"Apple Intelligence Is A Failure"

Apple | Intelligence is like any other distributed computer system, layered within Von Neumann dervied binary summing algorithms and the OSI Layer Stack. Private Cloud Computing is sophisticated and currently the best implementation for safe delivery of AI moderated query and it is far from a failure.

Large Language Models are pattern recognition machines that work at a rapid rate. Once you feed back into the client the specifics of what you want, in response to it's response and on, the experience becomes a game of catch and throw, like a harmony between two resolutions. It's breadth of knowledge is astounding but in reality it is far from perfection. This lies in the nature of each LLM having so many parameters, so many tokens and so much scope to draw inference from; for you as the end user.

To the best of my knowledge, the way it all works is:

1. Providers accumulate and train on massive datasets of both copyrighted and non copyrighted works

2. There are rules based standards of ethical query intake and response (I can’t do that, what your asking for goes against terms etc)

3. Session Traffic within the model is moderated via tokenisation; the segmentation of data into meaningful arrays that inform the output matrix on what relates to what from within the datasets of the module.

4. Neurally trained networks that have now for generations been refined with respect to the input - process - output flow of basic functions and theirhierarchical spawn in the application layer.

These are all server side based meaning they have massive resources that are provisioned to them to perform tasks under race to sleep. That means it waits, it sprints, it snoozes. To us, that mimics intelligence. It's just sweeping a big data pool and challenging inference. It's not magic, it's just damn powerful and really deep and wide and meaningful as far as the interconnect goes.

5. Pattern recognition is emergent as a result. Conditioning client side output, establishing and correcting the output's derived meaning and resulting inference value (by way of your followup input tokens) for LLM means that heuristic response curves are inherent in the design.

Simply put - it seems human because we've entrained it with human data and human response. It is just text though.

6. We are fallible. Human bias and perception, positive and negative reinforcement, visual response field, human memory persistence and intent are constantly part of the flux of human perception - which means we are actually, for want of a better term, pre to the LLM speed, depth of knowledge and output that will lead us to be satisfied with an outcome.

But you can't just swing the bat.

Pitch a question at any current state of the art LLM and it will infer and infer and infer until you reinforce it with more tokens (the forms of input validation within a heuristic network that shape how it comes back to you).

If you take the first response and don't demand it reiterate, you will get hallucinations. As the end user you need to point the wheel in the right direction - you need to be narrow and focussed and feed it back like a teacher would a student - otherwise it doesn’t have very specific tasks input into it and a human centred step by step evaluation of output response, for example, will yield second rate outcomes.

When the end user actively reviews the response and manually determines what part of the outcome is useful or irrelevant and either feed back or simple cull the irrelevant and make human use of the good, excellence reigns.

Let it run amok without feedback - and at this point in the flow you hit an inference - success wall. Did the human get the response it wants or did the agent miss?) - without the step by step approach the model will fail to some extent (sometimes wildly and poorly) and still most users will spam chat in andin and in but not give the detailed feedback where it dropped the ball across the feedback models - this is a misgiving to the potential outcome you could receive if you just give it the back and forth like you would in any kind of human interaction.

Remmeber human data = human response curves when you feedback with tokens, or specifics, or simple negative reinfrocement.

In comparison to billions of transistors and petabytes of storage on a dynamic network edge- the human brain is a predictive organ that is subject to amultitude of illogical or disordered presumptions | even in the best of us. It is also connected via multiple sensory and nervous pathways to the brain in a multi layered and multi directional interconnect. Enter Polyvagal Theory by Stephen W Porges if you want more on that paradigm. The world and it's boundary informs us of so much through prosody and play. Machines don't get that and likely never will without us solving the quantum paradigm.

7. With respect to whether wifi can make you hallucinate or have a psychogenic experience. No. It just cannot. Unless you have ionising radiation near the brain, which will just increase inflammation or tau.

A full duplex input output interconnect for an OSI mediated Neumann summing algorithm (an immersive brain computer interface with an user application later) is, dare I say, a literal pipe dream decades if not centuries off. Simply put - an AI can't solve problems it doesn't already have answers to.

It's math.

The brain and body interconect and it's intrincacies are the product of living and visceral organisms functionally both independent and at the same time dependent on each other, connected via afferent nerves that swing in both directions between the body, organs and the brain.

This makes the brain - mind - body interconnect an evolutionary prediction machine (God's finest, IMO) ; where phylogenetic expression and neural array complexity indicates that even though two individuals have the same general layout of physical brain medium - the brain is a matrix - a literal multiverse, and it is connected to another multiverse, the matrix of the body.

Intrinsic nerve pattern propagation is critical to cortical function. What's seen in animal and human brain studies is that the layout of each individual cortical matrix is discrete and unique for each and every human, because environment and genetics and conditioning and play are factors that are unique and dynamic, from conception until death ; so even with invasive procedures like the Musk Neuralink, which has amazing potential for triage and mediation for specific medical conditions ; we cannot and likely will not see a whole body brain immersive experience due to the demands on the visual and neurocorrtical field, the complexity of the matrix within the intracellular tissues of the brain, their interconnect and how we are wired and fired as units.

We are organic, complex, habitual and responsive to the world and its boundary around us. We learn. Machines do math.

Cathery | I am an independent Actor, Writer, Voiceover and have aspirations in systems architecture. My opinions are my own, and are not endorsed by any brands that I mention in my written work. Thank you for stopping by my corner of the internet. More to come.