adinhitlore

Assuming it exists? in addition: if it's without phone app just windows app or is this super retro these days and everyone wants iOS and android? 10x.

So like chatgpt.com but on your mobile/computer without internet access necessary, $0, absolutely free? This isn't a question if it's possible but if it's a good idea? I will. 100% privacy and 0$, why not?

To be precise in 1903 they predicted it will take one million years in the best possible case (sarcasm i guess) for humans to build useful heavier-than-air craft. Well it took 70 days. The very same year the Wright brothers arrived or rather "left"...upwards with their airplane:

https://en.wikipedia.org/wiki/Wright_Flyer

https://www.reddit.com/r/AskHistorians/comments/1ocua1b/in_1903_the_nyt_published_an_editorial_declaring/

or is just me? looks like jobs or personal real estate sales are failing me hard...if i was younger (now late 30s) i would've considered a fake firefighter stripper path, for now i'd persist on this cursed route.

So over the past months I've experimented with many potential alternatives to transformers, one of which is the so called Spiking Neural Networks which, to the ones unfamiliar with them, is an approach to AI which seeks to make AI as similar to the human brain as possible, in other words while transformers or RNN, LSTMs are result-driven (not seeking to simulate human 1:1) - SNNs actually try to mimic the way humans learn exactly. The idea to use them for NLP isn't entirely new (google SpikeGPT), however to the best of my knowledge no one yet has tried this:

* inclusion of episodic long term memory in SNN every-n-number-of-tokens behind (my own idea...); * implementations in C and C# ports without torch/tensorflow (SpikeGPT is in python with torch); * several types of 'attention' and training modes and memory modes; * training/learning without backpropagation; * CPU-friendly in the sense that while it's still kind of slow (unfortunately) at least GPU isn't mandatory

Here are the screenshots of both the c# windows forms implementation and the C/cygwin port...and 2 random screenshots of claude sonnet 4.6 and gemini pro 3.1 about the program:

https://imgur.com/a/SAQqKmm

why is the text generated from seed still far from perfect? 2 reasons: very small corpus and the c# has <100% accuracy.

However the big nice surprise: It seems like grammar and semantics are both learned, this coupled with my idea to include a way for long term episodic memory a long context outside the tiny 'ctx' window can be extended easily to thousands of tokens behind without decrease in speed - could make it a practical program. Generation is also very fast.

future work:

* BPE, right now it's just words tokenizer...not good for code; * did i say "code"? It may actually be a total failure for coding...or maybe not: completely untested; * The program actually has 2 versions, the other one noticeably deviates from this one and it has c and even f# port, however the f# just doesn't work...it always produces complete gibberish...major bug. * never tested on actual neuromorphic CPU, just goood ol' intel universal laptop ones; * python port should be possible; * finally the big test: large text corpus (megabytes) and accuracy over 95% <- ultimate test.

this is a linear RNN I've worked on, with the goal to solve the long range memory problem in RNNs as well as other improvements such as constant speed in generation...the rest of the text is copy/pasted as the original since due to the limited size this is massive overfit (perplexity is under 1.20):

https://i.imgur.com/p6AmBrq.png

Another way explain this is: "like Mamba but with good long term memory and fixed inference/generation speed". At the moment in C only but I bet it can be ported to python.

Hypothetically speaking a .bin model file <4 GB can be very competitive, the rest is just interface like a desktop app. This may be:

1. free; 2. solves the privacy issue so you communicate with your chatbot offline, lack of privacy is huge issue amongst end users. 3. crazy but even in 2026 people don't always have internet everywhere.

This turns the LLM industry from SaaS subscription to a "download gta 4" kind of business model, like a steam video game business.

I know there are many open source models, but they're designed to be 'build yourself and train them on your corpus' kind of solution...it's for professionals, not a ready working solution.

or even better...you can follow the money (aka bill gate's case back in the days_: contact big companies like intel, hp, lenovo etc to embed your AI into their hardware, so they market it as "for free offline AI assistant" while your contract with the company gives you millions, the vast majority of people have no idea what "45% on humanity last exam" means so even if your model isn't gemini 3.1 pro it will be considered a plus, and if it is then even better since it means people get something better without the hassle of paying online to sites they didn't know exist yesterday.

It's also option for IoT devices like watches, scooters, cars, even the fridge - again i'm sure this is a thing but almost anyone these days relies on SaaS even if you download the chatgpt app or use copilot on windows - internet connection is needed and the model is server-side not on your machine.