And with that I close this chain of controversies, which I don’t need actually.
There is a bright future ahead.
Peace.
In the last post
there was reported an unlikely double time coincidence concerning the behaviour of unknown CS arXiv moderators, when faced with two articles submitted to them in the same time: Anthropic’ arXiv:2310.13548 and my article, see version submitted here, which was rejected over the weekend with no argument and the submission deleted immediately. I completed it then with a section about how to counter a valid argument from AI summary, see updated version, and is now available as
Buliga, Marius (2023). Argument from AI summary: How does asemantic computing differ from traditional distributed computing?. figshare. Journal contribution. https://doi.org/10.6084/m9.figshare.24420967.v1
Then, one hour after I made that last post, I received out of the blue a message from arXiv mods that they are willing to change their decision if my article is accepted by a journal. That’s coincidence number two.
Mind that the first coincidence does not imply that the conception of articles are in any way related. It is just that the moderators treated them differently, even if, by chance, my article presents an use case of AI which is resistant to the other article so called AI sycophancy.
OK, then, you can read the revised version of my submission, where I include a cute self-referential section about how to counter an argument by AI summary, using the input from arXiv moderation. I won’t bother you with repeated links.
Now though, what if I check with phind for clarity of the argument from AI article?
I prepared a md version of the revised article, where I added also the reference to the Anthropic’ AI sycophancy article arXiv:2310.13548 and then, because here I can’t share md files, I saved that file as
Then I asked phind. I proceeded like this:
In the tradition of the argument from AI summary, here is the archived evidence.
UPDATE: Remark that you can copy the prompt from the archived evidence and then save it as a file.md (delete only “Summarize the following text:” string). There you go!, you have the md version of the revision.
…
Well, is it clear enough so that even a sycophant AI can reflect back a conformal summary, or that an anonymous moderator cannot claim that it not interesting?
You tell me, because I also reproduce it in the following.
The text by Marius Buliga, titled “Argument from AI summary: How does asemantic computing differ from traditional distributed computing?”, discusses the concept of Argument from AI summary and asemantic computing.
The Argument from AI summary is a method to determine if the written communication of an idea is clear or not by using an AI tool as a mirror which reflects back the message sent by the human author of the idea. The human author asks an AI to provide a summary of the thesis shared by the author. The summary is then checked by the author. If the author confirms the summary, then the conversation with the AI can be used as proof that the communication of the idea is clear enough [Source: Text].
Asemantic computing, on the other hand, is a computing method that operates locally in time and space, without needing to understand the overall state of the system. It does not require global control or meaning, and it uses Turing machines and graph rewrite systems. The data in asemantic computing is processed without any inherent meaning. Asemantic computing does not require confluence, which is the property of a graph rewrite system where if we have a graph that can be rewritten into one without any further possible rewrites, then this final state is unique [Source: Text].
The text also provides an example of how to counter a valid argument by AI summary. The author presents a real-life example of how an institution can counter the author’s argument using a reversed time order. The institution accesses its own black box and offers a response that the request was not valid, without any evidence of the proceedings. Finally, the institution produces a judgement decision dismissing the thesis, attributing the lack of resources to the author [Source: Text].
The text concludes with a bibliography of sources referred to in the article [Source: Text].
telegra.ph > Argument-from-AI-summary-How-does-asemantic-computing-differ-from-traditional-distributed-computing-10-26
Your custom link…
archive.ph > n909G
Your custom link…
How does the counter argument by AI summary process work and what are some potential implications or consequences of this approach?
The counter argument by AI summary process works as follows:
Potential implications of this approach include:
Potential consequences of this approach include:
Their article is arXiv:2310.13548 and my article, see version submitted here, was rejected over the weekend with no argument. It was completed with a section about how to counter a valid argument from AI summary, see updated version, and is now available as
Buliga, Marius (2023). Argument from AI summary: How does asemantic computing differ from traditional distributed computing?. figshare. Journal contribution. https://doi.org/10.6084/m9.figshare.24420967.v1
See also the post here How to counter a valid argument by AI summary, to see arXiv response.
Why is this relevant?
Because of the following reasons.
This is a very strange behaviour here! Of course that arXiv:2310.13548 is a very interesting paper, but why not mine? After all they are very much related and mine is not sensible to their’ sycophant AI phenomenon.
What do you think? Let me know.
UPDATE (26.10.2023): Yesterday, Oct 25th, hours after I posted this, I received out of the blue the following from arXiv moderation:
“Our moderators will consider an appeal if this submission is published or accepted in a conventional journal with a resolving DOI (Digital Object Identifier) or link to the journal’s website showing the status of the work.”
What? Second time coincidence? Anonymous moderators, more likely, covering their…. Can’t be the arXiv behavior.
I tried “summarize arXiv:1103.6007” with phind [0], archived answer [1] and I discovered I’m the proud author of the latest most important AI paper: Attention is all you need.
First I thought that’s yet another anti-chemlambda item.
Then , what if it’s a bug?
For no matter which article I tried, it always responds with an unrelated deep learning article summary [3] [4] [5].
That’s like… Sometime in the future, suppose you need an urgent heart surgery. The medical insurance proposes to be done by the latest AI driven robotic surgery device, what can be better? (or cheaper…)
The operation goes well. You wake up with puffed lips and silicone implants.
Wouldn’t that be nice?
I write this because it seems that arXiv articles are not freely available to everybody (AI tools included). So, if even arXiv does this gatekeeping, it means that the others, less green scientific publishers, will do even worse.
Hence Sci-Hub AI to the rescue :)) ok ok, anybody with data access, capable of an easy to use interface and might to solve the problem first and be useful than to gatekeep, to the rescue!
Would it even be possible? Is a technical problem, of course it is possible.
I’ve done more recent experiments with asking phind.com to “summarize arXiv:…”.
Here, for the first time, it does a rather good job with arXiv:2007.10288, see archived answer [6]. It mentions access to arXiv via export.arxiv.org.
Maybe, I thought, Phind adapted and solved the “bug” (I think is more complicated than a bug…). So I tried again the first query, about arXiv:1103.6007. No! it says the article is not online…
Maybe only the more recent arXiv articles are accessible? Let’s try arXiv:1206.3093. Here is the archived answer [7]. Is funny! Is written by Jean-Pierre Serre and it’s in french! But a half of the answer is reasonably close to reality.
OK, newer: arXiv:1305.5786. Here is the archived answer [8]. Well it works, that’s reasonable as an answer.
But when I tried new ones, it was not reasonable. Confusion in the case of the new arXiv:2304.14158, see here the archived answer [9]. Remark the same split brain behaviour as previously, where the search indicates the correct link but the answer speaks about something else.
After experiences with phind.com summarize arXiv:… , I tried to see if phind can summarize (and has access to) github md writings.
So I asked phind to summarize [10]. Here is the archived answer [11].
Remark that now the answer is given by phind v4 model, while in previous experiments the answers were by gpt 3.5, turbo or 4 models.
Further is a version of the answer minimally corrected by the author of the text: only one small thing is not correct. That is remarkable!
In particular phind does understand the Wittgenstein joke. I wish that human programmers follow the phind’s lead.
“The document you’ve asked to summarize is a discussion on Wittgenstein’s philosophy, specifically his thoughts on language and reality, and it also includes a story about the rhinoceros.
The main section of the document is titled “A Wittgenstein joke”. It’s a joke that’s framed, told, and punchlined. The framing sets up the idea that the book’s main point is that what can be said at all can be said clearly, and what we can’t talk about we must pass over in silence. The telling section consists of various excerpts from Wittgenstein’s work, which focus on the idea that there must be objects for the world to have an unalterable form, and that the existence and non-existence of states of affairs is reality. The punchline section closes with Wittgenstein stating that the thoughts in his book are unassailable and definitive [10].
The story of the rhino thing is a historical account of how the first rhinoceros to arrive in Europe in over a thousand years created various rhinoceros objects. The rhinoceros was seen by scholars, the curious, and its image was depicted in various places. This image, created by Albrecht Dürer, became an iconic representation of a rhinoceros, even though it contained some errors. Despite these errors, the image remained popular until the late 18th century when more accurate representations of rhinoceroses were made [10].
The reality thing section of the document discusses the idea that reality is a discussion or trial that happens in places. The author believes that objects, which are reifications of these discussions, are evidence used in these discussions, [this part wrong:] but they are not objective[end of wrong part]. The author argues that this is a failed philosophical program, as it first relies on objects and then tries to bring back interactions for objects. The author suggests that this has led to the meaning of things and objects being mixed or even reversed [10].”
Socrates discusses with Wittgenstein. At some point Socrates claims “A”. In response, Wittgenstein replies with “your A is B”.
Socrates then, aside, asks an AI to “summarize my argument A”. The AI answers by producing a text. Socrates reads the text, archives the conversation with the AI, for reference.
Then Socrates tells Wittgenstein the following.
“I listened your reply “your A is B” with attention and I believe that you misrepresent my A, by mistake or on purpose. Or perhaps I am at fault here because what I think about the subject is not clearly enough communicated in my argument A.
I don’t know which one it is, therefore I asked this AI to summarize my argument A, in the form I told you.
Here is the AI summary, as archived and also as corrected by me.
It seems that the AI clearly writes what I intended when I communicated A to you. There is only one small detail here, which is not important for our discussion, see this correction.
I ask you then: Wittgenstein, why do you reply with “Your A is B” when even an AI can get the gist of my argument A but you behave as if you don’t?
Your reply, in the form you told me, does not respond to my A. “
The discussion continues…
I tried the same initial question, ie “summarize arXiv:1103.6007”, this time with ChatGPT Online [12].
The result is the following:
“In the paper “The New Physics of Bird Flight: From Flight Planes to Acrobatics”, the authors explore the physical principles behind bird flight. They discuss how birds are able to generate lift and control their movement in the air, drawing comparisons to the principles of fixed-wing aircraft and helicopters. The paper also highlights the acrobatic maneuvers performed by birds during flight and the underlying mechanisms that enable them. The authors propose a new framework for understanding bird flight based on the interaction between aerodynamic, biomechanical, and neural factors. They believe that this interdisciplinary approach can contribute to the development of advanced flying machines inspired by bird flight.”
Of course that this has nothing to do with arXiv:1103.6007. Moreover I was not able to find any paper titled “The New Physics of Bird Flight: From Flight Planes to Acrobatics”, with google or ddg search!
It’s pure invention, this AI does not even try to locate an arXiv ID, is just useless drivel.
Now, I know that there are services online, to summarize arXiv articles. But they are either demanding a login or are exclusive to chrome.
Why would that be?
The answer, in my opinion, is very simple. Is the same answer as, for example, the one about Open Access: why would the open access research articles should be Gold (ie paid by the authors) or otherwise not recognized as publication? Or even if a technical solution appears for OA, with no cost, is then prosecuted? Because otherwise one major pillar of the society would no longer be controlled.
Now we see, in the making, the phenomenon where these new AI are trained on huge quantities of data but they are unable to process a tiny, tiny quantity of data. Indeed, the collection of all research articles ever written on Earth is tiny compared to what they use to train these AI.
So why? Because, while talent appears more or less evenly distributed, the scientific superiority and access are one of those sensible pillars of the society.
What would it look like if we could trace and compare the genesis of the ideas in research? Can’t be a public, free service, right? Most of research would appear to be badly written drivel, compressible to an even tinier quantity of data which will turn out that is authored by an evenly distributed collection of people.
This is an important question. It is also an example of argument from AI summary. Such an argument aims to establish if the communication of an idea is defective, by asking an AI to provide a summary. The summary is then vetted by the author of the argument, as proof that the communication of the idea is conformal with the intentions of the author.
With such proof, the author may ask then: if the AI gets it, why don’t you? [ie the dialogue partners]
Here is the AI answer to the question from the title and here is the archived answer.
The edits are only a link to the AI used and the replacement of the text “geeksforgeeks.org” with “asemantic-computing.md” for the links provided in the AI answer (only the links name are changed, not the links themselves, of course).
Start!
Answer | Phind V6 Model
Asemantic computing differs from traditional distributed computing in several key ways:
In the Argument from AI summary was proposed a new use of an AI tool, as a virtual check for clarity of a thesis communicated by the author of it, in a rational discussion.
Then, in How does asemantic computing differ from traditional distributed computing? was given an example of this kind of argument, for the thesis of asemantic computing.
This was turned into an article, in initial form [15], which was submitted to arXiv (where else? as you well know, I am a long time user of this excellent tool of dissemination which is, or should be, outside of the legacy publication business).
As a response, I obtained a real life example about how to counter a valid argument by AI summary.
The article [16] is a newer version of [15], completed with a new section. It can be cited as
Buliga, Marius (2023). Argument from AI summary: How does asemantic computing differ from traditional distributed computing?. figshare. Journal contribution. https://doi.org/10.6084/m9.figshare.24420967.v1
In order to understand the how the argument can be attacked, let’s analyze how the argument is built, in time.
As a part of a discussion, the author then shares A, B, C. For this article:
The structure used involves a human (the author), a black box (the AI tool), a judge (the human) and evidence of the process (A, B, C). It is important that the black box is available to all the participants in the discussion.
In order to counter the author argument, the receiver uses the same structure, in a reversed time order. Suppose that the receiver poses as an institution (therefore not a human). The receiver has access to it’s own black box (a part of the institution itself), but differently from the AI tool, only the receiver can access this black box.
As a consequence of this counter of the argument, the discussion ends, in such a way that the receiver does not have to provide reasonable further arguments to the author.
As an illustration, is reproduced here the answer to the submission of this article in the cs.AI, cs.HC sections of arXiv.org, a very reputable repository of articles which serves the research community by providing fast dissemination of research. Some times the streamlined submission process becomes an unsolicited editorial one, as they are not a journal, like in the case of the initial version of this article. (As an author who uses arXiv since a long time, this is the first time when I am part of this process, perhaps due to the unconventional nature of this article.) I wish to thank the institution for the provision of a real life counter attack of a valid argument by AI summary.
The response is edited (only) to separate it into the 3′, 2′, 1′ time reversed steps which were explained previously. Links and general statements were edited out.
[2] https://chorasimilarity.wordpress.com/tag/anti-chemlambda/
[10] https://github.com/mbuliga/writings/blob/main/wittgenstein-and-the-rhino.md
[12] https://gptonline.ai/chatgpt-online/
[14] https://github.com/chemlambda/molecular/blob/main/reading/asemantic-computing.md
[15] https://chorasimilarity.files.wordpress.com/2023/10/asemantic-answer-20-10-2023.pdf
In the Argument from AI summary was proposed a new use of an AI tool, as a virtual check for clarity of a thesis communicated by the author of it, in a rational discussion.
Then, in How does asemantic computing differ from traditional distributed computing? was given an example of this kind of argument, for the thesis of asemantic computing.
This was turned into an article, in initial form available here, which was submitted to arXiv (where else? as you well know, I am a long time user of this excellent tool of dissemination which is, or should be, outside of the legacy publication business).
As a response, I obtained a real life example about how to counter a valid argument by AI summary.
The article, available here, is now completed with a new section.
UPDATE: the article can be cited as
Buliga, Marius (2023). Argument from AI summary: How does asemantic computing differ from traditional distributed computing?. figshare. Journal contribution. https://doi.org/10.6084/m9.figshare.24420967.v1
…
In order to understand the how the argument can be attacked, let’s analyze how the argument is built, in time.
As a part of a discussion, the author then shares A, B, C. For this article:
The structure used involves a human (the author), a black box (the AI tool), a judge (the human) and evidence of the process (A, B, C). It is important that the black box is available to all the participants in the discussion.
In order to counter the author argument, the receiver uses the same structure, in a reversed time order. Suppose that the receiver poses as an institution (therefore not a human). The receiver has access to it’s own black box (a part of the institution itself), but differently from the AI tool, only the receiver can access this black box.
As a consequence of this counter of the argument, the discussion ends, in such a way that the receiver does not have to provide reasonable further arguments to the author.
As an illustration, is reproduced here the answer to the submission of this article in the cs.AI, cs.HC sections of arXiv.org, a very reputable repository of articles which serves the research community by providing fast dissemination of research. Some times the streamlined submission process becomes an unsolicited editorial one, as they are not a journal, like in the case of the initial version of this article. (As an author who uses arXiv since a long time, this is the first time when I am part of this process, perhaps due to the unconventional nature of this article.) I wish to thank the institution for the provision of a real life counter attack of a valid argument by AI summary.
The response is edited (only) to separate it into the 3′, 2′, 1′ time reversed steps which were explained previously. Links and general statements were edited out.
UPDATE: submitted this as an article to arXiv. UPDATE 2: as of 21.10.2023 it appears as “on hold”.
This is an important question. It is also an example of argument from AI summary. Such an argument aims to establish if the communication of an idea is defective, by asking an AI to provide a summary. The summary is then vetted by the author of the argument, as proof that the communication of the idea is conformal with the intentions of the author.
With such proof, the author may ask then: if the AI gets it, why don’t you? [ie the dialogue partners]
Here is the AI answer to the question from the title and here is the archived answer.
The edits are only a link to the AI used and the replacement of the text “geeksforgeeks.org” with “asemantic-computing.md” for the links provided in the AI answer (only the links name are changed, not the links themselves, of course).
Start!
Answer | Phind V6 Model
Asemantic computing differs from traditional distributed computing in several key ways:
I am very glad that the Romanian Academy is a member of CoARA. Why is this important, for me and for my colleagues?
There is a long time and ongoing fight against the assessment of research as if it is the activity of production of publishing units, ranked according to journal impact factors (JIF) or article influence score (AIS).
These criteria, initially adapted from criteria of value of journals, are against the scientific method, which asks for independent validation of research and not for vanity or authority arguments about where and in what vicinity articles are published.
Such simplistic criteria are prone to gaming.
One of the unexpected consequences of such so called objective criteria is that it turns research institutions into factories of articles.
Or, if the researchers become workers in such factories of articles, then a factory with 200 workers will be about two times more productive than a factory with 100 workers. No matter what!
The bigger factory wins, which leads to the squeezing of fundamental research out of funding.
No more fundamental research! It’s not economically viable!
We, researchers, are now cornered in the wrong place, because we felt, for honest reasons maybe, to the trap of using these criteria exclusively.
…
I signed in personal name the precursor of this European coalition, the San Francisco Declaration of Research Assessment (DORA). Therefore I am more than happy to see how my Institute of Mathematics of the Romanian Academy (IMAR) will avoid to turn into a generic factory of articles and it will follow it’s mission in fundamental research in mathematics, as the tradition of this prestigious institute is.
…
Research is not only about writing articles, although an important part of the research is to scholarly communicate the research results.
But today it is possible much, much more , besides articles, which are only a story of the research. Repositories of programs, exhibitions, data sets, experimental data, all this is possible to be shared as Open Science.
Or, the exclusive reliance on journal impact factors is something which keeps us away from communicating research.
You might ask: what about mathematics? Are these so important to research, or the article is enough?
The answer is that yes, even for mathematical research the sharing of much more than the research article is very important.
First, a constructive proof in mathematics IS a program. Programs are proofs!
Then, mathematics does not exist in a void. The first programmers were mathematicians. Now, in the present, mathematics is more necessary than ever in fields like Computer Science, Machine Learning, Biology or Chemistry. Conversely, new research subjects from these fields demand new mathematics.
It is natural then to accept programs, data sets and experiments, be them with primes, algebraic curves or computer science “tainted”, as an important part of the mathematical research communication.
I tried the same initial question, described here, ie “summarize arXiv:1103.6007”, this time with ChatGPT Online.
The result is the following:
“In the paper “The New Physics of Bird Flight: From Flight Planes to Acrobatics”, the authors explore the physical principles behind bird flight. They discuss how birds are able to generate lift and control their movement in the air, drawing comparisons to the principles of fixed-wing aircraft and helicopters. The paper also highlights the acrobatic maneuvers performed by birds during flight and the underlying mechanisms that enable them. The authors propose a new framework for understanding bird flight based on the interaction between aerodynamic, biomechanical, and neural factors. They believe that this interdisciplinary approach can contribute to the development of advanced flying machines inspired by bird flight.”
Of course that this has nothing to do with arXiv:1103.6007. Moreover I was not able to find any paper titled “The New Physics of Bird Flight: From Flight Planes to Acrobatics”, with google or ddg search!
It’s pure invention, this AI does not even try to locate an arXiv ID, is just useless drivel.
Now, I know that there are services online, to summarize arXiv articles. But they are either demanding a login or are exclusive to chrome.
Why would that be?
The answer, in my opinion, is very simple. Is the same answer as, for example, the one about Open Access: why would the open access research articles should be Gold (ie paid by the authors) or otherwise not recognized as publication? Or even if a technical solution appears for OA, with no cost, is then prosecuted? Because otherwise one major pillar of the society would no longer be controlled.
Now we see, in the making, the phenomenon where these new AI are trained on huge quantities of data but they are unable to process a tiny, tiny quantity of data. Indeed, the collection of all research articles ever written on Earth is tiny compared to what they use to train these AI.
So why? Because, while talent appears more or less evenly distributed, the scientific superiority and access are one of those sensible pillars of the society.
What would it look like if we could trace and compare the genesis of the ideas in research? Can’t be a public, free service, right? Most of research would appear to be badly written drivel, compressible to an even tinier quantity of data which will turn out that is authored by an evenly distributed collection of people.
Am I wrong? Prove me wrong.
Oh, incidentally, I tried also to see what happens when I ask the AI to summarize a piece of genius made text, like the madeleine passage from Proust. Result? Incompressible or turns to bland, tasteless, unsubtle, commercial quality text. For another time.
Socrates discusses with Wittgenstein. At some point Socrates claims “A”. In response, Wittgenstein replies with “your A is B”.
Socrates then, aside, asks an AI to “summarize my argument A”. The AI answers by producing a text. Socrates reads the text, archives the conversation with the AI, for reference.
Then Socrates tells Wittgenstein the following.
“I listened your reply “your A is B” with attention and I believe that you misrepresent my A, by mistake or on purpose. Or perhaps I am at fault here because what I think about the subject is not clearly enough communicated in my argument A.
I don’t know which one it is, therefore I asked this AI to summarize my argument A, in the form I told you.
Here is the AI summary, as archived and also as corrected by me.
It seems that the AI clearly writes what I intended when I communicated A to you. There is only one small detail here, which is not important for our discussion, see this correction.
I ask you then: Wittgenstein, why do you reply with “Your A is B” when even an AI can get the gist of my argument A but you behave as if you don’t?
Your reply, in the form you told me, does not respond to my A. “
The discussion continues…
UPDATE: Archived AI summary of this post.
chorasimilarity 