Boris Steipe @boris_steipe@qoto.org

@jon

No, no ... it's not about whether it makes a difference. I see the training data as something of a huge library that contains a significant and growing share of human thought. The way I see this, that is actually quite beautiful. Of course, I share many of your concerns about abuse, but then again – aren't our ideas like our children? We let them into the world to change it on their own terms. We hope that they are principled, and strong, and make a difference - but we could not wish for them to be ignored.

🙂

@Colarusso

David I have many thoughts on http://sentientsyllabus.substack.com and many people have found that useful. In particular the last two notes on educational objectives and on authorship are key to the discussion.

Good luck!

@jon

I am genuinely curious why you would want to do that. Thank you.

@dougholton

Great - this is a very practical synopsis. I have one suggestion though: as I explain in more depth here ... https://sentientsyllabus.substack.com/p/how-much-is-too-much ... there is a risk of starting out with a prompt: you are giving up the precious opportunity of a "first-thought", an unbiased engagement with the topic that prepares the mind. Thus I would ask students to first think for themselves, perhaps write down a few bullet points, and then prompt ChatGPT. The algorithm sounds very authoritative, and it takes quite a bit of effort to see its arguments clearly, to sort them by importance and relevance, and to check their logical connections.

You want to avoid them to get locked-in to the generated text's way of seeing thing.

So: "Do start all engagements with thinking for yourself."

🙂

#SentientSyllabus #ChatGPT #HigherEd

@soniashah

Happens a lot. I wrote about that here: https://sentientsyllabus.substack.com/p/chatgpts-achilles-heel

I can't complain about the new culture of fact checking that this requires.

#SentientSyllabus

@appforce1

Re. TTS - I trialed Elevenlabs.io yesterday and was very impressed with their quality.

@emilymbender

I respectfully disagree for two reasons: (1) XAI (explainable artificial intelligence) is a very active research field for exactly this reason - to get the kind of introspection that one needs to get things right. But even if your statement is limited to the current state of GPT type models (2) it only takes the very minimal step of connecting the output to a bog-standard search engine to get sourced and referenced information. And that is current reality with search engines like Neeva, Phind and Perplexity.

Also, "not designed to get things right"? I prefer to think of it as superimposed truths - I have called that "Schrödinger Facts" https://sentientsyllabus.substack.com/p/chatgpts-achilles-heel - and explain there why even this level of factuality is not useless.

🙂
#SentientSyllabus #ChatGPT #HigherEd #

@Frederik_Borgesius

Sehr guter Artikel, vielen Dank!

Der Amerikanische Kongressabgeordente Ted Lieu hat gestern in einem NYT OpEd zur Schaffung einer regulierenden Behörde nach dem Modell der US FDA aufgerufen. Da scheint sich einiges zu bewegen. Wie immer hinkt es den Realitäten etwas hinterher. Aber wie Hacker und Engel richtig schreiben wird es richtig schwierig werden hier eine Balance zwischen Bürger und Industrieinteressen, zwischen Innovation und Meinungsfreiheit etc. zu finden.

Dabei entsteht aber ein neues Problem: fundierte Lösungen brauchen Zeit, und die Entwicklung schreitet gerade so dermaßen rasant voran, dass die Rechtsvorgaben schon technisch überholt sein könnten bevor sie überhaupt in die erste Lesung kommen.

#SentientSyllabus #ChatGPT #HigherEd #AI #legalframework

@OverlyHonestEditor
That's not what they mean. What they mean is that the technology behind it allows a user-interface that is much more compelling, and threatens the ad-revenue model that Google has used to build its business. The answers we get from ChatGPT are superb - making sure that the are also true, and backing that up with existing links is a very small engineering challenge. It's actually even being solved right now with services like You.com, Perplexity, Neeva – and the one I think is the best so far among them: Phind.

🙂

#sentientsyllabus #chatgpt #google

@markcmarino@mastodon.socia

This is nice, Mark - but I have a suggestion. As I explain in more depth here ... https://sentientsyllabus.substack.com/p/how-much-is-too-much ... there is a risk of starting out with a prompt: you are giving up the precious opportunity of a "first-thought", an unbiased engagement with the topic that prepares the mind. Thus I would ask students to first think for themselves, perhaps write down a few bullet points, and then prompt ChatGPT. Remember, the algorithm sounds very authoritative, and it takes quite a bit of effort to see its arguments clearly, to sort them by importance and relevance, and to check their logical connections.

You just want to avoid them to get locked-in to the generated text's way of seeing thing.

Cheers -

#SentientSyllabus #ChatGPT #HigherEd #writing

@chrismessina

Yes they do, but that's not the correct approach IMO - it identifies the page as a journal, the company as a publisher, and it omits the author. Neither is correct: authorship is a hybrid of prompt, AI, and data; the site is not a journal, but a communication channel like an eMail; and OpenAI is not a publisher since they have no specific agency in the process - and actually exclude accountability via their TOS.

The IMO correct approach is to define a NEW category "Synthesized Communication" and structure the genertaed text like a "Personal Communication":

@misc{chatgpt_2023,
author = {ChatGPT},
title = {Synthesized Communication},
howpublished = {\url{https://chat.openai.com/chat}},
note = {Accessed on 2023-02-01},
}

🙂

#SentientSyllabus #ChatGPT #HigherEd #AI #Education #Authorship

Many comments are posted on OpenAI's work towards a classifier that distinguishes generated text from human written text. According to OpenAI's own announcement (a) input texts were not taken from an adversarial source, i.e. no deliberate attempts were made to obfuscate AI authorship; (b) there was a false positive rate of 9%, i.e. for 9% of the texts the tool evaluated a human written text as being AI generated.

These two points make the tool currently unfit for an academic misconduct allegation, where the false positive rate has to be practically indistinguishable from zero, and where efforts to obscure the authorship will be encountered.

Indeed, there is a short video in a response to OpenAI's tweet, in which a researcher copies ChatGPT output, runs it through GPT-3 – and the AI author is no longer recognized.

Also, the true positive rate of only 26% (i.e. what fraction of AI generated text was recognized as not having a human author) is rather sobering. Though this is hardly surprising: AI generated text is optimized to appear natural.

I touch on the topic in previous analyses at https://sentientsyllabus.substack.com – and misconduct will be the focus of the next newsletter at https://sentientsyllabus.substack.com

#SentientSyllabus #ChatGPT #HigherEd #AI #Education #Plagiarism

@jentrification

No - and yes. The input does not get fed back into the the "training data", that cuts off in 2021, but the conversations are used in experiments for meta-improvements, i.e. how to improve its ability to dialogue safely, helpfully, and respectfully.

@Weltenkreuzer
Ich habe eine Umkehrung des Sokratischen Dialogs im heutigen #InsideHigherEd Artikel von @susan_dagostino vorgeschlagen. Nicht die KI steht in der Rolle des Sokrates, sondern wir selbst; wir sind als herausfordernder Sokrates, gefordert unsere Argumente klar und überzeugend auszudrücken, um hinter den oberflächlich plausiblen Antworten der KI tiefere Einblicke zu entwickeln. Das geht erstaunlich gut - die KI ist geduldig, kann sich klar ausdrücken, und hat ein breites Wissen. Die Aufgabe ist nicht trivial, so ein "Gespräch" kann tatsächlich interessant werden.

https://www.insidehighered.com/news/2023/01/31/chatgpt-sparks-debate-how-design-student-assignments-now

#SentientSyllabus #ChatGPT #HigherEd #AI #Education