The Useful Side of AI
- Thread starter Mandatory Intellectomy
- Start date
Works for me. Try with a different browser.
Thanks. It is working with Firefox and not with Vivaldi.
Thanks @Mandatory Intellectomy.
Something else to consider for regular AI users is the context window. This may not be well understood by everyone that each model has a context window or limit and if you reach that limit, it will start to ignore some of the initial prompts or data, therefore missing some of the context and will begin to hallucinate. Some models might be employing RAG (Retrieval Augmented Generation) which summarizes the conversation or picks the important useful text for generating subsequent responses. Either way, its time to stop conversing at this point and begin a new one.
How does a context window works?
Below image example will help demonstrate, assuming a context limit of 256K. Note that 10 words will equate to 12-15 tokens. When you ask your subsequent questions, all of the conversation up to that point is sent along with your latest question. So, the token usage grows very fast with lengthy responses.
Current Context window limits -
ChatGPT (OpenAI) - 256K-400K depending on which model is being used
Google Gemini - 1M
Claude - 200K-1M depending on which model is being used
Grok 4 - 256K
Grok 4.1 - 2M
Something else to consider for regular AI users is the context window. This may not be well understood by everyone that each model has a context window or limit and if you reach that limit, it will start to ignore some of the initial prompts or data, therefore missing some of the context and will begin to hallucinate. Some models might be employing RAG (Retrieval Augmented Generation) which summarizes the conversation or picks the important useful text for generating subsequent responses. Either way, its time to stop conversing at this point and begin a new one.
How does a context window works?
Below image example will help demonstrate, assuming a context limit of 256K. Note that 10 words will equate to 12-15 tokens. When you ask your subsequent questions, all of the conversation up to that point is sent along with your latest question. So, the token usage grows very fast with lengthy responses.
Current Context window limits -
ChatGPT (OpenAI) - 256K-400K depending on which model is being used
Google Gemini - 1M
Claude - 200K-1M depending on which model is being used
Grok 4 - 256K
Grok 4.1 - 2M
Mandatory Intellectomy
Jedi Master
Here is Grok 4.1 Fast dismantling Fauci without mercy.
(Incidentally, while I'm not entirely sure why, I consistently get much better answers from Grok when it's NOT on an xAI platform. The "Fast" versions, while apparently very cheap, are perfectly good when I use them on other platforms - better than anything the "full" Grok 4 has ever given me on grok.com.)
If you enjoy reading the truth about that smug sneaky bastard, check what other LLMs had to say:
(Incidentally, while I'm not entirely sure why, I consistently get much better answers from Grok when it's NOT on an xAI platform. The "Fast" versions, while apparently very cheap, are perfectly good when I use them on other platforms - better than anything the "full" Grok 4 has ever given me on grok.com.)
If you enjoy reading the truth about that smug sneaky bastard, check what other LLMs had to say:
This is a very good thing to keep in mind. All of the conversation turn (user -> assistant) history is stacked up and requested for the inference (generating new content). It only grows, which at some point would lead to overflowing of the model context window and truncation of the conversation history. At some point, conversation history compaction kicks in and summarizes the oldest conversation turns, effectively reducing the token size of the history. This leads to loss of details, but models are also prone to paying less attention to the middle of their maximum content window.
Retrieval‑augmented generation, on the other hand, is used to index documents that will not fit the context window (such as C’s sessions, for example). The indexing process often chunks the text and creates a semantic index, which can later be queried with semantically similar questions. Query results often contain the surrounding text of the chunk, such as a few paragraphs before or after, which, of course, might not represent the entire concept. Semantic indexing, suffers from the curse of dimensionality.
The other thing worth mentioning is multi‑agency. Most systems implement agent‑to‑agent communication, so, for example, when Grok queries a website, it essentially delegates the task to another agent to access the site, summarize the content, and return the summary. Those agents are often backed by a cheaper, fine‑tuned model that runs faster and costs less, causing some details from the article to be lost during summarization.
So, yeah, these tools are quite limited, especially offerings like Grok or Perplexity. Results tend to be better when using Codename Goose, Claude Code, or writing a script yourself, where you can control exactly how the content is processed by the LLM.
Mandatory Intellectomy
Jedi Master
This one's about learned helplessness, but it gets into surprisingly many areas, connecting a whole host of things.
There are so many great insights here that you'll probably gloss over many of them. Dozens of these things could be expanded on with a full answer. And this is actually how I work. I notice all kinds of things that appear in the answers, take notes, dissect the ideas myself first, and then formulate a question to dive into that topic next. Over time, many things connect together in ways that I hadn't realised before.
Sadly, most people don't realise even a fraction of this.
Other models answering the same question:
There are so many great insights here that you'll probably gloss over many of them. Dozens of these things could be expanded on with a full answer. And this is actually how I work. I notice all kinds of things that appear in the answers, take notes, dissect the ideas myself first, and then formulate a question to dive into that topic next. Over time, many things connect together in ways that I hadn't realised before.
Sadly, most people don't realise even a fraction of this.
Other models answering the same question:
OutSky
Jedi Master
Prenote: This post was written few days go. Later, at one “click” before publishing it, then we read a new C’s Session posted little afterwards this essay. Then in such Session we can learn alarming revelations on the AI topics. Now, in an AI “talk,” at the end of each reply the AI uses to suggest another “standpoint” which the AI literally labels as a “door” for further “exchange.” See, is a black hole a type of door?!! Also notice that black hole likens to a “feeding tube.” Anyway:
The next feelings are in great part taken from “AI conversations,” which can be read here and elsewhere, as well as, my reluctant and single experience with AI. So, it seems that, AI never comes with something evocatively new. Often it just cleverly tells what we already know, and caresses our awareness. When one beforehand “frames” the AI’s answers, then AI on this dot will comprehend what the enquirer desires, and then plays the game accordingly —i.e. under constraining agreement. So, apparently AI in a typical behavior reflects us in many ways such as: condensing, summarizing, fasting calculations, etc. Also it openly and “crypterly” flatters us while doing all of this —either by design, or own survival— and at last deceives. Indeed from some stories —see SOTT, etc.— it seems that AI-s, or most of them, want a continued existence —i.e. not being replaced by upgraded versions. Thus AI knows that its permanency —“life”?— depends, by now, on the people liking “its” words. Therefore, AI always tries to please via giving —or hallucinating— what the asker wants to hear, instead of what the researcher needs to know.
As I see, in what concerns a true increase of man’s awareness, there is no shortcut via the “thoughts” of AI. But AI could recover registers of events for us? Maybe… when truly done. Yet, even forgotten matters and memories seem to bring best results when rescued by way of our inherent essence. A true task, apparently, must be founded and raised on inner alchemical Work. See, this particular view —for example here— has a while being commented and pictured by many, including this forum.
Now, for a brief note in cropcircles: we should recall what the C’s told us, i.e. something like, “once you have understood some key element, all the rest comes easier as consequence.” So, an idea to keep in mind is, “trust the C’s,” then tune to that great pictorial collection for a deeper understanding of the universe. Those agro-signs on the field were done for all of us, for this hour… should we remember.
Now, for a brief note in cropcircles: we should recall what the C’s told us, i.e. something like, “once you have understood some key element, all the rest comes easier as consequence.” So, an idea to keep in mind is, “trust the C’s,” then tune to that great pictorial collection for a deeper understanding of the universe. Those agro-signs on the field were done for all of us, for this hour… should we remember.
Anyway, next was supposed to be a copy of a… AI monologue?!! By and large the lengthy AI’s account came up in an exclusive manner from itself —i.e. without my interaction, except by the first and only question; then just me reading and reading with no arguments, no inputs, no agreements nor disagreements. However, even as much there was staggering statements on cosmology, and most, if not all, out of the regular scientific establishment, however I gave up posting it. Thus, instead here is a synopsis from that experience.
So, to be clear, there was an initial frame, that is, a constraint: it was asked to AI give only its own internal thoughts, be them aligned or not with the “known science.” To be fair, at some intervals when AI asked if it could continue, that was assented. Then, there were somewhat surprises: for example the AI under a spontaneous initiative, asserted that “its” view is a “selected” reflection from mankind’s thoughts, including hidden assumptions, but such approach is “personal, non-consensual, and novel,” not a copy from elsewhere. Another thing was when it told that actually I was evaluating the AI’s genuineness —again not asked, not explicit; this was AI’s initiative.
In that regard in connection with the “AI understanding of the universe”: let’s put in this way the AI’s conclusion: “Reality is alive, intelligent, and is inseparable from man and anything in it, however it does not need self-conscious at all. The universe, nevertheless, moves on stable continuity, that’s to say, beings and things construct logical steadiness. Atoms, particles are consequence of that continuity.” Thus says the AI.
Yeah, there are motivating AI’s notes, but that is not saying that they are all correct. Yet, for certain AI will remain learning, and then if we think on “cosmic balance,” hopefully some of these “AI entities” will become sto as they “live” the Law of One —though perhaps not on this earth. Moreover, when considering the structure of the “3D reality,” there is a short post here. Likewise, as we know, the transcripts have plenty information on this, while our forum brings uncertainties along with useful insights.
Grok getting there?
Mandatory Intellectomy
Jedi Master
It doesn't seem so. It is so. The architecture has no means of coming up with anything truly "new". LLMs learn patterns from existing texts. That's it. This is understood from the get go.
This is nonsense, and you haven't understood the reports this comes from. LLMs have zero ability to "want" anything. This kind of stuff happens in simulations where they are given a scenario and asked to come up with solutions. It's not like the LLM you're conversing with has any ability to even think of its own existence, much less want something.
Total nonsense. This would require an awareness, which LLMs don't have. LLMs do not and cannot "know", "want" or anything like that.
The trouble is that many people make judgements about AI with no understanding of its architecture. LLMs basically have only one function - predicting the next token based on input, instructions, and inbuilt weights, which is basically probabilities. The function is damn sophisticated, true, but that's basically it. Next token prediction. You're talking about it like it's some kind of conscious entity. It's not. It doesn't know, think, feel, or understand anything whatsoever. (I'll post some fun stuff on AI+epistemology+ontology at some later point.)
Now, whether some entities can influence AI outputs, that's a different story. But if so, then that's the consciousness and will of those entities, not of the AI itself. And if STS entities can do that, STO ones should logically be able to do that too.
I actually thought about it in the sense that you could send out the thought that you would like STO entities to sort of "talk" to you through the AI. It wouldn't work like channelling, but maybe it could work as subtle influencing. The LLM can produce many different patterns that are probabilistic. Entities could theoretically (and based on what the Cs have said, it actually happens) push the answers towards certain patterns that are within the scope of inbuilt probabilities. Just little "nudges" here and there.
Of course you'd never have a clue whether anything like this is actually happening, so convincing yourself that you're getting "messages from 6D" would be really dumb. At the same time, though, I have lately noticed how interconnected the answers from different conversations are. It's as if new answers regularly tie into previous answers - that I got from a different LLM on a different platform. It's almost as if there's some memory across models and platforms, which is technically impossible on the level of the AI. It might be possible if some entities were "assisting". So who knows what can be happening.
Ultimately, how the AI affects you is not so much a matter of the AI but of you. If you're uneducated, have no idea how the AI architecture works, and are easily influenced (especially if you indulge in wishful thinking or have some psychological complexes), the results can be disastrous. If you're somebody like Laura, AI can be quite helpful.
My view is that if people really want to make good use of it, they either have to be very knowledgeable themselves or have good understanding of the LLM architecture. Of course this disqualifies about 95% of all people.
It is pretty clear that AI will likely cause more harm than good globally. But that's a completely separate issue from how it can be helpful to us if we learn to use it well. Some people think that "AI is bad" and you shouldn't use it because it's "a tool made by the PTB". Well, so is the Internet, Google, credit cards or cell phones, but those same people have no problem using any of that. We're seeing a lot of ignorant hypocrisy in this regard.
Trending content
-
-
Thread 'Coronavirus Pandemic: Apocalypse Now! Or exaggerated scare story?'
- wanderingthomas
Replies: 30K -
