I’m an engineer and this topic is my bread and butter. Please avoid engaging in discussions you know very little about . Your last post was incorrect and a display of deep lack of knowledge.
I know what ai is, is not, and the relationship of applicable terminology.
My posts are intentionally simplified for the layman albeit clear and correct and only to those that have a minimal understanding of the topic and like to learn more.
At the end, generative AI is here with wide practical combat application. The steps we’re currently in are:
- integration with environmental inputs (prompts and RAG—a term I avoided using before). This is an hardware issue not an AI issue. Iran has solved that with their drone sensor subsystems.
- development of a (specialized) foundation model (or off the shelf ) or can be internally developed with good data. Probably supervised since training data volumes are not readily available. vs unsupervised.
- minimizing LLM size for local non-external access. We’re already there since the foundation model and LLM will be highly specialized.
These are all SOLVED TODAY. Anyone can obtain off the shelf OR develop internally.
This post is more or less identical to my previous ones and submitted for emphasis.
I want to add here that all the bullets below are the roots of confusion and lack of knowledge here:
First the old stuff:
- The notions and theories of neural networks date back to the 1940s.
- The notions and applications of machine learning date back to the latter half of the last century.
The next set of bullets are where the vast number of ‘knowledgeable’ people base their incorrect suppositions today.
The huge leaps have been since 2021 with the disruptive introduction of 3 new concepts. These bullets have raged and raced from theory to deep and practical application today in 2024. These three things and the lightening advancement and application are the root of the confusion we see here on this forum:
- transformers presented by an almost unnoticed paper by Google in 2021
- which lead to foundation models
- which lead to LLMs
The paper was later taken by OpenAI which resulted in GPT x.
These final three bullets where PRACTICAL application of GENERATIVE AI is today. Yes lightening advance in about THREE years.
We are at a stage of application now—not theory. Most countries of a reasonable scientific base can build foundation models . These foundation models are in the process of miniaturization. Not theoretically but in practice. Google’s Gemini already has a local version embedded in their new android phones. Apple has an advanced onboard foundation model running which will be introduced later this years replacing Siri.
I emphasize these things are TODAY and widely available for creation by various bodies including Iran.
So no, none of this is science fiction like many believe. We’ve made unprecedented advancement in ai.
Ironically, LLMs are already endangered. Generative ai will be based on ‘general world models’ (GWMs), a new type of foundation model, within a few very short years— much more applicable to UCAVs.
