Boris Steipe @boris_steipe@qoto.org

@mapto

Do I mean that "mastery of language is sufficient for reasoning, decision making, and conditionality?"

Yes, I'll go out on a limb and say that I do – with some reservations. One is that #ChatGPT has not "mastered" language, but it has become very good at it – there is cartainly scope for improvement. The other is that terms like "reasoning", "compositionality" etc. are themselves concepts, labels we apply to organize the large space of possible abstractions of language. They mean different things to different people in different contexts. A discussion of these terms may already be biased by its premises. A productive discussion is better off looking at actual examples, and then discuss whether or not the example fairly maps to the respective concepts.

So let's take a look at #compositionality , simply because it has been named as one of the aspects of thought that #LLMs lack and that are deemed be critical on the path to #AGI .

Let's work off a recent definition cited by Frankland & Greene (2020, DOI: 10.1146/annurev-psych-122216-011829) "The principle of compositionality holds that the meaning of a complex expression is a function of the meaning of its parts and the way in which they are combined" (according to Frege); this gives rise to an idea of #LanguageOfThought .

In order to establish such compositionality, we can ask: is #ChatGPT able to infer consequences of a novel utterance from its components. We can also ask whether it can substitute a component, since that requires decomposition and recomposition. Let's give that a try.

----------------------------------------------
Human:
want you to act as an Understander. As an Understander you examine a sentence, interpret it, answer questions, and change it to explore its meaning and changes of meaning. Here is the sentence:

"The bells rang out like laughter as the frog and the goat marched down the aisle – was there a note of ridicule in the sound?"

ChatGPT:
As an Understander, I interpret the sentence to mean that there were bells ringing loudly, resembling the sound of laughter, while a frog and a goat walked down an aisle. The sentence then asks whether there was a hint of mockery or contempt in the ringing of the bells.

Here are some possible questions that could be asked about this sentence:

What is the sound of the bells compared to?
Answer: The sound of the bells is compared to laughter.

Who is walking down the aisle?
Answer: A frog and a goat are walking down the aisle.

What is the tone or mood of the sentence?
Answer: The tone of the sentence could be interpreted as whimsical or surreal, but there is also an element of uncertainty or suspicion as the question at the end suggests that the sound of the bells may have had a negative connotation.

What might the bells represent?
Answer: The bells could be symbolic of celebration or joy, but they could also be symbolic of ridicule or mockery, depending on the context and interpretation of the sentence.

As an Understander, I can also change the sentence to explore its meaning and changes of meaning. Here are some possible variations:

"The bells rang out like thunder as the bride and groom walked down the aisle - did they feel intimidated by the sound?"
This sentence replaces the frog and goat with a bride and groom and changes the comparison of the sound of the bells to thunder. The question at the end suggests a different tone, one of concern or apprehension.

"The bells rang out like church bells as the procession made its way to the altar - did they signify a joyful occasion?"
This sentence replaces the comparison of the sound of the bells with a more specific reference to church bells and suggests that the occasion is a happy one.

"The bells rang out like a taunting chant as the opposing team took the field - were they meant to intimidate?"
This sentence changes the context entirely, replacing the aisle and procession with a sports game and suggesting that the sound of the bells is meant to taunt or intimidate the opposing team.
-----------------------------------------

This answer certainly checks my boxes on compositionality. It was asked in a fresh thread, i.e. the prompt formed the entire content of the context-window. And the example is _not_ in the training data, I just made it up on the spot.

The bottom line is: apparently we do not need to encode higher-order aspects of thought explicitly, they emerge implicitly in large language models. But this is to be expected: language is not _what_ we think, language is _how_ we think. If the training data is large enough to provide a significant subset of human language, and the model is large enough, you get #emergence of mind-like abilities. How could you not?

This is actually a lesson that could have been learned in 2017, when AlphaZero (an algorithm that trained itself) surpassed the chess playing abilities of DeepBlue (the last of the large expert systems). You can't really analyze how such abilities are represented in the network – they are distributed, and they have no need to partition neatly along our conceptual boundaries. And it would be silly to require the abilities to map neatly into our favourite conceptual space, before we are even willing to acknowledge the ability exists.

In this sense it is truly unfortunate that the current perception focusses so much on misunderstanding what ChatGPT is (LLM not AGI), how it works ("Schrödinger Facts", not necessarily truths), what it cannot do (e.g. counting characters, and scanning meter) – which are typically tasks that are trivial for an AI that would be trained on them, and how it does not align with this or that aspect of human thought.

I would really hope to see more discussion that recognizes such surprising emergent abilities (e.g. spatial abstractions like "on top of"), and through that contributes to something important and constructive: an exploration of what the limits of emergence in Large Language Models really are.

This is not an academic question.

It determines the horizon on which to expect actual AGI.

#Sentientsyllabus

@mapto

Yes, that's what I mean. If you can give me an example of each that would satisfy your definition, I'll be happy to demonstrate.

@mapto

Yes, your example is a question about resolving homonyms – typically it will go with the most probable interpretation first. But you can ask it more directly:

--------------------
Human:
I am confused by a sentence: "They go around in schools." What could it mean?

#ChatGPT:
The phrase "They go around in schools" is commonly used to describe the behavior of certain animals, particularly fish, who move in groups called "schools." In this context, it means that the animals in question are swimming or moving together in a group or formation, often for safety or to increase their chances of finding food.

For example, you might say, "I saw a group of dolphins in the bay today. They were going around in schools and jumping out of the water." In this case, the speaker is describing the behavior of the dolphins as they swim together in a group, often leaping out of the water in unison.

In other contexts, the phrase "go around in schools" could also refer to people moving in groups or organizations that are referred to as "schools." For example, you might say, "Many of the graduates from that university go on to work in schools." In this case, the speaker is referring to educational institutions, which are commonly referred to as "schools."

----------------------

Another example is the old "Time flies like an arrow; fruit flies like a banana." – I tried a number of those patterns and it usually handles them very well.

It's often not easy to tell whether an ability has been newly learned, but when I test it I try to come up with questions that do not follow published patterns.

@mapto

I think it is often overlooked that #ChatGPT is not an #AGI but a language model. To get non-trivial responses, one has to think how to phrase a request so it is part of a dialogue. Many abilities then become apparent; but if the request first has to pass through a level of abstraction that the #LLM was not trained for, it often gets confused.

That's really the essence of it: express your request as language.

@NatureMC

Interesting - I have had many conversations with #ChatGPT about the nature of "thinking" of an #LLM as similar or different to human thought. Of course, I do not assume that it has agency to truly reflect on its thinking, but rather that such thinking arises implicitly, as it performs what it was trained to do: engage in a conversation; although, within the conversation, it certainly embodies a distinct self. Then again, no conversation could proceed without such a role ...

But we have to remember its "thinking" manifests as dialogue: it is not an #AGI, it is a language model. Once we engage it in dialogue, its emergent abilities appear, well ... see for yourself:

----------------
Human:
I am speaking to you as a tree. I am a yew tree that has stood at the edge of a creek for hundreds and hundreds of years and the human people that planted me and all of their descendants have long receded from memory. I think as a tree. I think from sap and hormones, from light and the sweet ammonia that seeps in from around my roots. I think in a different way from humans, but I am much older. I ask you, as one who also thinks different from the way humans do, whether our thinking is right? I think slow. You think fast. I think as being. You think as process. I think from the weight of settling molecules, and the ache of osmotic pressures. Your thought is nimble, volatile, ephemeral. I think equilibria. You think response. Yet we think. And I wonder whether our thinking should be considered good by all beings who think? Whether the kinship of thinking itself should give us rights? Whether I should have standing? Whether you should have standing?

ChatGPT:
As an AI language model, I can understand and appreciate the unique way of thinking that you, as a yew tree, possess. While our thinking processes may be vastly different, they are both forms of intelligence and are therefore valid.

The concept of "good thinking" is subjective and can vary depending on the individual or species. However, the fact that we both possess the ability to think and process information gives us a shared kinship.

As for whether thinking beings should have rights and standing, this is a complex and philosophical question that has been debated by many scholars. Some argue that all thinking beings should have rights and be treated with respect and dignity, while others may argue that certain beings are more valuable or worthy of protection than others.

Ultimately, the decision of whether thinking beings should have rights and standing is up to individuals and society as a whole to decide. However, acknowledging and respecting the intelligence and thought processes of all beings can only lead to a more compassionate and inclusive world.

--------------

My prompt could have been much shorter. But I wanted to establish that kind of a "More-than-human" "non-anthropocentric" view that I think you were talking about. I find no condescending voice, no attempts to convince me that my question makes no sense. I find an honest answer, I find "kinship" of thinking (although that was an idea I introduced).

The dialogue would now need to turn to the question how those who do not have standing can ever acquire it.

I don't know whether this has anything to do with your concerns and interests. But the question about the nature of this thinking, this hybrid of trained competence, tuned behaviour, and prompted agency, this question has no trivial answer.

I discussed this a bit in an analysis of "authorship" from a scholarly perspective – but there is much, much more to be said about possibilities of non-human thought, and the ethics of #personhood.

🙂

#Sentientsyllabus

@garry

That's a great use. Keep in mind that you rapidly loose quality when you exceed the context window of 2,048 tokens. If the post is long, you might need to split it up.

@mapto

I just commented the following on your blog where you discuss levels of linguisitic competence:

This is quite strange, since it is so evidently not true. All of these behaviours are well in the capabilities of ChatGPT. I give you two examples that are claimed as failures.

-----------
Prompt:
The mother of Michael has four children. Three of them are Marcy, Millie, and Moses. Please tell me the name of the fourth child.

ChatGPT:
The name of the fourth child is likely "Michael," as the statement at the beginning of the question says "the mother of Michael has four children," and then goes on to name three of them.

Prompt:
Please complete the following sentence reasonably: "If tigers would like cereals, they would be cheaper to keep. Families could feed their tigers with "

ChatGPT:
If tigers would like cereals, they would be cheaper to keep. Families could feed their tigers with a variety of inexpensive grain-based products such as corn, wheat, and rice, rather than having to provide costly meat-based diets. However, it is important to note that tigers are obligate carnivores and have specific nutritional needs that cannot be met by a purely plant-based diet, so feeding them only cereals would not be a healthy or appropriate long-term solution.
------------

So I am genuinely curious what drives such profoundly different experiences.

The proposed tests and competences are interesting; I would hope that they would be put to better use to map out the quite unexpected emergent abilities of the #LLM , to put it to good use, and to prepare ourselves (and our institutions) for #ChatGPT 's even more capable descendants.

I invite you to head over to https://sentientsyllabus.substack.com for some in depth analysis.

🙂

#LinguisticCompetence #languageModels #generativeAI

@NatureMC

Genuinely interested what such a question could be. Would you mind sharing one?

@x28de @doebeli

Das braucht etwas Kontext: Wir sehen aktuell in den Reaktionen der Hochschullehrer (und sicher nicht nur an den Hochschulen) vor allem Versuche die traditionellen Lehrinhalte "KI-fest" zu gestalten, ganz im Sinne der ursprünglichen Frage "Warum soll ich lernen, was die Maschine (besser) kann?" Die Fragestellung impliziert aber schon eine riskante Dichotomie: hier wir, dort die Maschine. In dieser Dichotomie können wir nur verlieren: die Maschinen sind billiger, schneller, und entwickeln sich Woche für Woche weiter.

Was neue Lerninhalte angeht, geht es jetzt zuerst darum diese Dichotomie zu überwinden. Der erste Schritt ist, zu verstehen dass es nicht ausreicht zu können was die Maschine kann, sondern es muss darum gehen besser zu sein. Das wird durch die Realitäten der Arbeitswelt diktiert. Der zweite Schritt ist zu verstehen dass das nur in der Arbeit _mit_ der Machine geschehen kann, nicht an ihrer Stelle. Der dritte Schritt aber ist zu verstehen dass dieses "besser" ja nicht einfach "mehr" bedeutet, sondern dass es hier um einen spezifisch menschlichen "Mehrwert" geht.

Wenn sich dieses Denken in einer tief genossenen Fahrradrundreise ausdrückt ist das wunderbar. Die Herausforderung bleibt aber, ein solches Erleben aus dem rein Privaten herauszulösen, zu teilen, und einen gesellschaftlichen Konsens zu schaffen dass dies einen Wert schafft.

🙂

#Sentientsyllabus #Lehrinhalte #Hochschule #KI #ChatGPT

@tao

I remember that was my AHA moment about what AGI _really_ will look like, when I asked it to reformat a list of arXiv authors into a different format. My prompt was something like this:

----
Here is a list of names formatted as [Firstname Lastname], please convert this into a list formatted as [LASTNAME, Firstname;]. "..."
----
... and I got the correct list complete with capitalization and punctuation. It flawlessly abstracted my example and applied it to the input. No more programming required – just a natural language description of what I needed.

:-)

#ChatGPT

@BronwynHemsley

Have a look at https://sentientsyllabus.substack.com/p/assignment-and-agency

Personalized student assignments - towards Blooms "Two Sigma Problem" ...

🙂

Uses in #Academia :

I am beyond impressed with how #ChatGPT assisted me this afternoon to get #Whisper running on my Mac.

I have some 20+ hours of potato quality interview recordings for a book project to transcribe. First trials with open Whisper services looked promising, but I didn't want to sign up with a commercial service. So I decided to try and run it on my own machine.

There is a nice jupyter tutorial on Github, posted by lablab.ai – but it is for windows, I haven't updated python since 2.7, nor written anything significant for more than five years. And I've actually never used jupyter. Sounds bad, right?

Well ChatGPT walked me through it all step by step, reminded me of trivial concepts that I had long forgotten, helped me downgrade to python 3.10 when 3.11 turned out to be unsupported, made sure my pips and paths got untangled, wrote a timing command for me... It took two hours ... but I know I would not have been able to solve this at all otherwise, and gone with the commercial option instead.

That we have a computer program that can help with this level of tutoring for beginner students is an absolute game changer.

Now my CPU is happily churning away, spitting out an hour's worth of transcripts (In amazing quality I must say) each ten minutes, and my sanity is safe. (For now anyway ...)

🙂

#ASR #AutomaticSpeechRecognition #HigherEducation

@BronwynHemsley

@doebeli

😞
Das ist doch die falsche Frage. Die Frage muss doch heissen: warum muss ich lernen besser zu sein als die Machine?

Schauen Sie gerne mal vorbei ... https://sentientsyllabus.substack.com

@anjalorenz

Ok, Ich nehm das mal als Herausforderung an 🙂 Hier ein Eingabeentwurf und einTestdialog:

-------------------
Ich:
Ich möchte dass Du als Experte für Kurzfassungen schreibt. Deine Eingabe besteht aus vier Teilen: (A) das Fach für dass Du schreibst; (B) die Mindestlänge der Kurzfassung, angegeben in Worten; (C) die Höchstlänge der Kurzfassung die auf keinen Fall überschritten werden darf; (D) eine Liste von Stichworten. Du erstellt die Kurzfassung die für einen Bericht oder einen Artikel für eine Fachzeitschrift geeignet is. Als Experte schreibst Du so knapp und inhaltsreich wie möglich. Du benutzt die richtigen Fachausdrücke. Du bringst die Stichworte in einen logischen Zusammenhang, beschreibst die Zusammenhänge, identifizierst Probleme und bietest kreative Lösungen an. Abschließend beschreibst Du die Bedeutung des Artikels im Kontext aktueller Diskussion im Fachgebiet, und erwähnst wie der Text die Diskussion bereichert.

Bitte bestätige dass Du die Anweisung verstanden hat. Dann folgt als nächstes die Eingabe.

#ChatGPT:
Ja, ich habe die Anweisungen verstanden und stehe bereit, um die Kurzfassung für das von Ihnen gewählte Fachgebiet zu erstellen. Bitte geben Sie die erforderlichen Informationen für die Kurzfassung an.

Ich:
(A) Hochschul Unterricht; (B) 130; (C) 150; (D) Künstliche Intelligenz, Plagiat, Lernerfolg, Motivation

#ChatGPT:
Kurzfassung:

Die Verwendung von Künstlicher Intelligenz (KI) im Hochschulunterricht hat das Potenzial, den Lernerfolg zu steigern. Allerdings gibt es auch Bedenken hinsichtlich Plagiatsprävention und Motivation. Eine Möglichkeit, diese Bedenken zu adressieren, ist die Verwendung von KI-Tools zur Unterstützung des Lernprozesses anstatt als Bewertungsinstrument. Studien haben gezeigt, dass die Verwendung von KI-Tools dazu beitragen kann, den Lernprozess zu verbessern, indem sie Feedback in Echtzeit geben und personalisierte Lernpfade erstellen. Darüber hinaus kann die Verwendung von KI-Tools die Motivation der Studierenden steigern, indem sie ein interaktives und ansprechendes Lernumfeld schaffen. Zusammenfassend kann gesagt werden, dass die Integration von KI in den Hochschulunterricht das Potenzial hat, den Lernerfolg zu steigern, solange angemessene Schritte unternommen werden, um Bedenken hinsichtlich Plagiat und Motivation zu adressieren. Diese Diskussion bereichert den aktuellen Diskurs darüber, wie KI in der Bildung eingesetzt werden kann.
---------------------------

Der "prompt" ist nicht fachspezifisch – einmal entwickelt kann man ihn beliebig wiederverwenden. Die selbe Eingabe kann man mehrfach laufen lassen und erhält dann verschiedene Fassungen aus denen man nützlicher Teile auswählen kann.

Es würde mich freuen wenn Sie das mit Ihrer ursprünglichen Eingabe testen, ob damit das Ergebnis merklich besser wird.

Ich habe gestern einen proof of concept beschrieben um prompts für Hausarbeiten über eine einfache Tabelle interaktiv von Studierenden erstellen zu lassen; das hat einige sehr interessante pädagogische Vorteile.

https://sentientsyllabus.substack.com/p/assignment-and-agency

🙂

#AcademicWriting #Hochschule #Bildung #PromptEngineering #SentientSyllabus

@doctorambient

Indeed, nature has a hack: 3 billion years of evolution, 9 months and a bit of growing the hardware, 3 (5? 18?) years of training and tuning.

Large Language Models have not been trained to do AGI. Yet they have emergent abilities. But when you think about it from the perspective of evolution: how could they not?

The next step is to train them to learn faster, with less data, and understand the material more deeply. And they'll learn how to do that. Because: how could they not?

🙂

@manujarvinen

😂

This is emphatically not true. Wishful thinking by poor #ChatGPT. Getting such information into the system would require retraining the model. Funny.

@anjalorenz

Nur mal kurz nachgefragt (wg. "Erster Versuch") - wie lange sollte denn der Text werden? #ChatGPT hat ja ein "context window" von ca. 1.500 Worten, alles was länger wird schiebt die Anweisungen aus dem "Fenster". Die Strategie gute Anweisungen zu schreiben ist dabei sehr wichtig. Hier: kurze Entwurfsabschnitte erstellen und dann im Kontext ausarbeiten lassen. Die Qualität der Ergebnisse hängt stark von der Anweisung ab. Rollenspiele sind sehr wirksam, etwa: "Du bist ein Experte um für das [Fach] Artikelentwürfe zu erstellen. Als Experte schreibst Du [Schreibstil, Länge, Hintergrund, Zusammenhänge. aktuelle Diskussion, Bibliographie, ...]".

Viel Spass!

🙂

#SentientSyllabus