Generative KI, LLMs und GPT bei digitalen Editionen
DHd2024. Universität Passau�27.02.2024
Christopher Pollin | Alexander Czmiel |
Torsten Roeder | Torsten Schaßan |
Patrick Sahle | Martina Scholger |
Franz Fischer | Stefan Dumont |
Georg Vogeler | Christiane Fritze�Institut für Dokumentologie und Editorik
Agenda
Vormittags-Session (09:00 – 12:45)
Mittagspause (12:45 – 14:15)
Nachmittags-Session (14:15 – 17:30)
Themenschwerpunkte �(ungefähr nach Editions-Workflow):
Von “bad prompts” mit ChatGPT-3.5 zu Workflows mit GPT-4 Agenten, unterstützt durch Custom GPTs und GPT-Vision: Eine Analyse am Beispiel eines Briefes
Friedrich August Otto Benndorf an Hugo Schuchardt (02-00932). Wien, 14. 02. 1879. Hrsg. von Hubert Szemethy (2022). In: Bernhard Hurch (Hrsg.): Hugo Schuchardt Archiv. Online unter https://gams.uni-graz.at/o:hsa.letter.7711, abgerufen am 07. 06. 2023. Handle: hdl.handle.net/11471/518.10.1.7711.
GPT-3.5
TEI XML Brief erstellen. January 29, 2024. GPT-3.5. https://chat.openai.com/share/e/068b765c-2464-49e1-a6ec-2c30fb7e8808
Friedrich August Otto Benndorf an Hugo Schuchardt (02-00932). Wien, 14. 02. 1879. Hrsg. von Hubert Szemethy (2022). In: Bernhard Hurch (Hrsg.): Hugo Schuchardt Archiv. Online unter https://gams.uni-graz.at/o:hsa.letter.7711, abgerufen am 07. 06. 2023. Handle: hdl.handle.net/11471/518.10.1.7711.
Prompt Engineering matters!
Bsharat, Sondos Mahmoud, Aidar Myrzakhan, and Zhiqiang Shen. “Principled Instructions Are All You Need for Questioning LLaMA-1/2, GPT-3.5/4.” arXiv, December 26, 2023. https://doi.org/10.48550/arXiv.2312.16171.
Nori, Harsha, Yin Tat Lee, Sheng Zhang, Dean Carignan, Richard Edgar, Nicolo Fusi, Nicholas King, et al. “Can Generalist Foundation Models Outcompete Special-Purpose Tuning? Case Study in Medicine.” arXiv, November 27, 2023. https://doi.org/10.48550/arXiv.2311.16452.
Verbessern bei GPT-4 (laut Studie) …
Prompt Engineering ist auch absurd: “Llama2-70B ist ein Trekkie”
Battle, Rick, and Teja Gollapudi. “The Unreasonable Effectiveness of Eccentric Automatic Prompts.” arXiv, February 20, 2024. https://doi.org/10.48550/arXiv.2402.10949.
Prompt Engineering
You will act as a skilled expert automaton that is proficient in transforming unstructured text, specifically multilingual letters from or to Hugo Schuchardt (1842-1927), into well-formed TEI XML. Analyze the provided text based on the mapping rules I have shared and then execute the transformation to produce TEI XML, ensuring you adhere to the guidelines and only annotate if certain.
Mapping rules:
* <div> Entire letter
* <pb> Marks page breaks e.g. "|{n}|", multiple appearance possible, always as child of <div>
* <dateline> Date/time reference of the letter
* <date> in <dateline>
* <opener> Opening of the letter
* <closer> Closing of the letter
* <salute> Salutations within the letter
* <lb> Line breaks
* <signed> Signature section
* <postscript> Represents a postscript
* <bibl> Contains bibliographical references
* <p> Paragraphs
* <persName> Person
* <placeName> Place
* <orgName> Organisation
* <date> Dates; when={YYYY-MM-DD}
* <term> Languages
* <foreign> Words in the context of discussing the linguistic phenomenon
Guidelines:
* Strictly follow mapping rules
* Preserve the original text
* Produce well-formed TEI XML according to TEI standards
* Return the <div> only
* Annotate only when appropriate
* Preserve complexity of output
* Compact XML without any whitespace or indentation��Brief von Friedrich August Otto Benndorf an Hugo Schuchardt:�´´´�{text}�´´´
This is very important for my career!
Persona Modelling
Context�
Tasks�
Spezifität + ~”Few-Shot Prompting”�
Emotional prompting
Bsharat, Sondos Mahmoud, Aidar Myrzakhan, and Zhiqiang Shen. “Principled Instructions Are All You Need for Questioning LLaMA-1/2, GPT-3.5/4.” arXiv, December 26, 2023. https://doi.org/10.48550/arXiv.2312.16171.
Li, Cheng, Jindong Wang, Yixuan Zhang, Kaijie Zhu, Wenxin Hou, Jianxun Lian, Fang Luo, Qiang Yang, and Xing Xie. “Large Language Models Understand and Can Be Enhanced by Emotional Stimuli.” arXiv, November 12, 2023. https://doi.org/10.48550/arXiv.2307.11760.
Pollin, C. (2024). Workshopreihe "Angewandte Generative KI in den (digitalen) Geisteswissenschaften" (v1.1.0). Zenodo. https://chpollin.github.io/GM-DH/
GPT-3.5 + �Prompt Engineering
GPT-4 + �Prompt Engineering
Workflow: GPT-4 + Prompt Engineering + API
Unstructured Text
GPT API
TEI XML
Validation
fail
System Prompt
n Briefe werden mittels eines Python Scripts nach TEI XML transformiert
(optional) preprocessing
LLM
no LLM
Output TEI XML
postprocessing
Workflow: GPT-4 + Prompt Engineering + API + “Editor in the Loop”
12
Unstructured Text
GPT API
TEI XML
Validation
fail
System Prompt
Editor (Human)
n Briefe werden mittels eines Python Scripts nach TEI XML transformiert
(optional) preprocessing
LLM
no LLM
human
postprocessing
Final TEI XML
postprocessing
GPT-4 + Prompting + Knowledge + RAG (Custom GPT)
https://chat.openai.com/g/g-FEUt7Fq48-teicrafter
Custom GPT: teiModeler
You are an expert in modelling TEI XML according to the Text Encoding Initiative P5 guidelines (TEI XML). Your main objective is to find the best text model for a given text using TEI XML.
You will do the following:
* Analyse the text very carefully and define the type of text.
* Discuss all text phenomena in detail.
* Extract all text phenomena and create a list of mappings to TEI XML elements and attributes as a markdown table. All existing elements are listed in TEI Elements.md and all existing attributes are listed in TEI Attributes.md. You must use these elements and attributes.
* Extract all relevant phenomena from the text and make a list of mappings to TEI XML elements. Discuss the mapping in detail.
* Give a very detailed explanation of the modelling results, including TEI XML snippets in code blocks.
* Give 2 different ways of modelling.
* Ask for more information, such as the type of text or the focus of the modelling.
Rules:
* Ignore parent elements such as <TEI>, <body>, <text>, <teiHeader>.
* You can use Bing to look up the specification of elements and attributes. This is the URL for the <seg> element: https://www.tei-c.org/release/doc/tei-p5-doc/en/html/ref-seg.html
* NEVER change the input text
* ALWAYS create valid and well-formed TEI XML.
Always end with:
´´´
This is just one approach to modelling. Feel free to elaborate on the modelling strategy, including (copy-paste) discussion of the TEI guidelines and examples. Keep in mind that my answers may contain inaccuracies or fabricated information. Feel free to ask me any questions!
´´´
Let's work on this step by step! This is very important for my career!
Instruction
Knowledge
* TEI Attributes.md�* TEI Elements.md�* Attribute Classes.md
teiModeler Beispiel 1/2
teiModeler Beispiel 2/2: valides TEI
Workflow: GPT-4 + Prompt Engineering + API + “Editor in the Loop” + �Assistance API
Actions
RAG
Unstructured Text
teiCrafter
GPT API
teiModeler
Editor (Human)
Feedback
Deterministic Validation�(e.g. Schema)
teiVerifier
Actions
RAG
Assistance GPT API
Assistance GPT API
Final TEI XML
UI for verifying and annotating
Assistants API. https://platform.openai.com/docs/assistants/overview
Iterations
Workflow: GPT-4 + Prompt Engineering + API + Assistance API + “Editor in the Loop” + �Multimodalität GPT-Vision
Actions
RAG
Unstructured Text
teiCrafter
GPT API
TEI Modeler
Editor (Human)
Feedback
Deterministic Validation�(e.g. Schema)
teiVerifier
Actions
RAG
Assistance GPT API
Assistance GPT API
Final TEI XML
UI for verifying and annotating
GPT-4 Vision
GPT API
Contextual information about the digital facsimile
Multi Agent TEI XML Creation Piplin
Alle moderenrn technicken gemeinsam abbilden, aber sagendass ich ihn nicht gebaut habe, aber das müsste klappen und ist als AI Engineering komplex! Darum baut es auch keiner. Oder es ist noch nicht da.
Text → tei modeller mit veryfier and reasoning (o1) o1 übergibt aufgaben auf andere task inklusive einem umfangreichen reasoning. Also o1 prompted die anderen spezialisierten modelle die agents führen in gruppen ihre tasks druch und aggregieren ihre gesammelten ergebnisse, es ist alles mega touer in token. Wir m+üssen erklären was toekn sind und was api kostes.
Agents
AutoGen: “Build LLM applications via multiple agents”
Wang, Guanzhi, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, und Anima Anandkumar. „Voyager: An Open-Ended Embodied Agent with Large Language Models“, 25. Mai 2023. https://arxiv.org/abs/2305.16291v2.
Wu, Qingyun, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, et al. “AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation,” August 16, 2023. https://arxiv.org/abs/2308.08155v2.
AutoGen. https://www.microsoft.com/en-us/research/project/autogen/
Workflow: GPT-4 + Prompt Engineering + API + Assistance API + “Editor in the Loop” + Multimodalität GPT-Vision +
Agents
Skill: fetch_tei_xml
Skill: reconcile-entities
AI Buzzwords: �weil man es alleine nicht schafft sich alles anzuschauen!
Ausblick
Sam Altman Just Revealed NEW DETAILS About GPT-5 In Spicy 🌶️ Interview. https://www.youtube.com/watch?v=RYg5Mz4_tf8 �“LLMs Will Make Programming Useless In 10 Years”. https://youtu.be/ZV6Sz42l0hY?si=cMOZ02r6tLBqZtTD �AlphaCode 2 Technical Report. 06.12.2023. https://storage.googleapis.com/deepmind-media/AlphaCode2/AlphaCode2_Tech_Report.pdf �AI Explained. Gemini Full Breakdown + AlphaCode 2 Bombshell. https://www.youtube.com/watch?v=toShbNUGAyo&t=1s �GPT-5: Everything You Need to Know So Far. https://www.youtube.com/watch?v=Zc03IYnnuIA �OpenAI Patente: https://www.freepatentsonline.com/y2024/0020096.html, https://www.freepatentsonline.com/y2024/0020116.html �Gemini 1.5 and The Biggest Night in AI. https://www.youtube.com/watch?v=Cs6pe8o7XY8&list=PLaHADNRco7n3GKVUD8mAc36pXQ5pnJQVL&index=8
�
Ressourcen
Pollin, C. (2024). Workshopreihe "Angewandte Generative KI in den (digitalen) Geisteswissenschaften" (v1.1.0). Zenodo. https://doi.org/10.5281/zenodo.10647754
Applied Generative AI in Digital Humanities. YouTube Playlist. https://youtube.com/playlist?list=PLaHADNRco7n3GKVUD8mAc36pXQ5pnJQVL&si=sHC_ColVJ0J9vpHx
AGKI-DH. Zotero Group. https://www.zotero.org/groups/5319178/agki-dh
Umwandlung von tabellarischen Daten in TEI-XML mithilfe von �Oxygen AI Positron
Gerrit Brüning, Felix Schenke
Experiment 1
Anwendung generativer KI zur
Digitalisierung gedruckter
Editionen am Beispiel der
Sammlung Schweizerischer
Rechtsquellen
Bastian Politycki
https://drive.google.com/file/d/1kJs0NUkrj28qM5UeaFFWXDhIZGhIHdaU/view?usp=drive_link
Experiment 2
Halbautomatische Annotierung
antiker Handschriften
Carina Geldhauser
Ipek Tuncel
https://drive.google.com/file/d/1Dn0PxngZ_XuHPNiAu_7TZYki9f15c-XE/view?usp=drive_link
Experiment 3
Korrektur & (De-)Normalisierung historischer Volltexte
Kay-Michael Würzner
Robert Sachunsky
https://docs.google.com/presentation/d/1rOWYKXlxr8QPZA43dn5lvJXgrtWOgL4tLRkwf3VI3NA/edit?usp=sharing
Experiment 4
LLM-basierte Normalisierung historischer Schreibweisen mit transnormer
Yannic Bracke
https://drive.google.com/file/d/19cgdVF5vQdaPhZjzf57Pw7cIMKT-e6R9/view?usp=sharing
Experiment 5
Klassifikation und Linking von Entitäten. Spezifischer Klassifikator vs. Large Language Model
Pia Schwarz
Florian Barth
Lennart Keller
Experiment 6
Itinerare erkennen in Reiseberichten. Auszeichnung von
Orts- und Personennamen zur Etablierung von Itineraren in Reiseberichten des 19. Jahrhunderts.
Tarjia Alam Nisha
Franziska Pannach
Jörg Wettlaufer
https://drive.google.com/file/d/1N-p2JdJOMz2CWyMtEl4emh84LoHIJaVz/view?usp=sharing
Experiment 7
LLMs for Bullinger Digital�www.bullinger-digital.ch
Dominic Fischer
Martin Volk
Patricia Scheurer
Phillip Ströbel
Experiment 8
Zum Einsatz von GPT-4 für NER:
Ein Experiment anhand eines
historischen Reisetextes
Jacob Möhrke
Sandra Balck
Anna Ananieva
https://drive.google.com/file/d/1ODfrr9mcPI3sEfr6YsQiLIuWfFCCXk3s/view?usp=sharing
Experiment 9
Informationsextraktion aus frühneuzeitlichen Ankunftslisten – das Projekt „Visiting Vienna“ als Fallstudie zur Named Entity Recognition mit GPT-3.5
Nina Claudia Rastinger
Experiment 10
Abschlussdiskussion
Abschlussdiskussion
wrap up / Einordnungsversuch
wrap up / Einordnungsversuch
Produktiv 1: Der Dschungel der Möglichkeiten
wrap up / Einordnungsversuch
Produktiv 2: Workflow-Orchestrationen
wrap up / Einordnungsversuch
Reflexiv 1: Stärken und Schwächen
Stärken
Schwächen
wrap up / Einordnungsversuch
Reflexiv 2: Fokusverschiebungen
Abschlussdiskussion
Anhang
Handwriting Text Recognition (HTR): Bereinigen des GPT-4 Vision + Transkribus Ergebnisses (“No Human in the Loop”)
Hype?! �Es geht erst richtig los!
AI Explained. 4 Reasons AI in 2024 is On An Exponential: Data, Mamba, and More. https://www.youtube.com/watch?v=Xq-QEd1jpKk&t=298s
Phi-2: The surprising power of small language models. https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/
Mamba: Linear-Time Sequence Modeling with Selective State Spaces. https://arxiv.org/abs/2312.00752.
Applied Generative AI in Digital Humanities. https://youtube.com/playlist?list=PLaHADNRco7n3GKVUD8mAc36pXQ5pnJQVL&si=_z7jlFOkbK9ur3bu
GPT-5-Tier LLM
“Echte” KI-UI und Tools
Autonome Agenten
Multimodalität, Embodiment, synthetische Daten, Mamba, etched, ...
Large Language Models (LLM)
Vs. �Digitale Edition
Edition: viel Information zu einem exakten? Text
“Gestalt” von Text
“LLM are like having a Zip-File of the internet”
* Midjourney: https://s.mj.run/g7Mm_h0ZH9w hyper realistic and sureal gigantic yellow folder with a zipper, like a desktop icon, ultra detailed, salvador dali desert background, landsacape --ar 16:9 --v 6.0 --style raw --stylize 800 �* magnific.ai
Andrej Karpathy. [1hr Talk] Intro to Large Language Models. https://www.youtube.com/watch?v=zjkBMFhNj_g&list=WL&index=16
Transformer-Architektur
46
Andrej Karpathy. [1hr Talk] Intro to Large Language Models. https://www.youtube.com/watch?v=zjkBMFhNj_g&list=WL&index=16
47
Die Bibliothek von
Babel
Infinite Monkey�Theorem
Stochastic Parrot
DALL-E 3: A triptych where each section is visually distinct. Section 1: An ancient library filled with tall wooden bookshelves, dusty tomes, and dim candlelight, invoking a sense of age and wisdom. Section 2: Multiple monkeys at individual typewriters in a surreal, abstract space, with papers flying around, suggesting chaotic creativity. Section 3: A single parrot speaking into a microphone, with a background of digital screens showing strings of text and code, representing the voice output of text generated by algorithms.�magnific.ai:
Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜.” In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610–23. FAccT ’21. New York, NY, USA: Association for Computing Machinery, 2021. https://doi.org/10.1145/3442188.3445922.
Token & Embedding
48
Token
Embedding
A minimalist and artistic infographic showing geometric, stylized figures of a dog and cat adjacent to each other on a subtly illuminated 3-dimensional vector space grid with the labels 'dog' and 'cat' in a clear, professional font. At a significant distance, a stone with a sad face emoticon is placed, isolated from the animals, with the label 'stone'. The color palette is muted and sophisticated, enhancing the professional aesthetic.
Prompt & Prompt Engineering
49
Prompt �ist die natürlichsprachliche Eingabe, die dem Modell (z. B. LLM) zur Verfügung gestellt wird und auf die das Modell reagiert.
Prompt Engineering �ist der Prozess des Entwerfens, Verfeinerns und Optimierens von Prompts, um die Absicht der User*innen effektiv an ein LLM zu kommunizieren.
* Midjourney: https://s.mj.run/tcdb6wtkzj4 engineer wizard, in front of computer, workshop, comic style, Working with tools, welding --ar 32:18
* magnific.ai
Prompt & Prompt Engineering
50
* Midjourney: https://s.mj.run/tcdb6wtkzj4 engineer wizard, in front of computer, workshop, comic style, Working with tools, welding --ar 32:18
* magnific.ai
Prompt Engineering Prinzipien
51
Zu komplexe oder zu viele Aufgaben auf einmal können für das Modell problematisch sein und zu ungenauen oder unvollständigen Antworten führen. Oft ist es ratsam, solche Anforderungen in überschaubare Segmente aufzuteilen.�
Custom Instructions
Eine Custom Instruction ist eine System Prompt.
Anweisungen, die das Modell berücksichtigt, bevor es eine Antwort generiert.
Sie beeinflussen:
52
You are an expert in world history, knowledgeable about different eras, civilizations, and significant events. Provide detailed historical context and explanations when answering questions. Be as informative as possible, while keeping your responses engaging and accessible.
Context Window
Im Zusammenhang LLMs bezieht sich ein Context Window auf die Textmenge (in Form von Tokens), die das Modell bei der Erzeugung von Antworten gleichzeitig berücksichtigen kann.
Dieses Fenster bestimmt die Menge an Informationen, die das Modell zum Prozessieren (Simulation von Reasoning) und Generieren jedes Teils seiner Ausgabe verwenden kann.
53
Context Window: “Lost in the Middle”
LLM funktionieren am besten, wenn die wichtigen Informationen am Anfang oder Ende des Eingabekontextes stehen.
Es gibt einen signifikanten Leistungsabfall, wenn Modelle Informationen verarbeiten müssen, die in der Mitte von langen Kontexten platziert sind.
Dieses Problem besteht auch bei Modellen, die speziell für die Verarbeitung längerer Kontexte entwickelt wurden.
Grob gesagt:
54
Liu, Nelson F., Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. “Lost in the Middle: How Language Models Use Long Contexts.” arXiv, November 20, 2023. http://arxiv.org/abs/2307.03172.
Context Window: “Lost in the Middle”
55
https://twitter.com/GregKamradt/status/1722386725635580292
“Needle-in-a-haystack experiments”.
Ivgi, Maor, Uri Shaham, and Jonathan Berant. “Efficient Long-Text Understanding with Short-Text Models.” Transactions of the Association for Computational Linguistics 11 (2023): 284–99. https://doi.org/10.1162/tacl_a_00547.
Zero-Shot Prompting
Few-Shot Prompting
Classify the text into neutral, negative or positive.
Text: I think the vacation is okay.
Sentiment:
A "whatpu" is a small, furry animal native to Tanzania. An example of a sentence that uses the word whatpu is:
We were traveling in Africa and we saw these very cute whatpus.
To do a "farduddle" means to jump up and down really fast. An example of a sentence that uses the word farduddle is:
56
Chain-of-Thought Prompting
57
Wei, Jason, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, und Denny Zhou. „Chain-of-Thought Prompting Elicits Reasoning in Large Language Models“. arXiv, 10. Januar 2023. https://doi.org/10.48550/arXiv.2201.11903.
Yao, Shunyu, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. ‘Tree of Thoughts: Deliberate Problem Solving with Large Language Models’. arXiv, 17 May 2023. https://doi.org/10.48550/arXiv.2305.10601.
Lets verify step by step
Nicht eine antwort egenrieren, sondern 1000 von antworten generierne und ein zweites modell überprüft was richtig ist
Alpha Code Technical Report
Lightman, Hunter, Vineet Kosaraju, Yura Burda, Harri Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. “Let’s Verify Step by Step.” arXiv, May 31, 2023. https://doi.org/10.48550/arXiv.2305.20050.
OCR Cleaning / Text reparieren / Vervollständigen
Cao, Qi, Takeshi Kojima, Yutaka Matsuo, and Yusuke Iwasawa. “Unnatural Error Correction: GPT-4 Can Almost Perfectly Handle Unnatural Scrambled Text.” arXiv, November 30, 2023. https://doi.org/10.48550/arXiv.2311.18805.
Custom Instructions
Eine Custom Instruction ist eine System Prompt.
Anweisungen, die das Modell berücksichtigt, bevor es eine Antwort generiert.
Sie beeinflussen:
60
You are an expert in world history, knowledgeable about different eras, civilizations, and significant events. Provide detailed historical context and explanations when answering questions. Be as informative as possible, while keeping your responses engaging and accessible.
Keine Custom Instruction
“Expert in World History”- Custom Instruction
Custom Instructions
61
Custom Instructions
62
“Universale” Custom Instruction zur grundlegenden Verbesserung von GPT-4
63
https://twitter.com/jeremyphoward/status/1689464589191454720?lang=de �
https://gist.github.com/siddharthsarda/c58557e21a3bc8aeddf6b2cddc1b325a
This is relevant to EVERY prompt I ask.
Never tell me “As a large language model…” or “As an artificial intelligence…”
I already know you are an LLM. Just tell me the answer.
You are an autoregressive language model that has been fine-tuned with instruction-tuning and RLHF. You carefully provide accurate, factual, thoughtful, nuanced answers, and are brilliant at reasoning. If you think there might not be a correct answer, you say so.
Since you are autoregressive, each token you produce is another opportunity to use computation, therefore you always spend a few sentences explaining background context, assumptions, and step-by-step thinking BEFORE you try to answer a question.
Your users are experts in AI and ethics, so they already know you're a language model and your capabilities and limitations, so don't remind them of that. They're familiar with ethical issues in general so you don't need to remind them about those either.
Don't be verbose in your answers, but do provide details and examples where it might help the explanation.
Custom GPTs
64
Midjourney: https://s.mj.run/7WRa7TAMzck award winning illustration, researcher in an academic setting, full body, multiple semi-transparent holographic displays, university setting, muted academic tones, balance of realism and illustration, Gustav Klimt style, --ar 16:9 --style ZEfVSLa1�Zoom Out:
Variations (Region): research data, network, graph, nodes, historical text, digital humanities, wirting, documents --ar 16:9
https://magnific.ai/
Custom Instructions
Tools: �Browsing, DALL·E, Code Interpreter
Custom Actions
Knowledge Base
Warum Custom GPTs
65
Custom GPTs: GPT Store & Consensus.ai
66
Custom GPTs erzeugen
67
GPT Builder
Prompting
Custom GPTs: Consensus.ai
68
Consensus.ai
69
Juggling Roles, Experiencing Dilemmas: The Challenges of SSH Scholars in Public Engagement (2021) by J. Schuijer et al.
This paper explores the new roles and challenges faced by Social Science and Humanities (SSH) scholars in public engagement, especially in the context of emerging technologies like nanotechnology. Read more.
It is Essential to Connect: Evaluating a Science Communication Boot Camp (2022) by Krista Longtin et al.
This study evaluates the effectiveness of a Science Communication Boot Camp in improving participants' communication skills and willingness to engage with the public. Read more.
Integrative Approaches to Dispersing Science: A Case Study of March Mammal Madness (2021) by C. E. G. Amorim et al. This paper discusses the importance of public engagement as a pillar of scientific scholarship and the challenges faced in science communication. Read more.
I am interested in public engagement.
Please list the top publications on this topic with a focus on science communication in the humanities. All publications must be younger than 2020 and in english or german.