To transcribe audio and video to text

• Amberscript offers a transcription service for audio files and subtitles videos in multiple languages (including Swedish). The service is web-based and offers both machine and manual transcription and subtitling. Transcribe audio and video to text.

To find relevant articles based on a text or topic

• Keenious is a search tool that utilizes AI to analyze the text in your document and suggest relevant research articles based on the topic.

AI chatbots — tools that chat and "talk" to the users

• ChatGPT 3.5 (free). Conversational bot.
• ChatGPT 4 (paid). Advanced conversational bot.
• EditGPT. Helps edit text.
• CoPilot. Similar to ChatGPT 4 (free).
• Gemini. Conversational bot that analyzes text, audio, and images.
• Claude. Conversational bot that analyzes text and pdf files.

AI tools to create images

• CoPilot. Conversational bot that generates images and code.
• DALL-E. Generates images based on instructions.
• Midjourney. Generates images, requires a paid license.

To summarize text in a PDF document

• CoPilot (Microsoft Edge sidebar)

To generate code

• GitHub CoPilot. Generates and suggests code.

To translate and improve text

• DeePL. Translates between languages and improves English and German text.
• Grammarly. Improves grammar in texts.
• Claude. Conversational bot that translates in multiple languages.

To search and summarize information from articles and other texts

• Perplexity. Summarizes results from articles.
• Elicit. Searches and summarizes text based on example articles.
• Research Rabbit. Works similar to Elicit.

Date of text: 2024-04-15

Amberscript
Amberscript Speech-to-text service

transcription, text-to-speech, subtitles

What is it?

Amberscript offers a transcription service for audio files and subtitles videos in multiple languages (including Swedish). The service is web-based and offers both machine and manual transcription and subtitling.

Amberscript provides a free trial and then costs SEK 1.50/minute.

Potential uses

• Creating video content that is fully accessible to people with hearing impairments
• Interview transcriptions 

Key features

• GDPR-compliant, company. The service is delivered by Amberscript BV (Dutch company) and SUNET is a contracting party. The user is solely responsible for providing the content to be transcribed. After the transcription is complete and delivered to the web interface, Amberscript does not retain any information. The user must also remove the source material and delivery from the web interface. It is important to note that only the owner can view the files; neither team admins nor central admins have access to users' material. The agreement for a personal data assistant has been made with the Amberscript supplier, also stating that personal data will not be transferred to countries outside the EU/EEA. The data is encrypted both during transmission between the user and Amberscript, as well as in storage. Amberscript helpdesk on privacy and security
• SU_Medarbetarwebben_it-tjänster_tal_till_text

Tips & tricks
Amberscript might be less inaccurate in other languages than English and also make mistakes when multiple accents have been recorded. A thorough and careful examination of the text is strongly recommended.

DEMO
This short demo shows how to upload your audio file and begin your project
This short demo shows how to create and edit your project 

Date of text: 2024-04-15

search, relevant literature

Keenious

What is it?

Keenious is a search tool that utilizes AI to analyze the text in your document and suggest relevant research articles based on the topic. Finds relevant articles based on provided text. Keenious is provided by Stockholm University Library to SU students and staff.

Potential uses

• Find relevant literature on a topic of interest
• Identify related research 

Key features

• Keenious allows focusing on a document or even on specific parts of a document to search for relevant literature. 
• There is an option to add several filters for your search, such as keywords
• Literature and references management 
• Stockholm University Library Services

DEMO

Keenious official product video

ChatGPT—3.5 (free version)

Date of text: 2024-04-24

ChatGPT—3.5

What is it?

text generation, code generation

ChatGPT is an AI-powered conversational chatbot developed by OpenAI, designed to process and generate human-like text. ChatGPT is also capable of writing and debugging code. The paid version (GPT-4) includes additional features.

Key Features

• ChatGPT is a very flexible tool that can be customized to assist the user in a wide range of tasks such as rephrasing texts, generating summaries and outlines, and translations. 
• The quality of the outputs is often related to the quality of the instructions the user provides to the tool. 
• ChatGPT, like other AI-powered chatbots, is prone to errors (mostly subtle), especially when it comes to accuracy, and providing correct references. However, because of the tools' convincing writing style, an uncritical user is likely to fall for the errors.

ChatGPT-4 (paid version)

Date of text: 2024-04-24

ChatGPT—4

What is it?

text generation, code generation, image generation, data analysis

This version of ChatGPT-4 is larger and more powerful than GPT-3.5 and is trained on a broader dataset, which includes more diverse data up until its training cut-off in 2023. Unlike GPT-3.5, which operates solely on text, GPT-4 has multimodal capabilities and can process and generate both text and images, enabling more diverse applications.

Key Features

• If the user asks for an image,  ChatGPT-4 can generate images using the text-to-image tool DALL-E 3.
• The user can upload data and get help analyzing it. Specifically, ChatGPT-4 can generate Python code, execute it, and output the results, allowing the user to access the code. 
• The user can create their own personal "GPTs" that are tailored to interact with the user in a specific way. It is also possible to use other users' GPTs through the "GPT Store." Since the GPTs can be connected to external tools through APIs, there are many user-created GPTs in the store. For example, Andreas Jemstedt has created a GPT that aids in creating multiple-choice questions with detailed feedback for each response alternative. The GPT can be accessed for anyone with a paid ChatGPT account.
• If you use the ChatGPT mobile app, you can also have a verbal chat with It. In this mode, you speak, and ChatGPT replies using text-to-speech technology.

EditGPT (browser plugin)

Date of text: 2024-04-24

What is it?

editing text; grammar check; proofread

The EditGPT plug-in is an additional feature designed for the ChatGPT models, which allows users to request specific edits to previously generated responses or texts. 

EditGPT

Key Features

• EditGPT allows users to specify how they want the text to be changed, mentioning specific edits or giving general directions like "make it more formal" or "shorten it."
• The plugin activates when you are on the ChatGPT webpage. Once you submit a message, a button appears beside the text saying ChatGPT. Once you click it it shows you all the differences between the message you submitted and what ChatGPT responds to. For example, if you submit a text you want to proofread, it will mark all changes from the submitted text and ChatGPT’s version of the text. See the image for example:

Our own comments 

Andreas Jemstedt: If you want to use a chatbot to generate something that needs to follow complex instructions, ChatGPT-3.5 will probably fail, while ChatGPT-4 and Google Gemini do quite well. Both the free and paid versions of ChatGPT require that the user either use it for purposes where errors are not a big problem, or where they are able to validate the answers themselves.
Linda Barman: Remember to never post student-written texts or material into ChatGPT.

Similar to ChatGPT 4, free.
Date of text: 2024-04-24

Microsoft Copilot

What is it?

text generation, code generation, image generation, data analysis

Microsoft Copilot is an AI-powered conversational chatbot that relies on ChatGPT-4 and DALL-E 3, combined with Microsoft Bing search engine and text-to-speech technology. Similar to ChatGPT and Google Gemini, Copilot uses anonymized conversations for training its AI model, which implies that any input provided is utilized for its training process—unless you possess an organizational account. Hence, it's advisable to refrain from sharing sensitive information during interactions. Microsoft Copilot has also been integrated into various Microsoft tools in Microsoft 365 , while there is also a Copilot app  as well as a Copilot Pro version, but these are not discussed here. However, the following describes a feature of the built-in version of Copilot in the Microsoft Edge https://www.microsoft.com/edge/ web browser.

Key features

• Because it is based on ChatGPT-4, it shares many of its features, including the ability to create images with DALL-E 3 and search the web. However, the user cannot upload files, but there are workarounds for this (see below). 
• Like other chatbots, it is prone to subtle bugs that can be difficult to detect.
• Copilot allows voice input and spoken responses alongside text interaction.
• Copilot offers the choice of response style—More Creative, More Balanced, or More Precise—when generating output.
• Source links can be provided in relation to the output that Copilot generates -based on Bing search.
• Copilot offers image generation without requiring a paid Pro account.

Our own comments

Andreas Jemstedt: I only use Microsoft Copilot outside of workshops. For my chatbot needs, I prefer Google Gemini or ChatGPT-4. However, it's a great tool if you want a powerful AI chatbot that you don't have to pay for. A useful function of the built-in version of Copilot in Microsoft Edge. In Microsoft Edge, a web browser developed by Microsoft, Copilot is available as a "sidebar" that can be accessed by clicking the blue swirly-looking icon in the upper right corner. It opens the panel, and you can use the Copilot chatbot as described above. However, in this mode, the chatbot can also access the webpage you currently have open. Because Microsoft Edge can open and read PDFs, this version of Copilot can access your PDFs. However, it may sometimes be reluctant to actually read the document and may need very clear instructions to do so.

1. A version of Copilot can be seamlessly integrated into Microsoft Office 365, such as Word, PowerPoint, Excel and Teams.
2. In the Copilot mobile app, users can opt for either GPT-3.5 (faster) or GPT-4 (more accurate), with Copilot Pro subscribers having access to GPT-4 Turbo for even faster and more accurate responses.

Date of text: 2024-04-24

Gemini

What is it?

text generation; code generation; image generation; data analysis

Conversational bot that analyzes text, audio, and images.

Google Gemini is an AI-powered conversational chatbot that is built on a family of large language models created by Google DeepMind, announced in December 2023 and it’s considered the successor to prior Google AI projects like LaMDA and PaLM 2. Gemini allows for using combinations of multimodal inputs and outputs such as text, images, audio, video, and code. Similar to ChatGPT and Copilot, Gemini uses anonymized conversations for training its AI model, which implies that any input provided is utilized for its training process—unless you possess an organizational account.

Key features

• Gemini can process and generate content across different formats like text, code, audio, images, and video
• Developers can integrate Gemini models into their applications through Google AI Studio and Google Cloud Vertex AI to enhance their capabilities.
• Like other chatbots, it is prone to subtle bugs that can be difficult to detect.
• Gemini allows voice input and spoken responses alongside text interaction.
• Gemini offers three versions of output that the users can access and choose from.
• Source links can be provided in relation to the output that Gemini generates - based on Google search.

Our own comments

Alexandra Farazouli: It seems that this tool works very well also with translations in English and Swedish.

Date of text: 2024-09-20

Claude

What is it?

text generation; code generation

Claude is an AI-powered conversational chatbot created by Anthropic, designed to help with a wide variety of tasks including analysis, answering questions, math, coding, creative writing, and general discussion.

Key Features

Users can define a set of tools for Claude to use and specify a request. The model will then decide which tool is required to achieve the task and execute an action on their behalf, such as:

• Using a calculator for complex numerical reasoning
• Translating natural language requests into structured API calls
• Answering questions by searching databases or using a web search API
• Taking simple actions in software via private APIs
• Connecting to product datasets to make recommendations and help users complete purchases

Our comments

Alexandra Farazouli: It seems that this tool is very good at reasoning, engages the user to collaborate when answering prompts, and is quite accurate in its responses. This tool has no access to the internet.   

DeePL (free version)
Description:
Date of text: 2024-04-10

DeepL translation
DeepL write
DeepL also has a Pro version with more features, which we are not reviewing here.

proofreading; translation

What is it?

Translates between languages and improves English and German text. DeepL is a tool that started out as a translation service between many different languages. However, it can now also improve texts written in English or German and adjust their style, for example, by making them sound more academic or casual.

Key features

For both the translation and text enhancement, the user can select alternative words or phrases in the interface by clicking on words in the output. 
In DeepL write, all changes are highlighted, making it easy for the user to see what changes have been made and decide what to keep. 
While AI chatbots like ChatGPT or Google Gemini tend to make large changes to texts (sometimes even when they are instructed not to), DeepL mainly adjusts the text in terms of grammar, word choice, and small changes in structure.
DeepL is accessible as an add-on in Microsoft Word through its Add-ins feature. However, in our experience, it doesn't work as well as the web interface.

Our comments:

Jemstedt: DeepL is one of the main tools I use when I work with text. I love that it only improves grammar and word choices without messing with the message or structure of my texts.
A colleague at Stockholm University who works with translation told me that DeepL is a great translator, much better than, for example, Google Translate.

DEMO video Date of video: 2024-04-10

Date of text: 2024-04-24
Improves grammar in texts.

Grammarly 

proofreading, grammarcheck

What is it?

Grammarly is a writing assistant tool that helps improve users’ writing by checking for grammar mistakes, and typos, and offering suggestions for clarity, style, and conciseness

Potential uses

Proofread essays, reports, emails, and other written documents.

• Key features
• Grammar and spelling check
• Vocabulary suggestions
• Tone and style adjustments
• Clarity and conciseness checks
• Grammarly is accessible as an add-on in Microsoft Word and Browser through its Add-ins feature.

Tips & tricks

• Text improvers often generate content that although may read nicely, can include confusing words that do not make sense, therefore proofreading and critically reading their output is very important.
• Fun fact: The use of multiple text-improving tools shows that often they do not come in agreement with each other suggesting very different phrasings.

Date of text: 2024-09-20
Claude

text generation, code generation

What is it?

Conversational bot that translates in multiple languages.

Claude is an AI-powered conversational chatbot created by Anthropic, designed to help with a wide variety of tasks including analysis, answering questions, math, coding, creative writing, and general discussion.

Key Features

Users can define a set of tools for Claude to use and specify a request. The model will then decide which tool is required to achieve the task and execute an action on their behalf, such as:

• Using a calculator for complex numerical reasoning
• Translating natural language requests into structured API calls
• Answering questions by searching databases or using a web search API
• Taking simple actions in software via private APIs
• Connecting to product datasets to make recommendations and help users complete purchases

Our comment

Alexandra Farazouli: It seems that this tool is very good at reasoning, engages the user to collaborate when answering prompts, and is quite accurate in its responses. This tool has no access to the internet.

Date of text: 2024-04-24

GitHub Copilot

Guide about using it in Visual Studio Code

Getting free access as a university student or faculty 

code

What is it?

Github Copilot is an AI-powered code completion tool that is part programming chatbot and part integrated coding assistant. It can debug code, give code suggestions, add comments to code, and generate code based on natural language (e.g. "create a function that adds two numbers"). It works as an add-on to the following code editors: Visual Studio Code, Visual Studio, Vim, Neovim, and JetBrains. However, we haven’t extensively used all the aforementioned tools and the information here is based on our experience with Visual Studio Code.

Note that it will take you longer to get started with Github Copilot than the other tools in this guide because it requires a lot of setup.

Read how to get started with GitHub Copilot

Key features

• A panel where you can interact with a programming chatbot. For example, you can ask it to add features to your code, debug your code, or generate code from scratch.
• It can also assist the user in discovering new libraries or functions by accepting Copilot's suggestions and exploring their documentation. 
• As you code in the editor, it makes multiple suggestions, which appear as greyed-out code that you can accept by pressing the “Tab” key (default). 
• You can select your code and right-click to see many options, such as "Explain this" and "Fix this”.
• It can process comments and docstrings, allowing the users to describe their desired functionality and get code suggestions based on that description.
• When you select code and type in the chat panel, the chatbot responds as if the selected code was part of your instruction/question.
• Has access to all your open tabs in the code editor, so it can make accurate suggestions for variable names, functions, and more based on your open project.

Our commentsJemstedt: I started using Github Copilot when I was doing a lot of statistics. At that time, I was mainly using R, which I know a lot of other researchers use. Therefore, here is a guide on how to start using R in Visual Studio Code (where you can use Github Copilot). I find the "Explain this" feature really useful for learning about coding. I have used it many times to understand what copied or generated code does and how it works. It's a great learning tool.

Generates images based on instructions.

Date of text: 2024-04-24

DALL-E 3 (accessed through either ChatGPT-4 or Microsoft Copilot)

text-to-image

What is it?

DALL-E 3 is a text-to-image model that creates images based on the user's instructions. It can be accessed as a stand-alone tool. However, we have only accessed it through ChatGPT-4 and Microsoft Copilot.

Key features

• It has been trained on a large database of text and image pairs from open data sets and collected from the open internet, and then recreates the patterns in these data in line with the words used by the user. For example, if the instructions use the word “photo”, the output will look more like a photo, if the user writes “in the style of Van Gogh” it will look more like a painting by Van Gogh.
• If accessed through ChatGPT-4, the user’s instruction to DALL-E 3 will be altered before being sent to the tool. For example, when writing: “Generate a painting of a university professor on a sunny beach in the style of Joaquin Sorolla,” ChatGPT-4 might “improve” the prompt by sending “A painting of a university professor standing on a sunny beach. The scene captures the soft, vibrant light and the gentle movement of the sea typical of early 20th-century Spanish impressionism, reminiscent of Joaquin Sorolla's style. The professor, a middle-aged Caucasian male with short gray hair, wears a casual light blue shirt and khaki trousers. He holds a book in one hand and looks thoughtfully at the horizon, surrounded by the shimmering play of light on the water and the soft sand under his feet.”
• In ChatGPT-4, the user can regenerate parts of the image with a selection tool after the image has been generated.

Our comment

Andreas Jemstedt: I mainly use DALL-E 3 in my workshops because participants can access it for free through Microsoft Copilot. When I create images for my own work, I use Midjourney.

Date of text: 2024-04-19

Midjourney (no free licenses available)

text-to-image, generates images, requires a paid license

What is it?

Midjourney is a text-to-image model that creates images based on the user's instructions. It can be accessed as a stand-alone tool in the browser or through the chat/discussion board/communication software Discord. Midjourney does not have a free license, and the cheapest monthly plan is around $10.

Key features

Midjourney offers several features that are not available in, for example, DALL-E 3. For example, the user can upload images that can influence the generated images, it can create images with many different aspect ratios and not just squares, and there are several features for altering generated images in different ways.

Our comments

Andreas Jemstedt: I almost only use Midjourney when creating images, and this is because of the ability to create images with different aspect ratios. I mainly use my images in my presentations, so it is very handy to be able to create images that are wider than they are tall (e.g. 3:1) or vice versa.

Tips & tricks

• AI image generators often reproduce bias and offensive content in their output in sometimes less obvious ways.
• AI image generators’ output needs to be always fact-checked since they often generate inaccurate content. 
• AI image generators often generate content that although may seem fine, can include falsy depictions of reality (especially in details such as human fingers for example) that might include mistakes, therefore critically checking their output is very important.

Date of text: 2024-04-16

Perplexity (free version)

search

Summarizes results from articles.

What is it?

Perplexity (free version) is a Chatbot like Copilot or ChatGPT, but it primarily focuses on finding and summarising information. As such, it aims to be an alternative to Google Search rather than ChatGPT.

Key features

• Very quickly summarises web content such as homepages or research articles and presents the summary as an answer to the question you posed in the chat window.
• Can be “focused”, to make it mainly rely on academic sources to answer your questions.
• Links to the sources it uses.
• Perplexity allows users to organize their research threads, files, and data in collections, making it easier to manage projects or topics. 
• Perplexity offers the option to invite contributors and adjust privacy settings as needed.
• The free version gives you 5 “pro searches” each day, but there is also a useful free version. The company states that the pro searches are more capable, but we have not assessed the difference ourselves.

Our comment

Andreas Jemstedt: I find the answers to questions are generally of high quality and more reliable than when asking, for example, ChatGPT-4 to answer a question. When I use it, it replaces Google and I mainly use it to get an idea about a topic I have questions about, and then I go to the sources it mentions. I find it useful to get an overview of a field and find sources to start digging in.

Tips!

• AI chatbots might reproduce bias and offensive content in their output in sometimes less obvious ways.
• AI chatbots’ output needs to be always fact-checked since they often generate inaccurate content. 
• AI chatbots often generate content that although may read nicely, can include confusing words that do not make sense, therefore proofreading and critically reading their output is very important. 

Date of text: 2024-04-24

Elicit

search, relevant literature

What is it?

Elicit is an AI-powered search tool designed to assist users in finding relevant research articles based on the content of their documents. It analyses a prompted research topic through a document, that a user selects, and suggests relevant articles.

Potential uses

• Find relevant literature on a topic of interest
• Identify related research

Key features

• Elicit allows focusing on a document or even on specific parts of a document to search for relevant literature
• There is an option to add several filters for your search, such as keywords
• Literature and references management 
• Summarise articles

DEMO: Access the official presentation of the tool by Elicit cofounders

Date of text: 2024-04-24

Research Rabbit

search, relevant literature

What is it?

Research Rabbit is an AI-powered literature mapping tool that takes one or more initial (seed) papers and employs various methods (primarily based on citations) to suggest additional similar papers.

Potential uses

• Visualise networks of literature
• Identify related research 
• Explore influential authors in your field

Key features

• Research Rabbit allows users to create collections of research papers and by adding papers to such collections, ResearchRabbit learns users’ preferences and improves its recommendation. 
• Users can collaborate on collections with others.
• Literature and references management 

How to get the results you want when interacting with chatbots such as ChatGPT, Microsoft Copilot, or Google Gemini?

The instructions a user gives to a chatbot are sometimes called a “prompt,” and when the user designs their prompt to increase the quality of the chatbot's response, it is sometimes called “prompt engineering.” The OpenAI Promt engineering guide from the company OpenAI (that created ChatGPT and DALL-E), gives many useful tips on how you can improve your prompts to achieve your sought outputs.

OpenAI Promt engineering guide

A prompt library

Drs. Ethan and Lilach Mollick, faculty at the Warton School at the University of Pennsylvania, have compiled an open library of prompts. Many of them are for educational purposes. The prompts can be used as they are or as inspiration for your own prompts. You can access the prompts for free.

Prompt Library — More Useful Things: AI Resources

Explore!

What is out there? Ongoing research, news (reports from UNESCO, UNICEF AND EC). 

Elements of AI — free online course about the bascis of AI

Text: Andreas Jemstedt