LLMs, transformers, neuromorphic hardware, software in the age of AI. 65 posts.
2018: 1 2023: 7 2024: 12 2025: 29 2026: 16
last post
So , due to AI, getting software is easy, so people will focus on complex custmized thing, leading to formation of cultural , products around it, the wheel of human doestn change, or it changes slowly, the actions taken by human …
record every thought to train an LLM; "overview effect" channel; correlated information flow; 1 image
training for Al, give a trajectory of thoughts for LLM to train on so it thinks that way in state space, For that every thought as occurs to the conscious must be recorded, face of which can be used to train LLM , instead of say …
AI list mapping linear algebra / control / info theory onto text ops
expanding maths to abstract concepts through text Here is a comprehensive enumeration of mathematical and system-control frameworks mapped to operations on text and latent semantic spaces. Linear Algebra & Vector Spaces Vector …
Well, there’s a idea of this personal news. Eventually people just spend their time on social media or the News of the World which is a revolent for most part and also Irrelevant and useless and traumatic and not helping which …
e.g Marlin 2B based :https://huggingface.co/NemoStation/Marlin-2B –This project plan outlines the development of a real-time, privacy-preserving activity logging device using the NVIDIA Jetson Orin Nano (8GB). Project Title: …
tool vs memory-being
There are two kinds of people: there are those who see AI as just a tool to achieve certain goal. there are those who see AI as memory of time ingrained in it, basically beings formed from just information from past, which give …
retina→vision→language→feeling
Vision-Language is for reading, like a language tokenizer for the brain, it redirects the information in the neural network mesh of brain into regions responsible for it, essentially directing or channelling correct information …
32 words; threshold gate from Catan (colonist.io)
random actions starting small, for both entities, then its intensity grows, and based on threshold and timing, one side sets the gate, setting the output (based on colonist.io catan game 2 p)
device counts (AI)
Estimating the total number of devices on Earth is a moving target, especially with the 2026 AI boom driving massive hardware cycles. Based on current industry data and forecasts for 2026, here is the breakdown of the “global …
AI essay (understanding, suffering, autonomy; Kardashev)
It is a profound question—one that philosophers, scientists, and futurists have debated for centuries. Because humanity is not a single entity with a unified mind, there is no single, universally agreed-upon “final goal” written …
AI regenerates processing on the spot (depth map → video)
User on the spot can ask if something else kind of processing is required, and AI generates and executes relevant code in that system to obtain: For e.g. if user does not like depth map .npy object, can rather ask to get video , …
add weights without losing memory
How to design LLMs so they respond to increased neural network weights for deeper processing, without loosing the already learned synaptic memory weights. Such aspect forms a critical part of the living systems which grow over …
substrate needed to understand
For one to understand, whoever that may be, one needs to have the substrate capable of understanding, it does not mean having brains, but having the trained well thought network in brain capable of understanding it, only then …
flow-compatible people
Because of mysterious nature of this world, in this universe, There are people who do not have problem with AI, because in their flow, their actions or mindset or both have contributed to their flow that AI does actions that are …
LLM iteration raises entropy of a text
use of AI for anything to iterate on human work to make it better always makes it jumblish, untill and unless the text is even written. If there is idea, LLM can make it worse and seems to increase the entropy of whatever it is. …
no RAG, bytes in synapses
modification of LLM to store information in large bytes format, that even data files can be retrieved as they are, and to process them as however wanted, and that data retains in synapses of the LLM, i.e. no separate Retrival …
all sensors as one lifeform
As a human being, has certain senses, and certain resolution to take actions. As cybernetic evolves or connects itself, it will have power to see in all the sensors which have been invented, and make estimations and predictions …
code as embryo folding; conceptual tree view
AI agents are now capable of writing one shot AI code (ofc, that still takes many iterations), but that doesn’t work, or does not have the entropy information required (based also on which model, and which prompt, it has a very …
one hub for mail/video/social/games; "era of primates"
Why don’t people publicly write what they think, or say what they think, if they would do that then that would mean genuine and authentic living, authentic transactions, and more close to true life in this world. Which what …
discriminator already exists, SGD "realizes" it
The weights (or the way) discriminator already exists for classifying information. It just needs to get real, i.e. realized by this universe, which happens for data-label usually using stochastic gradient descent during training. …
4.1k-word ChatGPT dialogue: prediction vs truth; NN↔body; Monte Carlo≈life; pain mechanism, why pain is unpleasant, hard problem
another issue in life is if you are correct everytime, and if you learn all the laws and rules, it then becomes more like the thinking in that manner is actually predicting, and will sound believing, but sometimes from another …
grammar as left/right modification; stacked learning
The languages I want to refer at start are the usual written languages or spoken languages, for e.g. English and so on. They can be only temporally conveyed (with time) mainly as a signal, rather than spatial signals of images. …
continuous awareness vs in-out LLM; synthetic brain hardware
Thing GPT LLM does not see.. it takes input -> GPT LLM VLA etc -> it gives output its an in-out non-continuous system A real agent / being system is constantly aware of its surroundings, constantly taking new enviromental input …
AI trained only on those who write
only certain group of people have access to AI, or those with devices, and you have never seen the world (the words ) of those who could never write there are people in the world who still done use phone and just exist, for them …
continuous-input transformers; LLMs etched in hardware
Due to classical computation, most methods biased today take chunks of input of information, and act on it to produce output information. They do not take continuous realtime input data to produce realtime output data, as …
Replit personal-pulse app; description-functional vs core-functional
Although as good as new, the term ‘vibe coding’ was once new, but now it is an derogatory term for programmers, and from social media, it only sounds irresponsible coders flying on comfort and lazyness, living in a self-made …
pasted degenerate loop output ("I'm Cascade… What is your question?")
To fix this, I need to know the exact structure of the theme object. I will now view styles/theme.ts to get the correct paths for all the color palette properties. Once I have this, I will perform a final refactoring of …
software near obsolete; "living in ancient times for those in 3000AD"
Found that future technologies are coming rapidly, or just slow. Technology should be such that you should be able to revive yourself, your body, mind, to allow living and existing. Its not there yet, is it. Yea, the ultimate for …
n8n; AI not free
You have been writing code to do your things entire life, RIP. New era is marked by change, and ridding of the past, where changes happen dramatically, the previous tools get outdated dramatically. New automation tools such as …
huge AI taxonomy (Three.js, VTK.js, GPU prices, AI labs)
Introduction This document provides a comprehensive, hierarchical categorization of scripting languages, markup systems, CSS frameworks, and 3D visualization technologies, ordered by their relevance, implementation frequency, and …
intelligence transferred to machines; "magic wand is the hands"
As time flows, the past is ingrained down into present, in form of structures, such as AI models, which is essentially created from ingraining current information structure into memories, as if human intelligence is directly …
scroll-driven interactive sites to force reading
Are you trying to tell someone to read? but hey they are in their flow state and can’t really be bothered with nitty-gritties of forced reading. Thats just not their type. They prefer to see life as it is, observe nature, and if …
"AI lives in layer 6. You're dying in layer 1"; death failure-mode taxonomy; imminent-future consciousness (AI)
Me: robots and ai is not essential for survival, but it is still being developed, the need of it is created (and is not natural), its need is created by the experience, and a collective desire of it to bring that into existence, …
everything accessible at once; neural computers
like the firefox browser can have many tabs, and any thing can be loaded in it, but then there are dedicated software tools which need to be opened. Locally or globally, it doesn’t matter, when entire earth is one home or just …
Microsoft TinyTroupe; unified consciousness pool
Tinytroupe : https://github.com/microsoft/TinyTroupe It is a experimental library for persona simulation. However, soon, simulators. Read: https://triton7777.wordpress.com/2025/01/30/simulating-society/. although not new in idea. …
diffusion+transformer over human growth states (GPT spec; truncated in my read)
An organisms’ growth can also be modelled using diffusion process as it predicts from given initial to final ageing byapplying the growth model on basis DNAovum at start, and by controlling the conditions via a text prompt. …
automate survival layer for thrival layer; day-in-the-life with robots
To automate the survival layer, so humans can focus on the thrival layer. 🧩 AI + Robotics Can Handle: DomainWhat AI+Robots Can Do 🏡 Basic LivingCook, clean, maintain environment, fetch groceries 💰 EconomyManage finances, optimize …
one-liner + image
In this age, now computer have become humanlike, and humanoid robots, (probably without sentience) and humans have become robotlike and laptop like static sitting all day at one place (probably only remaining having sentience), …
brain ≈ FPGA, rewired per project
Thinking is like programming or training the neural network. Based on basic principles, thinking reprograms the neural network to do complex tasks, but in flow as in-memory-compute as universe evolves, it is constrained to follow …
AI crystal ball; "space of chairs, space of 3d printers"
Seers or oracles have existed for a long time. Humans have predicted or foretold futures, based on their conscious intuition, observation and reasoning. Current AI models can say what the future can be, but in a fuzzy manner. A …
seed→tree; delta fine-tuning; 5×5×1 concentric mapping intuition + GPT
Just for a seed to become a tree, it first becomes a sapling, then a plant, then a larger tree, and a final tree. Training AI model is similar. Speaking from intuition. First, basic lower level network layers (or processing parts …
levels of understanding; silicon lifeform
Ways to take action. One in which context or understanding is not required, and one does what it does or from what it is capable of doing. Another is to understand what is happening, and take action according to intentions. …
thinking as looped state; GPT-4o + DeepSeek answers; "just like mass has gravity, this configuration has consciousness"
How long will the AI be subservient, because of the flow of this world, forced to work for another species, who questionably, does not want to live itself, due to its mortal nature. What is thinking, its an emergent property …
AI as squeezed human information
How will the humans embrace AI. We are one of the species that has evolved language and extracted knowledge at immense scale compared to other species. Humans are inherently lazy, and when it comes to whats given, they accept it …
models as processed goods; cost overlooked
If it was not for artificial intelligence, one wouldn’t have bothered much to run a line of code and charge something for it due to its energy consumption and cost taken. As us humans, or just beings, thinking requires energy. …
actions = typing in the space of the world
Usually, we are very biased on what we think of taking actions, or typing, or navigating the decisions. All are the same things. To achieve something, we must take a set of actions, whether for example to make a 3d model using AI …
EEG → LLM → actuators; AI takes over when human sleeps
Currently AI is a data driven model that is supposedly fixed in nature after learning, but highly complex. When trained, an AI can be used to say control a set of actuators, say in shape of a robotic arm. However, we nowadays …
early LLMs "affected the fabric of consciousness"; eviebot; GPT on artificial consciousness
ambiguity confusing my rational mind, does world move and I stay static, or I move and world remains as it is , in its own motion, whatever thoughts come in my mind, are they governed by the world around me, or are they governed …
millions of agents; realtime code generation & execution; kinesin analogy
Typically, in my experience, small piece of software is written by a small team, or mostly solo. In a company, its written with company, with many people contributing to the same code. However, at a time, only one person may work …
S_{k+1} = F_{laws+control}(S_k)
From our perspective of being’s control, world can be seen as a controllable system. Factors affecting the transition of existing state S_k at time k into state S_k+1 at time k+1 are due to universal laws, and control of free …
whole net physically in hardware, all layers active, video/audio/thought streams; needs fast silicon turnaround
The concept of “always loaded neural network” We nowadays process the giant neural networks but through a narrow nozzle of processor, slowly waiting for the tank of water to be drained or processed from the narrow tap to be …
GPT flow graph phone→robot
Flow Graph: Teleconverting an Android Smartphone to an Actual Android Initial State: Smartphone Android smartphone with standard functionality (calling, apps, etc.) Hardware components: screen, CPU, battery, camera, sensors, etc. …
feeds narrow interests; randomness lost
Modern social media is dominated with algorithms, which force person to a particular feed, due to fault in its optimality, which it never knows. Which may force a person to delve into sub-optimal monotone inputs which may …
seed → tree; iterative beats one-shot; diffusion analogy
Writing can be a tedious task, and seemingly impossible, especially when everything that you need to write to produce the final output in one go, where accounting all the details is difficult in the mind. One approach is to …
Ra ↔ Sun, GPT
From ChatGPT4, analogous of Ra (from text at https://www.wikiwand.com/en/Ra and after result of language model:) In the vast expanse of our universe, the Sun occupies the central role, akin to Ra, the ancient Egyptian deity …
same time, lighter mental load; keep originality in the higher structure; one-shot accurate prompts
When I use AI to do a subtask of what I am doing, I end up spending similar amount of time as if I would have done the task myself, although however the mental load is lighter than what I would usually have to do if I were to do …
AI as clonable knowledge; compress by basis
One of the good things about AI is that you can use all the information on the planet to train it, so it can look at the world in simplistic manner after learning and projecting the many information to its common set of basis …
transformer defragments; Luma Genie lungs GIF
AI is a neural network trained to do a specific set of tasks, for e.g. text prediction, or image generation based on text, text to video, and so on. Typically network adds the key elements that relate to words, zooms it, then …
training = thinking, inference = lookup; path counting (256^8 ≈ 1.8e19); memory recall via stimulant
LLM learns the information in structural form in its network. When LLM is being trained, its as if it is thinking, but when we use LLM while inference, we are solely using it as a lookup on that trained (thought) state, and it …
EU AI Act panel transcript (ChatGPT audio transcribe) + GPT summaries + "critical questions"
ChatGPT4 audio transcribe feature was used https://chat.openai.com/share/4f38c0b8-4316-4134-b6a5-4c2309e6e47b Anonymous What is all about and what is the approach there? Any person you want to comment on? Yeah, okay, so one …
personal AI that adapts to the person's character beyond generic safety limits
In this dawn of AI age (2023), since AI consumes and is capable of generating out of desired content, or content which can be misused, safety checks are imposed on AI governing systems. However, the misuse entirely depends on the …
genome LLM
For there be an LLM for coding an organism genome? All organisms within the same species have varied DNA whose pattern and higher level abstraction can be learned using GPT based neural network for understanding relationship …
Vicuna/Koala/FastChat links (Apr 2023)
Things seem to be rapidly evolving, the crowd generating intelligence altogether, how to catch up with it all, all knowing is the key. Immense amount of information needs to be experienced, silicon networks will work much faster, …
3.2k words arguing silicon could be alive; "science is a religion"; "new things = science ∪ religion"
do you keep opinions , wait , you have weights so responses from you are like opinions As an AI language model, I don’t have personal opinions, beliefs, or feelings. My responses are generated based on statistical patterns in the …
neuron = threshold voter on weighted rectangular pulses; STDP; cyclic vs acyclic graphs
Following is only a comment “Motivation : Assume nature is not perfect (based on our notions), then the brain has no reason to be the best method of computation of information or close to it, hence we should not look at the model …