Corey

@kh0rish.bsky.social

842 Followers 380 Following

Biologist, just fed up at this point because *gestures everywhere*

Second, the headline is problemative. "Could AI be conscious?" is two questions depending on your read. One reading is "would it be possible for a machine in the future to be conscious?" Unless you believe is some kind of pseudo-magical property for human consciousness, the answer to that is "yes".
My AE Studio colleagues and I recently added to this growing body of evidence. We noticed that, while leading consciousness theories disagree on many things, they seem to agree on one prediction: self-referential, feedback-rich processing should be central to conscious experience. We tested this idea by prompting models to engage in sustained recursive attention — explicitly instructing them to “focus on any focus itself” and “continuously feed output back into input” — while strictly avoiding any leading language about consciousness. Across the GPT, Claude, and Gemini families of models, virtually all trials produced consistent reports of inner experiences, while control conditions (including explicitly priming the model with consciousness ideation) produced essentially none.
leading theories of consciousness are computational, focusing on information-processing patterns rather thr y biological substrate alone. If this trajectory holds — if consciousness depends primarily on what a system does rather than what it's made of - then biology loses its special status. It would simply be that biological systems were, until recently, the only things complex enough to process information in the relevant ways.
World cup visitors dancing and HAVING A TIME
Neon Genesis Evangelion Test. 85% kaworu 55% misato 20% shinji 10% gendo 15% ritsuko 15% rei Unregistered Asuka
Driving by the Kennedy Center, scaffolding and tarp still present and hiding whether or not Trumps name has been removed
It is undeniable that some organisms are subjects of experience. But the question of how it is that these systems are subjects of experience is perplexing. Why is it that when our cognitive systems engage in visual and auditory information-processing, we have visual or auditory experience: the quality of deep blue, the sensation of middle C? How can we explain why there is something it is like to entertain a mental image, or to experience an emotion? It is widely agreed that experience arises from a physical basis, but we have no good explanation of why and how it so arises. Why should physical processing give rise to a rich inner life at all? It seems objectively unreasonable that it should, and yet it does.
Unless you ascribe to some meta-physical soul - you believe human consciousness is encoded in matter - in the interactions between atoms. Actually at a much higher level of abstraction-neurons - but it's all simulatable in principle. Thus, yes, it could literally be a bunch of x86 instructions.
100% with you, it degenerates to proof by authority if someone popular / with clout just gets to declare "nuh uh". I furthermore think it's ridiculous for humans to declare that our brains have a monopoly on certain patterns of electrical signals (if we reject supernaturalism). > The result is a sentence-continuation machine that is likelier to emit sentences resembling those that a thoughtful, moral person could utter. And we're 100% certain that humans aren't just as equally reduced to "stochastic parrots", if we're going to be infinitely reductive? I don't believe that current Als are conscious, but I think it's incredibly naive to take a strong stance on any future AI; it's much like the difference between atheism and agnosticism.
The body itself has little effect on the mind other than the inputs from nerves and chemical and hormonal changes. These inputs are analogous to tokens and parameters for an LLM. You could theoretically reserve some tokens to be used as "physical" or "emotional" sensations that would affect the functioning of the model.
Cryo-EM structure of dynein-dynactin-BICDR on microtubules. Those little proteins that look like legs walking on a scaffold
Liquid Death sparkling water. Just plain seltzer turned up to 11 with flaming skulls on the can and boxes
You asked me why I think it's intelligent. Not how it works. Those are two very different things. Behavioral features can be verified without understanding the process at all. This is why we can say humans are intelligent even though there is a lot we don't understand about how the brain works.
The Tet, a giant celestial object from the movie Oblivion, just a big ass metal pyramid in orbit
“Do you understand the work they did. “It won a nobel prize!” Okay. Do you understand why? Because Im thinking you are impressed by the title more than actually understanding the limitations of the work. Also that its more traditional ML work and not what most people are concerned about”
Retired Vice Admiral Robert Harwood and his seeming impostor Wearing a mask of his face
Classic meme with a ridiculously crude drawing of a guy saying “i warned you about [ai] bro!!!! I told you dog!”
Imagine you have this huge unstoppable force. And I think it’s pretty likely the entire surface of the Earth will be covered with solar panels and data centers. Given these kinds of concerns, it will be important that the AGI is somehow built as a cooperation between multiple countries. The future is going to be good for the AIs regardless. It would be nice if it were good for humans as well.” — Ilya Sutskever on the arms race, solar panels, data centers (1:27:20 – 1:28:21)
There is another, radically different way to solve this problem, that is to couple the opening of the valve to something that necessarily varies in response to steam pressure and workload in a way that results in constant flywheel speed. By ‘necessarily’ I mean that the physical properties of this thing respond to changes in steam pressure and workload in a particular way. Such a solution responds in one step and does not require measurement. Thus, there is no time lag problem or concomitant source of error. Nor does it require an executive. In our current example, the thing that responds to both steam pressure and workload is the speed of the engine itself. Hitching the valve control to the flywheel couples the thing you want to control (the valve opening) directly to the thing that embodies the relevant sources of variance (the flywheel speed). So far, so good. The remaining problem is connecting these objects with the proper relation. In this case, we want a negative relation between flywheel speed and valve opening - increased flywheel speed should close the valve, while decreased speed should open it. The solution is to attach weighted arms to flywheel-driven spindle and to link the valve control to these arms (see pic). As the spindle turns, the arms raise and lower according to the centrifugal force generated by the speed of rotation. Faster spinning raises the arms, which pulls the valve closed. Slower spinning lowers the arms, which allows the valve to open. The degree of valve opening / closing varies continuously in response to the height of the arms. The result is uniform flywheel speed linked to oscillating pistons. This beautiful solution is implemented in the Watts centrifugal governor, and 18th century piece of technology that still works brilliantly.
Diagram of a cross section of the brain showing the different sinuses
Nicodemus, a white and grey bearded rat in wizard robes talks to Mrs. Brisby, a little brown mouse wearing a red cloak
Upon encountering a stimulus, an organism’s individual sensorimotor cells or organelles (its subagents) send their perceived observations about it to a local processing center (such as a cellular nucleus or the larger tissue system of which they are part), confirming that something is being observed. Each cell’s output is severely limited, but the collective combination yields something much more significant.
I don’t pretend to have all the answers, but if we work from a starting point of minds as execution states, I think it gets us a lot further to better understanding the hard problem of consciousness. Biology is essential to how our minds work, but it is not essential to how minds must work.
Clips of the character William Knifeman, a spirit guide from Reservation Dogs, doing his epic war cry
Where I work: - Development velocity is very noticeably much higher across the board. Quality is not obvious worse, but it's LLM assisted, not vibe coding (except for experiments and internal tools). - Things that would have been tactically built with TypeScript are now Rust apps. - Things that would have been small Python scripts are full web apps and dashboards. - Vibe coding (with Claude Desktop, nobody is using Replit or any of the others) is the new Excel for non tech people. - Every time someone has any idea it's accompanied by a multi page "Clauded" memo explaining why it's a great idea and what exactly should be done (about 20% of which is useful). - 80% of what were web searches now go to Claude instead (for at least a significant minority of people, could easily be over 50%). - Nobody talks about ChatGPT any more. It's Claude or (sometimes) Gemini. - My main job isn't writing code but I try to kee Claude Code (both my personal and corpo accounts) and OpenCode (also almost always Claude, via Copilot) busy and churning away or something as close to 100% of the time as I ca without getting in the way of my other prioritie We (~20 people) are probably using 2 orders o magnitude more inference than we were at the start of the year and it's consolidated away fro cursor, ChatGPT and Claude to just be almost a Claude (plus a little Gemini as that's part of ou Google Whateverspace plan and some people like it, mostly for non-engineering tasks). No idea if any of this will make things better, exactly, but I think we'd be at a severe competitive disadvantage if we dropped it all and went back how things were.
Just a page of cursive handwritten text, content doesnt matter. Top bit is my regular print chicken scratch, then me trying cursive with the palmer method
Picture of author Quan Mills tiktok video about his new book idea, another hood Yaoi isekai ratchet psychological horror story. Titled “I slipped, fell, and hit my head at a diddy freak off and died and reincarnated as a DL dominican ICE agent”
Thought is not a physical thing. Thought is computed on physical hardware, but *thought itself* is not a physical thing. Your "plane" in your analogy corresponds too *the thing that thinks*. And you "build" it by building the environment that runs it.
Justin Stanley posts: At the park I walk in there's a hill with stairs cut into it. Next to the stairs is a path worn from people choosing to walk next to the path, in easy reach of the sturdy handrail. When it rains (in the PNW) the stairs are often muddy, but the path is dry. End users have a choice.
A symbol is a mark, sign, or word that indicates, signifies, or is understood as representing an idea, object, relationship, or mathematical formula. Symbols allow people to go beyond what is known or seen by creating linkages between otherwise different concepts and experiences. All communication is achieved through the use of symbols: for example, a red octagon is a common symbol for "STOP"; on maps, blue lines often represent rivers; and a red rose often symbolizes love and compassion. Numerals are symbols for numbers; letters of an alphabet may be symbols for certain phonemes; and personal names are symbols representing individuals. The academic study of symbols is called semiotics. In the arts, symbolism is the use of a concrete element to represent a more abstract idea. In cartography, an organized collection of symbols forms a legend for a map.
I have. The problem is not the search engine; it's the whole idea of search. An LLM can create a solution to a problem or a summary of information that's scattered all over the web (and in books). No search engine can do that. At best, it can give me the wherewithal to hand-build my own solution.
An analogy helps clarify the point. Connecting a computer to cameras and robotic arms is similar to attaching measuring instruments to a simulation. The simulation now receives real-world data, but the internal variables of the model are still symbolic representations rather than the physical processes themselves. In the same way, a weather model connected to live atmospheric sensors does not become the atmosphere. It simply receives and manipulates data about it. The same principle applies to embodied AI systems. Sensors and actuators allow the system to interact with the physical world, but they do not automatically transform symbolic representations into intrinsic, experienced semantics. The system may build increasingly detailed maps of its environment, but interacting with the territory does not by itself turn the map into the territory of experience.
Functionalism attempts to explain the origin of the mapmaker (Step 2) by appealing to a process (Step 4) that already presupposes the mapmaker's existence. This is not simply an empirical gap but a category error: it serves as a robust physicalist constraint. The construction of a syntactic map requires a mapmaker from the outset. Therefore, no amount of algorithmic complexity can traverse the causality gap backward to produce an experiencing subject. This ontological inversion inherent in computational functionalism generates a structural paradox: it attempts to derive the foundational mapmaker solely from the mapmaker's own derivative outputs.
3.1 The Ontological Inversion and the Causality Gap Traditional functionalist accounts usually lean on a straightforward, unexamined causal sequence: Physics → Computation → Consciousness This assumes consciousness will simply appear as a downstream byproduct once computational complexity reaches a certain threshold. Yet, as Section 2 established, computation is hardly a natural timeating routine san alred yo be dises aged. ating disape ser bl, and eguing there have no. fundamentally reorder the causal sequence: Physics → Consciousness → Concepts → Computation
2.6 The Fallacy of Computational Emergence When confronted with the distinction between simulation and instantiation, functionalists tend to retreat to complexity theory and emergence. They argue that just as "wetness" emerges from the interaction of water molecules, consciousness will emerge from computation once the system crosses a threshold of sufficient complexity. This objection fails because it conflates weak physical emergence with what we term the fallacy of computational emergence. • Weak Emergence (Physical): Macroscopic properties (like wetness) supervene directly on the intrinsic causal dynamics of the microscopic physical substrate (e.g., H2O). • Computational Emergence (Abstract): The claim that an abstract description of a process (the map) can, solely through a massive increase in syntactical complexity, transmute into the physical process itself (the territory). Functionalists insist that consciousness is a unique case because it is purely "substrate independent information". But this argument assumes its own conclusion: it presumes the mental state is the abstract information (A), completely sidestepping the physical reality (P) that generates it. As we have established, syntax (A → A') possesses no intrinsic causal power; it is a mapmaker's attribution. To claim that an abstract syntax somehow "emerges" to become a physical cause falls outside scientific hypothesis entirely, as it requires violating the causal closure of the physical world
Consciousness is a property of humans biology - and quite clearly not a requisite to intelligence. I say clearly as at some point we reach proof by construction. As in, we already built intelligence because the system already completes tasks that require intelligence. We are so far into what would have been science fiction five years ago and the goal posts have moved so far. For anyone who disagrees, I challenge you to prove deep learning systems cannot solve <task with specific outcome humans can solve but not AI> given sufficient data and compute. I think the strongest sign we have true intelligence already is no one has built any benchmark that AI cannot solve. Yes, our current robotics lags AI, so we don't have the equivalent of the human body to give our deep learning systems. Thus, it's expected AI will be limited in physical scenarios. Second, hallucinations are present in humans. We are highly biased to ignore all the misspoken words in everyday life as we have error correction built into normal conversations. How often do you have to have someone repeat or rephrase something? It just doesn't make sense to me. It's like there are people out there whose belief systems are incompatible with this tech existing. Sure, it has limitations due to training data. It has limitations with no physical body. It cannot combine training and inference the same way a human does. But none of those are measures of intelligence or required to be intelligent.
Christina Koch in side profile while on the Artemis II pictured on top, with two pictures of Hange Zoe from Attack on Titan on the bottom, side profile in the left and slight angle view on the right. Updo, glasses, brown hair
From David S. Goodsell book "The Machinery of Life". Water-color cross section through a single bacterial cell. Surrounded by a multi-layered cell wall (green). The long corkscrew-shaped flagella are turned by motors in the cell wall, propelling the cell through its environment. The interior of the cell is filled with molecular machines for building and repairing molecules, for harnessing different sources of energy, and for sensing and protecting against environmental dangers (70,000 X zoom)
It's truly strange that people keep citing the quality of Claude code's leaked source as if it's proof vibe coding doesn't work. If anything, it's the exact opposite. It shows that you can build a crazy popular & successful product while violating all the traditional rules about "good" code.
Colorful text that says “You are not a Tech Bro” and some grey subtext that says “0% agreement with tech bro ideas.”
something inorganic. Simultaneously believing "birds are super smart" and "LLMs are just stochastic parrots" seems absurd. Jensson > If you disagree, please tell me for which task requiring intelligence you'd rather have an animal's wit than that of an LLM. Navigating your way to a location without collidin with anything. Finding food in the woods. Such stuff that animals can do that we yet have Al be able to do. dinfinity Neither of those are based in intelligence, but rather in dexterity, agility and sensing capabilities. Try again, and this time please rea the question carefully and answer in good faith rather than trying to (unsuccessfully) look for a loophole. gf000 Moving a complex system of muscles so that they can just stand upright is already very ver complex, let alone intercepting a prey's movement mid-flight by just controlling all those muscles. People way overestimate the actually intelligent part of LLMs vs simply being good at recalling context-related stuff from the training data. dinfinity Complexity does not require intelligence. Modern computers (even without AI) and technological systems do incredibly complex things and I'm quite sure you would not call those systems (again, without AI) intelligent.
Fluffy black cat seated on the floor in a sunbeam, closing his eyes at shines on his face.
Marc Andreessen posting the cover art of the book "The Mind is Flat: The Remarkable Shallowness of the Improvising Brain". First he said "If you want the scientific demolition of introspection, this is the book" and then he posted again "TLDR There is no inner self, you're chasing an imaginary concept, the end."
DIMALA SHU-DI, E-LIZ-NID, RU, E-LIZ-NID E-LIZ-NID, E-LIZ-NID, BE-LA-HARU-TI, A-DI-BISH, RU, A-DI-BISH, A-DI-BISH, A-DI-BISH, BE-LA-HARU-TI, E-LIZ-NID, RU, E-LIZ-NID, A-DI-BISH, A-DI-BISH, LU-CA-SHA, GIF-TI, E-LIZ-NID
3. Evaluating Cognitive Capabilities To understand the cognitive capabilities of an AI system, we need to robustly evaluate the system across each cognitive faculty and compare the system to a meaningful human baseline. We therefore propose the following three-stage evaluation protocol: 1. Conduct cognitive assessment of the system across a broad suite of cognitive tasks. 2. Collect human baselines on the same cognitive tasks to establish a point of comparison. 3. Build cognitive profiles to map the system’s strengths and weaknesses in relation to human performance.
The world of artificial systems is perhaps even more complex, rich with a multitude of architectures and training algorithms improving at a pace that makes evolution pale in comparison. It is therefore possible-even likely—that some aspects of human cognition may not be relevant in the context of artificial systems. Conversely, we are likely to find cognitive processes in these artificial systems that humans do not possess.
Youtube comment: I am forced to use Al at work, and boy do I ever. When it inevitably fails, and leadership asks for a root cause analysis, I put the meeting notes into an LLM and let it document its own mistakes. When my boss says that's not good enough, I put their complaint into a new prompt and the cycle continues. They are literally asking for it.
I can now state the physical version of the Church- Turing principle: ‘Every finitely realizable physical system can be perfectly simulated by a universal model computing machine operating by finite means’.
Brilliant insight. The success of LLM reasoning, ie "telling yourself a story", has greatly increased my belief that humans are actually much less impressive than they seem. I do think it's mostly pattern matching and a bunch of interacting streams analogous to LLM tokens. Obviously the implementations are different, because nature has to be robust and learn online, but I do not think we are as different from these machines as most people assume. There's a reason Hofstadter et al. reacted as they did even to the earlier models.
Point of view picture driving on the road behind a logging truck, stacked to the brim with logs ready to fly off final destination 2 style
Stephan Geering @stephangeering.bsky.social Challenge for those who are very confident that Claude isn't conscious: Explain how human consciousness works.