Policy on use of ChatGPT

Beware! This might not be the post you are looking for. Must really be wanting it. I made it my duty.

@petri999 said in #28:

Well @dboing you text for so long that asked LLM to summarize :

Sent by Copilot:
Sure, here’s a summary in bullet points:

Ok. This is WIP10
(I deleted wonderful sharing about that "" syntax, but I don't think the room is ready for that).

Ok. before I start diggiing and chunking to approve. Here is my WIP (clink clanck clunk!) plan of approach

I plan on no more than one line to say yes, no, maybe. maybe some other day, I will make another pass. But for now. I am going to read the below quoted AI output and if I agree with the idea there I will say yes, even if the AI made a new point (unlikely, but I am diligent and transparent for this experiment to be most informative about my intent and the ouput relation. So, my first pass for my own limitations, is just reading below .. and yes no maybe. or what in the hell!?

@petri999
I have one question or 2. What prompt made this. exactly? how many bullets points? can it find some optimal number from the text input? (I assume from LLM afar, but ML familiar that there would be some metrics in there, and clustering possiblities for the new input text data. But my question of bullet points extraction, is not dependent, I just think it analogous in having that choice to make or provision for it to make a call from the input itself.. I am not LLM detail expert at all. But I might not need too, to ask such questions. (analogy with chess users of chess engines as analysis microscope instruments without a manual, somewhere lurking).

I think if user could have some control there to experiment, I could actually find it an attention engaging charade to figure out how to reoder or cut chunks and put them back in a more reader friendly order (as my storms might actually expand in various places, if I do the error of trying to fix above my typing pointer... (more nighmare for all). but sometimes it is my thoughts themselve that make me realize from my writing, that some logic element might have been missing.. etc.. I hope I don,t need AI for this to explain a bit my spamming appearance. Not an AI, not a smooth writer. and yes. not native english (worse native from a language that is too proximal in too many similar words, although in my language we have the correct ordering in our phrases.. fyi).

Cheating in Chess:
Sustainable cheating to appear as a human player with a normal rating trajectory and human-like fluctuations.
Ultimate goal: to invoke or evoke the appearance of a genuine player.
Question: How good would a player be without engines, given current chess engines’ abilities to play at human level?

Yes. But since I referred to the content of a clip which was not part of the input data, it dismissed that context, and did not notice that I was making or proposing an analogy between the clip statement about AI and our human forum expectations (helpdesk or discussion as here). And the question was in passing, a tangent. This is me writing.

It does see that the vertical order is not having a spatial meaning or ordering the prefix being more important for each sub-bullet point. It just missed the purpose of "cheating in chess" regarding this thread. Sorry, I want to make this interesting for me too.. so yes.. but. I hope this is not calling for another round of AI-ification....

Linguistic Helpers:
Different tools for language correction and enhancement.
Use of tools like Language Tools and MS Editor for correcting and rephrasing text.
Extensions of correctors help after initial thoughts are written.

All above: YES.

Whole-scale paragraph rephrasing tools may not always capture unconventional or complex thoughts accurately.
Potential issue: rephrasing tools might average out unique insights into more conventional language.

see notes below for my tourist impression. But yes too here. Very good. I had a caveat, but then did the bottom notes. Something about purpose. too late.. gone.. (this is why I am in hurry to write, before my thoughts get lost).

Honor Code for Language Tools:
Suggestion for an honor code in language tools.
Transparency in showing the prompt or method used for generating text.
Importance of maintaining the integrity of original thoughts, especially when publishing online

Yes

again not sure of the ordering, I might not even have a need for it. But it does seem to already have the elements. The "especially when publishing online" is a bit of a surprise (another pass, for myself maybe where did it find this in my verbal encrypted stream...)

comment from Petri about above and below Co-pilot experiment.

it did not too much sorter but it works other way as well you can give list of your ideas and ask concise paragrah for it lets try

thanks for the hints. Putting the prompt output review in the next post.

I agree with having that bullets first if I were to check on the elements first (perhaps that is already part of the process of digesting new user input, anyone? is that what the tokenificatoin is about.. clustering of some kind?)

It was better to make such map of elements than mesmerizing smooth prose where it might be losing the human newest input "outlier" information its massive data set trained model would not have room for or detection/discerning ability (there are 2 sensitive human design fudging point about that problem, and we only consider one of them, even in research, the other being stuck in marketland and whose data is it, but the science of the dependency of the sampling structure design or choice, I don't seem to hear much about it. It is not always the LLM architecture or the training pipeline. There is the big data fog at the dimension level, and what is representing what (I had to get there, to connect to chess learning question of generalization from training to what exactly?).

ooops.. did I at least say yes?

The bullet part is actually impressive, but it might be that my wall of text was just difficult from its size not from it readability.
Anyway I still find it useful, since I would get lost reading myself too (kind of unfair, right, I can pour, but I can read, well I also find it an annoyance, I wish I could sustain equivalent replies to my post in discussions).

I only realize now the remaining of post is Petris own take. using the tool to help the thread. 2 in 1. So only keeping the prompt part. I do have a question. Are you @petri999, asking for content creation along those constraints?

With prompt:
write short non verbose essay on use LLM for forum text using following points:
it is ok use LLM to proof read
is acceptable to ask LLM to write short decritption idead writer provides
it not acceptable to give broad description of idead being discussed and copy paste the output

Thanks for the bullets. And I take this question being a sort of summary of this thread traget question(s).
I will try to read the output. but wonder the actual source of it besides the above constraint input.

Done with my feedback on my intent versus Co-polit LAI bullet points.

But curious about confirmation that the actual @petri999 post second part only had the prompt constraints.

I would think they might have worked hard on some basic things to compensate for the data gaps or whatever was allowing the trained model to go very creative some concoction from its latent quirks given the data around the gaps. Ok, I am making suggestions or questions, I don't know much.

Here is anotherer they might have to work very hard to compensate for possible meaning (or "dimensions" of it) never actually being part of the whole modelling problem my question to the linguistic AI "universe") and actual coded model parts human design generalization.

So we might be reading what they wanted it to say. As this question is likely to have been on their deployment research... (something like that, no clue about the lingo there, this is above my curiosity, as my curiosity might often be in another universe altogether, I don't speak corporate at all, but sometimes I can smell its presence).

Does such AI hallucinate about well accepted scientific research that has lots of reproduced published data to confirm and the care of the writers to make their logic and discussions as explicit within their disciplines as the academic standard would have it (if they were not being under pressure otherwise, i.e. in some ideal world, or topic of research where that might be true).

But I do agree with what came out. Just, it might not be the generative model speaking.... lol. How could we know?

PS: I realize I am saving on the shift key that goes with having to aim i and shift same time, too busy struggle to express my wonderful limitation and pain growing sublimating new ways to figure out how to share and also test my them, better just give up on the pronouns. I find that typing distraction concise. So. Please feel free to put pronouns of your favorite kinds where there should be. But I accept criticism about sentences missing the main dish...

** I am guessing from crumbs and some familiarity with machine learning models at some mathematical level, and internalized visual support. I am making suggestions or questions.

going wild or free.

This is probably not the post one might be looking for. Your eye gaze suddenly goes up the Lichess top menu, lobby is calling you. Some blog might be jumping in trepidation for your attention. Go there. You will find what you thought was here.

i.e. asking questions I find related and that chess and this thread is inspiring me to humbly (really, I just love ideas, first at least, I seem to manage that.. Not abandoning all hope of actually doing something tangible, but I found in past, much of what I could do, started with spending time formulating questions, and finding more information, and so forth. (and bathing in other discussions, when I was not damaged). other precautions. If this is spamming. I will put it elsewhere.

Other lurking question of mine in case someone has some clue about that.

The nature of meaning

This is not philosophy despite the appearances, I am asking this as a research question. Chat GPT 3 and chess using PGN but not FEN (as often PGN don't come with it. given that we don't use illustrious "from position" PGNs that much. Yes. The MFTL brain, that which is at the core of this new technology. That technology is model of our animal brain boiled down to some basic characteristics. At some point what is an approximation or model of what, depends on our ability to formulate correctly both what we can obtain as data, and what we would want to figure out as a culture or community (science?) from it. Care in the questions formulation to make use of existing models. One has to drop some illusions of we know it all. I can't say it else. The old wisdom of acknowledging our limitations. It might not be a competitive winning attitude, as one might think it means having no ambition, I can say. I have been forces to think that way, and I have no merit. My life trajectory environment might have it. So believe me. I am speaking from that kind of freedom to look whatever you want. I just want to share and express and figure out by this motivating task of explanation, what is boiling inside me that I call creativity or imagination and theory.

Damn I need some real therapy. Thanks Lichess for holding my fort.

Now the actual thinking I just took infinite precautions announcing.

The nature of meaning that we are expecting such things to understand from the language data and mean of communication. I am having maybe a difficult relationship with language, native or not, just the fact there is only one string available while inside our minds it does not really work that way.

We at least have more than one homunculus worth degrees of freedom, and if I connect the crumb I read because of @NDpatzer excellent blog series, that imagination is now being declined at least with terms like "sensory recruitment", allow me to make some question by association and some hunch (being human, and transparent about my reasoning tools)

We have many homunculi, in various brain structures, not just skin maps but whole locomotion degrees of freedom (proprioception included, motor pattern from functional logic muscle relationships, if I might use my own words), and then at least for visual testable task, that sensory recruitment crumb I am clinging to: why would sensory recruitment be limited to the retina sensory pathways. That we can consciously talk about it, but what of the other internal dimensions of the senses .

How much flexibility do we have in transposing existing things and projecting ourselves in such internal models, without actually doing the motion or perceiving as real the signals. I am guessing.. but I find that the question of meaning can't just be all contained in language data alone. Or have human languages conquered the universe dynamics..

Yep. chess is a fascinating source of inspiration or divagation (french?). At some point, the pleasure of ideation is what sustains some hope for more. And when that fails, there is always the chessboard individual learner mysteries to recenter one's reality. some wood. Also, is that old news? anyone. I don't think I have the resources to figure that out. and keep afloat. help welcome. at least i would be able to build on such new informatino or pointers.

place you bets on whether I trigger some spam alarm.....

In summary of above post. The question is how much can language hold of the reality we consider part of our intelligence, so that we might consider the AI sentient. or at least actually able to answer all our questions. as we as assuming now, in the complete ignorance of the question of the training dataset conception and design (yes half the model that humans who make those things or more, is there, and we just don't even ask, well. some do. But not at the hype scale. So. Loads of research questions are left handing during the circus (ok caricature, but there is a layer of it, and it gets in the way of true humanity (not just the usual few) progress and purpose of keeping such pursuit going. Otherwise. Since everyone is going back to glorious narrative. Creative slaves of the world unite!

Hey I was doing self-derision about the therapy. Just having written that did the job... why I thank Lichess for existing, chess also implied.

News update. I started a guenuine web documentation based wishful long conversation with chatGPT consumer chat free website of the same brand.

I now understand better some of the human exasperating text production it can lead to, besides the good useful stuff. It has not self-awareness of its own quality of knowledge (they just dumps a big shapeless data blob onto a LLM, and bet of the huge size to sort all those reality representativity hidden issues to magically sort out themselves, or at least I can hunch away that impression, and not even close to any evidence that would make me regret blabbering that hypothesis).

It can shrink, but it can also expand vaguely at will or user control and the default seems to errr on the wall of text, but not my kind, where the problem I have is fitting all I am considering into some ascii noodle stream. with only CR or LF visual management. which also depends on my taking a breadth while take notes of my incomings things I want to share..
resulting in another kind of despair, as I mess my initial intentions, through considering alternate understandings of what I might just have written (or thought). Exasperation from not being able express myself, what I see but does not seem to fit with my verbal written skills. (or such few time windows and flukes when I might, that trying to reproduce, becomes its own frustration). And I understand, that I might be another source of wall of text frustration. But I think for the opposite causal factors. I always fear having missing an important logical or reasoning or hidden assumption not being shared by lichess unknown internet scope mind diversity and luggages trying to read this..

I find the AI have no problem with my breathless content. as they don't breath! and they are hunch machines. with uncharacterized huge massive blob of corpus (if not so, we would like to have access to the partitions and existing conceived and found characterization measures, as the things are on neutral navigation, powerless about their corpus information so that maybe a user could help the verbomotor machine, generate more logical AND information increase text things to read.

there seems to be a huge abliity to chew on text and it seems to be completley independ of informative content measure (that any human actually in pursuit and in effective digestion of any new information, might end up after prolonged interogation and overlapping input meaning, start figuring out, that their last hours might have been pain in vain (ok that might be me).

That figuring it out what is actionable information that seems cogent, if it had been me on the other side (lots of the human side projections structural design artefact going on, either oblivious to human psychology or the opposite (and I would bet it would be the chat interface layer more than the web content corpus. Language AI made chat bot. I think that might be analogous to putting the horse in the cart. and the whole things not one any road withiin any life sustained planet in sight.

In conclusion. And since one can tune the text wall verbosity to actual informative content after individual research project with many tete a tete on same topics.. I hope that will carry to my other attempts, when I get hopefull again (memory fading, not a myth, everyday, we can fool ourselves into forgetting certain things, and do it again and then same results etc.. you known the drilll)..

in summar of concluion. or in reality of conclusion that just derailed.

Lichess should impose its open source and open data philosophy to public forum and blog (even) about using such tools beyond mere correction. and rephrasing, shrinking or bullet condensing (those are not the problem, but knowing that there were used would increase confidnece in the wall of text not being pain in vain). And most importantly. if someone is going the other way. For example given few keywords, and then ask for a haiku, to maximally spread the actual human provided information in those keywords, say one word in each motif.. Then if link put on top of post having used quantiity of words out high ration to their input in, someone among the many lichess users coming into contact might warn the others by having checkied the context real human prior creative or informative content that was context to the produced well writtent and in the worst cases vague or not informative. (some people confuse general with vague.. I think. Maybe a lot of people, and maybe the web is strong with that "force", at least when wanting more depths as human. They still seem to be limited to small brain me.

ok. I might have super AI strength in wall of text. Same results difference methods.. sorry for the pain.

so. here. on the other hand and nonetheless, but in executive summary:

Obligation to make the long conversation that lead to the post content, a public link, and post that link. and maintain the link (it can be restricted to the converastion at the date of the link creation) through the life expectancy of thread.

If the chat tool being used does not allow that. then maybe obligation to put the verbatim of the conversation in some text hostinig website.

which I realize would require the chat bot to index there conversation. and not rely on their vague but assertive and reassuring claims that they have all the conversation in memory, or that one can request that they do for the verbatim. That they have their own inert memory indexing, might be good for them. but we would need here to ask for visible indexing .

in conclusion. The sloppy and untraceable rational typical presentation to try to fool us into a casual atmostphere of chat, might be great for the product propagation appeal, but I would ask here. that something of more transparent pressure be explored as policies.

There are serious efforts and even some using such AI but it is enough from small amount of pain in vain posts. to raise general suspicion itself in vain, as most economical assumption. More noise in the human to human communication is not acceptable.

so when using AI in good faith effort, please be aware, that it might be fooling you into making sense. And do share your actual true input out of your mind with it. and then feel the appreciation of sharing your true self through the sugary wrapper danger of loss of information.

can someone shrinnk the above in bullets of half liners. In that direction we are safe. Also maybe find out how to force an AI to be consistent with it own pretension of multipost memory in one convo. Ask it to put the bullets line into some meaningful groupings. it is good at that. as this does not really require any notion of non-linguistic source of meaning. Natural language either through population needs for efficient communication, or with grafted academy futile written form overhead rules, tends to contain a lot of associative "logic" (possibly common human semantics causative factors), or sequential syntactical roles of verbal stream segmentation dynamical relationships. sorry for the verbiage. best i coudl do. in this overspilling last words.

I am not saying it is useless. but the work to find the BS zone. is a long learning curve, in my own abliities to understand such trained model (whch should be having some reproducible internal mechanisc to figure out, even thourgh the smoke of proprietary and marketting and public concerns ad-hoc code patches or interface selling appeal (the chat as a person bait).

The LLM generative model, should have an invariant probablilty law for its behavior. Conversations should be reproducible. if we had at least transparency on the interpretable parameters values used thorugh a conversation..

But the more obscurity layers, the more a waste of energies it can induce for us. I think one had to be trained almost as much as in chess, to be able to get the soft spot of input nudging and restriction to be able to find the ration of human energy put in and the information gotten out to start making such technology usable. Clue: ;the amount of web pages about the art of prompt. prompt engineering. long tutorials wth lots of acrobatic wall of text in and out, and renetrant.

I even have doubts now about the coding tutorial I saw in the line of web articles trying to show how their one example did work. but as I realized above. there is not tracecablility conceived to go with the tool that can manipulate the thoughts of many with less energty or spin doctor HR work force budget . That is about my rambling above being in vain finally. It might be a populatoin pandora box . it was already bad before. now language is losiing some trust I would say. and not about my difficult relationship. i mean people even here. So humans. be careful.