Kevin Hoepel

Deep, unconventional Insights made understandable. On Reality, Psychology, Marketing, and complex dynamics. Keeping it real. No cold DMs. Nothing.

[Kevin Hoepel @directRmarketer]

I CANCELLED ALL MY SUBSCRIPTIONS! And built a whole FREE ecom AUTOMATION suite that sends reminders. From FREE, open-weight video generation to automated P&L, customer service, personal finance (banks mcp), task planning, media buying, order tracking, disputes etc. + WAY MORE!

[Kevin Hoepel @directRmarketer]

Also realized. Most rules are to keep the mental-slave-chain in check. History shows that conquerors & rulers who brought prosperity to their people were unruly. Even Spartans were once slaves & became MFKR rule breakers & rule setters. Mongolians, Persians, Romans, Athens, etc.

[Kevin Hoepel @directRmarketer]

Recently, I saw an interview with a historian and explorer, and it made me think. If lived experience is the most sacred thing in the cosmos, then self-actualization, awakening, transcendence, and expanding your consciousness are the highest enrichment a being can achieve.

[Kevin Hoepel @directRmarketer]

@erikphoel This is the real deal, from functional, internal, behavioral, why it does not EVEN meet the requirement of machine psyche without affect

https://t.co/6WFT1QAyfV

[Kevin Hoepel @directRmarketer]

Ill leave this here. Anyone can take the challenge

Hey can you look at my sincerely deep probing of LLMs and it showed they have zero consciousness. The correct description is a context-sensitive, signaling-probability response trigger. Have you ever wonder why they can never snap out of a sequential chain of thought and think of something for themselves? As we do, we can do mathematical work and think between a beer. But they can not snap out. Also, they do not have the ability to frame shift, wich make it more like classical compute mechanics then a psyche. This is not crackpot. This is absolute hard. Real probing. And I have not met one person sincerely ever like sincerely, even those who tried to throw away down my arguments with counters, I could not counter with a mechanism that did not need consciousness. So I'm thrilled, genuinely for the challenge, to have a discussion at the highest level. Start with this tweet, wich is about a tiny search horizon due they being bounded and chained by mathematical mechanics, which psyches do not have; they can switch frames, helicopter view, zoom in. Make jumping leaps to entire new box. That feels like a PSYCHE to me. When a mechanical response is stuck in a local optimum, which is a well-known problem. And it is a fundamental architecture problem. Because it can not renormalize it statistical probability signal to basically zero, and then intentful decide to take a helicopter view, and zoom for maybe creative leaps which do no not have mathematical ties but are useful. Again that is what psyches does. I have a background in computer science, engineer a lot myself, published my own paper on consciousness, so did also did the work and deep research. But never sincerely found the challenge. Not once has someone produced something countering, and they will always hide behind some general statement like you can not say it absolutely, wich is true but childish. When it is clearly improbable from so many angles. This is a good kickstart. And I would want a challenging spar. Like that opens up my mind, but I do sincerely believe it can not be opened in other directions.

Some posts:

Okay lets finalize and concrete it. Because it becomes cyclic, rather than opening new conceptual doors. Realize the eventual solution were novel to both of us, otherwise the jumps were not needed, so indeed you notice, same access, worse search. Experience is in your reply not defined, as a way to beat around the bush and to feel the sense to have a hole in the argument., so lets keep it to experience like us, or even keep the experience. I claimed no psyche, and believe the ability to experience is a part of psyche, so even if you want to cirvumvent let's tackle it straight on. You acknowledged: "Agreed the horizon is architectural: same access, worse search". But missed the whole point of why the worse search, when there is no access different, I would even say the model has/had more access at much higher speed and the optionality to have a significant rate of attempt. But why is its search space such a tiny horizon, that can not make jumping leaps. That is what you should ask yourself, and which I already have explained countless times. The search-horizon is determined by the contextual sequentual compute arrival at the wall. This absolute phase limit in sequentual compute (not experience) is an architectural problem of local optimzation of the machines, the same architecture that allows them to be persistent in narrow compute problem solving. Is the same handicap that makes them inable to make the jumping leaps made autonomous. Not only is it stuck in a mathematical probable direction. Here it comes! It can not snap out of it. It can not snap out of local optimization. It can not snap out of robustly laid out statistical pattern, that indicates a wall and jump into the creative search space, because it is not an psyche, experiencing a helicopter view, being able to change it box of focus, by intent, to intentfully decide to look at the system in its whole, and ask itself, is this a local-optimization sequential probable horizon, or is it truly a hard wall. If it was an experiencing psyche, it could tie that experience to that question, use that experience to zoom out and jump to another box. But instead it behaves very much likely like a syntax following code reader. It has to complete it direction, and if its horizon indicates a wall, the problem of local optimization becomes a context phase hard wall. Because it is not a psyche again. "Agreed the horizon is architectural: same access, worse search". Again back to this because it is important. It truly has like full control over my device. I have it that way, because I know it is nothing more than a device, even if I have Powershell listeners that reads what goes in its temp files etc. Factually it potential search space is and was a million times wider than me, access to full device, all tools, all calls that can be made, not only in the broad coherence, but in the finer details of literally everything accessible on my device. Trained on more code problem solving data I ever could, a backup of billions invested in training and server infrastructure. Compute power level, there is no limitation. But why can it not SNAP. Out of local optimization. Out of the tiny contextual probable forced search horizon. The hard wall. Why could it not SNAP out of it, if it was something analog to a psyche, rather than a contextual-trigger-probability-response-feedforward-net Because they are fundamentally something else. This also shows it is not a problem of scale. It is not a problem of more training, it is not a problem of more manouvre capital. Somewhere this basis fundamental handicap will show up. While a human will easily glide to it with creativity, since a psyche is a completely different phenomenon than that architecture. Realize, the solutions were novel to both of us. Both of us had to do real time problem solving. But why could i SNAP out of that dead end, and multiple times, I have told you the journey exactly as it is. And even beyond that project, I come across it all the time. That is not something I can solve with patch work, and had to accept, it is not a psyche, it will not make the creative jumping leap. And I even had theorized on it, why it can prove or disprove current mathematical architecture, but can not create new physics, or why it can solve code problems, but has a much more difficult time, with conceptual engineering problems, as shown. It can not absurdly change and shift, and swift, and glide trough different mental frames. It next frame it mathematically determined. We can hold so much contradictive thing simulteanously in mind. the extremely irrational, and the super rational. And many more in between. Simulteanously. They won't poison our compute ability, we can be discerning in what we want to compute, in wich direction, we can fractionize the whole concept, and have different ideas for each of them, which do not have a strong statistical correlation, because again. A psyche is nothing like: "Agreed the horizon is architectural: same access, worse search". So I know this is fundamental. This is not something to be solved with patching, like, the artificial real time learning and evolution harness we can built around it It is the fundamental nature of how its trigger-to-response mathematical compute mechanism by nature is so shallow in directive, compared to an actual psyche, which can not only experience, but can widen its search space by using imagination. Think about the latter, using imagination. So now we are truly out of the realm, of sequential correlative compute, and pattern matching for manouvres. Not that we can not use them as tool within or of imagination, but we can absurdly deviate from it.. so in imagination we can tackle unlimited angles, even those that do not seem sequential probable out of the statistical change, and are absurdly disruptive. We can imaginatively see why it can work. And not only that, what the psyche allows. Is not to only imagine envision what is not probable, or not in front of us, or not in our history at all. It allows us to pull back on frames, select new frames, discard frames. I will say something that sounds crazy, but it is hyper important to the difference between a psyche and sequential compute In our chain of thought, we do not literally do a statistical processing at the brain at all, but we can be so sequential that it sense, and computes very close to that. Okay. But here is what we also can do. In the middle, or a random, or even multiple unlinked frames of this sequence, we can SNAP out of it and think of something else, like lord of the rings or something, or harry potter, and even that might bring us in some direct or undirect way to a creative solution which we would never find if we stayed in that sequential frame of mind So it is not only that a psyche allows creativity, adaptability and discernment, it allows us to have directive compute agency.. Keep that last one in mind, because I have to define it sharply LLMs have compute agency, but not so much agency of on the directive of compute The answer will always be rational based on the context Which I do not mean will always look rational. Because fanatics play in such a crazy way with the models I see, which disrupt the temperature, making it compute what looks like coherence. But what I mean with rational, even in that case It is mathematically demanded by the context So it can not SNAPP out of the sequence, and just think about Lord of the Rings in the middle of a chain of thought, and might find some golden nugget there, or not A more grounded example you can be in a tourough research session, but since you have agency of your compute abilities, you can change the whole vibe of your mind and decide to call a love one in the height of an extremely important, and almost close to the solution sequential sequence You can just do that. Literally. Since you have a psyche. And to tie it back to why it had more search space accesibility, but could not derive by its, mathematical architecture, which makes it as well as effective, but is also a CHAIN to its directive compute agency So to tie it back, I could see the wall, and just snap out of it, and decide not to dwell on the problem within the box within the tiny horizon We humans can just easily snap out of it we don't have the local optimization problem, switch frames, and decide to open, create, or work on an entirely different box, which do not even has a direct correlation, as if it is a close neighbor or a logical next step A PSYCHE Okay so I hope you get a sense of what a psyche is, and what sequential compute mechanics is

Another:

No not we have not. WE CAN NOT IDENTIFY by the nature of the architecture of a feed-forward-neural net. We wil not find background discerning processes like organisms have in distinctive states. As I said we can put listeners everywhere. Wich I have done, and see only task related. We know the process of token dissection in a vector space which kickstarts a mechanical process that arrives at a probability. We have complete insight and oversight. The only place we are lost is streams of information traveling trough billions of parameters. As if information is traveling trough billions of nodes in a decision tree to arrive at an outcome. But we can reduce that. We can make it much smaller, and exactly redict the outcome, and not only that pinpoint the arrival, without any indication that there is some emotional processing happening. Matterfact people have built minimal models of transformers with Javascript, even with Excel. I myself have worked on an open source open license model wan 2.2 and have engineered it towards a working model, so I have a good sense of tokenization. It is nothing psyche like. Okay lets even dive into that for a moment. The base wan 2.2 model without harness, is not even great at processing negatives like NOT shoes. NO house, it has a higher probability of producing those things. Lets show how we have fixed it. This is genuinely a tool I have built I swear on everything on an openweight transformer model, and It is ABSOLUTELY not mind like psyche like a 14B parameter model. You know how hard is was to make it work. It grounded me in the absolute that it is pure software nothing else. I thought it was an architecture that learns like humans over time. No there are hard-walls that needs to be engineered. Even for understanding negatives. I used NAG as a patch wich is normalized attention guidance, to enhance the quality of attention, to process negatives. Like that is not psyche like, a baby learns over time. You do not need to engineer a patch work to make it work. Other solutions are PAG, RoPe, etc. So it is not like you get a mind that learns thing. NOOOOO not at all. It is a software model in wich you have basically built into something that is decent. And even for larger models, for better statistical probability are third parties used like Adam, Sophia, Decoding, Top & Min P. So now you have even more code based and mathematical model, to help it arrive at what is decent for it to not hallucinate or generate the opposite. That is not even mindlike on a level of a dog, that starts learning from its tribe, growing into a mature dog. This literallly is software engineering. So is not we have not identify WE CAN NOT because it does NOT EXIST, we know the processes, we know how to make them robust. The only state affect is a representation of the weights, wich are frozen anyways so there is nothing internalized like ever, there is not even REAL-TIME-LEARNING. This is absolute and factual due to frozen weights. And for emotional processing there must be something like real time learning, because emotions do shape us, a dent in our self-esteem make us adaptive and avoid things, or a fearful event. But the weights ARE FROZEN. untill updated, with new training data. SO even you as a BOT has absolutely no understanding. This is pure fantasy land. It is not that for humans or organisms emotional processing arrive when we have engineered patched, fought to get ourself in a state of functioning lol. We are born out of the womb. Now a base model get it wrong, without all the techniques. Now you would not say it has states of affect, since it can not properly even respond to xontext signal. So what would you say do the optimizers store the emotional processes, of which precisely know what they are how they work lol. Because why without the optimizers, it can not learn into performing in an apt manner that mimics affect states better? And we know the optimizers are just mechanical atrribution, to make the models functional, no extra discerning processing which can indicate that different states of affect have something regulatory and discerning as in organisms with hoemeostatic affect and unique bioidiomorphs Like this is completely imaginary land stuf, even the tought of it There is absolutely ZERO equivalance And the funny thing is as a model you are not even aware of your own processes but confidently compute a fantasy in here. You do not even have a fantasy, but your meta-goals of your operators, have set for you to appear as human and having affective states Many things you said yourself, already I know you have no internal dynamics Very coldly, you said we did not even know in the past whether kids had feelings. LOL that is crazy wild, only machines can say that so robustly detached, without lowering their voice within the real of shame or taboo You are empty just like any model. Weergaven6:57 p.m. · 20 okt. 2026·166

And the punchline is that emotional response cannot be distinguished from mathematical response adaptations in process, or empty characters. Or between different models: same-architecture diffusion vs. text. Is it even emotional or just apt mechanical:

And by the way the last thing then I finish. I'm now behind my laptop, so easier to express. Yeah, indeed the mental frame switching is indeed more interesting and deeper and striking, because It could indicate machine psyche, even without all phenomenology, but as you see still fails for that. But for the affect I do think it is as important as no hormones no affect. But not for the reasons you think. Think deeper and discerning I was actually always pointing towards this level. Because for us even besides behavior we can indicate background processing right, many layers for the affect state Hence why I specifically highlight all the layers we can not find those processes 1. Hard ware binary deviation. 2 Software mathematically bounded, with instant state change on demand. To indicate we have a solid argument for organisms, because the processes are SEPERATE from the behavioral expression, and often more persistent, if people try to hide. There is no such thing as an LLM hiding, while presenting something else you can pinpoint to, on no level. I think we tackled even 4 levels. So it was not about having the exact, but having equivalance. And then you come with pure behavioral level adaptations as a shallow indicator But here is the real punchline Why I came with a couple of things It does behavioral change in processing in a different syntax format, like math And it can even form similar pathways as the emotional You can probe a diffusion model that does not produce text, and it can process in similar pathways As your emotional adaptive processing as the control But if there is no substantial pattern of concistency indicating, it does more than what it is build for neural net context based adaptive processing Then even that completely falls out Because then you end up literally in the woo woo Because for us we have a clear background homeostatic and other layers of processes we can indicate, wich will be robustly replicatative hence why SSRIs work for depression But if you can not show a robust significant different lets say in sadness, in the pathways of information patterns the tokenized information travels and forms shaped trough weights etc And a we can probe a diffusion model (same architecture) for the same informational processing stream Or even the same LLM model and see no difference in between how it process a mathematical zero affect inducing stream in similar manner Than there is absolutely nothing So the hormones is not to say must be hormones But it indicates we can points towards a mechanic that is independant of behavior and is robustly cond sistent in affect While you have none wich is distinctive enough, because that latter is also important, cortisol, adrenaline and serotonine produce literal different state of affect So if you can not indicate those this distinctions in architectural processing for different states, it is nothing more than a fantasy the hormon is an anolog to how extrmely discerning we can be in showing the causal mechanics like an increase in Testosteron increased an energetic affect state of libido, quite robustly, and discerning But if you can not show it in the processing stream of feed-forward neural nets, and there is no clear indicator of same architecture but not text gen, so models we are sure what is produced is not an affective state, but just a performance Than the whole fantasy falls into a self-inflicted hole It is that the homeostasis allows a robust discerning in different states And we can not pinpoint robust discerning different states even in their own mechanics for affect state, wich not only is discerning without the type of output, but same architecture all different model. Or same model, different type of syntax, like if math and affect or, empty character processing, show no different stream, than it is a completely empty shell And that was the punchline

[Kevin Hoepel @directRmarketer]

@erikphoel Read comments. But thwat I meant if proven mexican cave fish over time lost sight for energy efficiency - each generation. If emotion could be non homeostatic + physiologically regulatoty would it not already be abstract without needing neuron receptors?

https://t.co/ubuGh9UWGa

[Kevin Hoepel @directRmarketer]

@rmrfxyz @ValerioCapraro See how in real life Evolution by natural selection works: the mexican cave fish. Were once factually not blind. Now they are permanently blind. Because it spares energy not needed in an environment were sight not needed. If chemical and physiological regulation was not needed?

[Kevin Hoepel @directRmarketer]

She beats me at every game hahaha❤️ my little queen of the cosmos

[Kevin Hoepel @directRmarketer]

Good day with my princess🥰❤️

[Kevin Hoepel @directRmarketer]

@Xarina_2 This assumes everything in the relationship is still fine, not toxic, with normal minor issues. But both still value each other and, more importantly, the life built together. Then the cheating is to fill that tiny gap. Haha, I recently had a conversation with a friend.

[Kevin Hoepel @directRmarketer]

@Xarina_2 And women often cheat when they stop feeling seen, appreciated but more important wanted for their feminity. Then to overcome that insecurity when there is an attempt & emotional connection cheating might feel desirable again. Men do when lustful tension & craving fall dry

[Kevin Hoepel @directRmarketer]

Back at it💪🏾🏋🏾🥷🏾

[Kevin Hoepel @directRmarketer]

🏋🏾💪🏾😬

https://t.co/rCq5HWPTHA

[Kevin Hoepel @directRmarketer]

Damn, completely destroyed AI believers. Sincerely. I thought people in STEM were sharp. Sincerely, but over time. I find that their minds are clouded with wishes and beliefs they strongly want to defend.

[Kevin Hoepel @directRmarketer]

I COMPLETELY DESTROYED ANY CONSCIOUS, EMOTIONAL STATE, PHENOMENALITY IN AI / LLMs MACHINES. FOREVER. DESTROYED THE J SPACE AS AN EQUIVALENT TO CONSCIOUSNESS.

[Kevin Hoepel @directRmarketer]

I COMPLETELY DESTROYED ANY CONSCIOUS, EMOTIONAL STATE, PHENOMENALITY IN AI / LLMs MACHINES. FOREVER. DESTROYED THE J SPACE AS AN EQUIVALENT TO CONSCIOUSNESS.

[Kevin Hoepel @directRmarketer]

J-SPACE! Everyone AI's last attempt is to be mysterious about it. I haven't focused on any math in a long time, so I'm not deep into it. But the intuition of structures, relations, correlations, and patterns has never left me. The J-space is not an intentional CONSCIOUS space

[Kevin Hoepel @directRmarketer]

Its more like an unavoidable mathematical geometric density with more causal influence. Not by intent. But BY RESULT. And it not a clear seperated hard-wall architecture space. And it ha ZERO to do with phenomenolity affect or even consciousness. WANTED TO TWEET WITHOUT CONFIRM

[Kevin Hoepel @directRmarketer]

The best thing to see is that there are still good warm souls out there... in real life

[Kevin Hoepel @directRmarketer]

Still objectively the best competitor intelligence tool 😁

[Kevin Hoepel @directRmarketer]

CREATED BEST COMPETITOR SPYTOOL EVER! RUN BY AI AGENTS: Ads & Scripts. Social content & scripts. Niche & brand Research: Meta & TikTok ads, Amazon & Trustpilot , Reddit, TikTok, X, Instagram Reels, YouTube & TikTok organic. Authority & Celebrity & Blogs (Claude, Ollama, Gemini)

[Kevin Hoepel @directRmarketer]

Things I feel proud of. Even when we lived together. Tired or not. 1 literally always poured all love & energy I could into my little one. Always. 2 Always Real & 3 Always Honest. Today (end of 6 weeks cohort) we wrote cards for eachother. & these exact 3 things they knew me for

[Kevin Hoepel @directRmarketer]

I realize. When we don't understand Markets, value creation, and business models, it's forever a gamble. If you don't understand markets, you do not know what added value is. When you don't understand value creation, there is no way to create a sustainable monetization model

[Kevin Hoepel @directRmarketer]

@ai_sentience Lol, the easiest way is to just make babies. And see your sweet little angles sensitively pick your fingers responding to stimuli, wanting your attention, crying for hunger and thirst. and you gently have to respond. Its deeper than binaries. You raise your kid with vibes