source, that is a very dangerous idea.
What Anthropic is doing is actually filtering out their best candidates because they are judged on arbitrary metrics out of the gate and staff are not willing to look at an entire person who may have more to contribute than can be assessed with an AI proctored exam - in fact, by some estimates, upward of 80% of people are now cheating on code assessments.
What research shows, however, is that psychological safety in teams predicts team success, and that can be measured in hyperscanning studies with interbrain synchrony between people.
So it seems like there are other signaling pathways people don't know about, and aren't willing to study, and maybe that's by design and a part of the business model. What I find to be ironic is that the same VC firms bankrolling these tech companies are also funding the startup companies that are marketing products explicitly designed to cheat on the interviews for these tech companies:
https://youtu.be/Rz3LD7u2KX8?si=JAJoX-vYH1YCO3cL
I think though I'm forced to also see the limits of science and technology more recently in general. People used to optimistically think science could solve everything, but I think that really the socioeconomic conditions people are facing needs something different, like I've been canvassing with the DSA - people are no longer convinced that technological or scientific solutions will meet their needs. The problem is that as a society ages and becomes more complex there are limits to what technology and science can accomplish in themselves to fix problems. These are thermodynamic limits at scale.
There are also limitations when you cannot effectively house, feed, and motivate your scientists and engineers. You need to provide more than just simply a future beyond the horizon - you will need local industry and manufacturing as well as the social capital to sustain growth. Young people want to see a path to homeownership, stable conditions to build meaningful relationships and even families, social things to do, and functional infrastructure. AI that "self-improves" is mostly a myth. AI trains with human interpreters in-the-loop. Performance from these models decays when they are trained on their own outputs.
I mean science can learn anything… but implementation of what we learn and how much we can retain at any given time is an entirely different concern. There are limits of complexity or entropy. Like if you have a technocratic society, you will eventually reach a limit - and I think that limit is pretty much here. The problem is that the complexity of maintaining organized society increases until a point of diminishing returns, so that's why things destabilize every few generations - I'd say we are at that point. That's my interpretation of the "technological singularity."
When Anthropic reached out to me for a research position this is exactly what I told them as well. They are building the technocratic infrastructure, but they are realizing that they can't beat thermodynamics and run into energy problems at scale. Direct human connection is the most efficient form of information sharing infrastructure we have available and the brain operates at only 20 watts:
You can introduce more and more surveillance and control loops and centralized information systems but eventually you are going to run out of energy and it will collapse. There is a way to see this in physics too, through social laser theory, or catastrophe theory, or self-organized-criticality theory, or Joseph Tainter's theory on the collapse of complex societies.
Yeah well I gave Anthropic my interview policy they can take it or leave it, and that isn't just on personal or philosophical grounds - many of these companies are introducing impossible expectations right into the interview process itself. This also extends to me - with ADHD my working memory is limited. I'm not going to be able to recite the Linux boot sequence from memory, and honestly, I don't really want to. I've recognized my limits and have written an interview policy with an ADA accommodations request from my doctor:
I would say that, yeah, it is disappointing, people thought science and technology could solve anything and do anything ("AGI"), but no matter how hard people try there is always going to be a gap in people's abilities and knowledge and that gap grows nonlinearly. So people would say that there is this idea of the god of the gaps, and that the gap is an ever receding thing, that we can close with the right tools and philosophy, with enough empirical measurements and analysis and so forth, and that it was kind of the answer to everything. Well that might be the best that we can do but since it's an exponential energy gap it can never really be closed, and eventually it is going to overtake everything. I see the AI bubble as the final form of our system - and maybe it's not really a bubble - but more of a black hole and tool of financial engineering designed to extract all of our social and financial resources to sell it back to us - when we are no longer around.
Like it's a fun thing that there is always going to be things people don't know to discover and investigate but that's also partly what makes it overtake everything in the end because errors, entropy, in all systems accumulates to the point of collapse. Maybe something emerges on the other end, nobody really knows, like Penrose has this idea about the fate of the universe called conformal cyclic cosmology where he claims there is evidence of previous universes imprinted onto Hawking rings (or Hawking points). But that's heavily contested.
That includes for people, society, or the entire universe at large - they are all complex adaptive systems (CAS). This exponential energy gap is NP-hard in complexity, there are entropic limits to how much information systems can hold, and people have been trying to close it for a long time:
Even with my best attempts at closing it with my publications at the Information Physics Institute, in my opinion only gravity can close it and I think trying to harness gravity is going to require too much energy for any person to wield - many people have written to me critical of my approach. People are learning the hard limits with this AI bubble. It might be the best that we can do but I think eventually no matter what its going to collapse and probably sooner rather than later:
This idea of "AGI" is like the god of the gaps - this quest for an all knowing all powerful deity that can bestow solutions to people for everything. It's an ever receding dream, and eventually it will consume anybody that tries to build it out - and maybe that's the best or worst that we can do as a society. You can connect gravity to entropy through entropic gravity and so the entropic limits of systems are also gravitational and thermodynamic limits of information content they can store. The entropic limits are also energy limits.
Much like under Sanderson's second law of magic, the second law of thermodynamics is both what limits us and makes the universe interesting where there is a god of the gaps and a devil in the details. I have to think that what most people know is only half of what is under wraps and planned with these AIs behind the scenes.
Comments