<

Scientific papers in plain clothes

by Mădălina
If you follow me, you may know that I've been paying close attention to artificial intelligence for the past two and a half years. No, this isn't yet a post about ChatGPT, but a post about what we can learn from AI researchers about how they present their latest scientific papers.

Shortcut from research to practice

Generative artificial intelligence is a hot topic and a lot of people are paying attention to it, whether they are simple AI users or builders of systems that use AI. This means that a very interesting phenomenon is happening in the field: the path from scientific work to practical application is very short. The transition from research to a new application is on the order of days. For example, most prompting techniques that many AI “gurus” promote first appeared in scientific papers. The concept of “anchoring” an AI chatbot’s responses to reality through a technique called RAG was first a scientific articleThe prompting technique in which the AI is trained to first make a response plan, the one that philosophically underlies today's powerful GPT o1 and Deepseek r1 models, was described - you guessed it - in a scientific paperA few days after its publication, there were hundreds of developers, meaning programmers, meaning not researchers, who were already trying to put into practice what they had read.
In other words, the community that uses the research "keeps an eye on" the community that does the research. And that creates miracles in science communication.

Scientific works dressed "in civilian clothes"

Large companies and universities quickly understood that they had to present research in two ways: once in a journal, then “in plain sight,” for the community. The most important scientific papers now have either their own webpage, or a dedicated blog post, or a detailed Twitter presentation, or a YouTube discussion. The presentation of a paper began to borrow from the characteristics of a new product launch.

What non-scientific presentations look like

Non-scientific presentations of scientific papers additionally have:
  • A language that is easier to understand by the community (not necessarily explained for a child, but explains as for a person interested in AI)
  • A different hierarchy of information: start with what is most interesting and then move step-by-step through the process
  • Pleasant visual appearance, with graphics and animations.

Examples

 

 

 

Lucrarea OpenAI despre generare de video
OpenAI: developing a model for video generation
Mapping the mind of AI

Anthropic, about mapping the brain of an AI

 

 

 

Stanford researchers, about a model that explores the integration of technology and artificial intelligence in the analysis and interpretation of classical musical compositions, focusing on Für Elise by Ludwig van Beethoven.

 

 

Stable Diffusion 3 Science Paper

Stability AI, announcing the scientific paper detailing how the text-based image generation AI model, Stable Diffusion 3, works

 

And other examples:

Anthropic, a YouTube discussion with researchers launched alongside the scientific paper on how AIs sometimes "pretend" to follow instructions

Coherence, a Twitter thread (X) where they develop the ideas from the latest scientific paper

 

Computer science already had this good habit

 

Computer science researchers and practitioners already have the prerequisites to communicate better with each other. And this is because there are very large communities that include both "camps": those who are creators of science and those who use science in software code.

 

Computer science research often appears on GitHub, and the one in the field of AI and Machine Learning also appears on Hugging Face, a site for AI practitioners that also has a special section for scientific papers.

 

But we are not computer scientists.

 

Okay, not all researchers work in a field as highly sought after and promoted as artificial intelligence is now. Furthermore, not all fields have such a strong connection between researchers and industry.

However, the principles we see here are valid for everyone:

  1. For most fields where something is being researched, there are communities of people who are interested in putting that research into practice: in your case, where are those communities? How do you tell them about your science?
  2. For most published scientific papers, there are ordinary people who would like to know more about those topics: how can science be presented in a way that is understandable to them? What images can I show so that they can understand? How can I structure the information to make it clearer?
  3. for every scientific paper, there should also be a "plainclothes" paper, as I said above: that is, a text with more patient and empathetic explanations, whether it's on Twitter, a YouTube discussion, or a Facebook post

 

 

"We are pleased to announce that..."

 

I have noticed that researchers in Romania also feel the need to announce when they publish a new paper, either on Facebook or LinkedIn, the social networks that are more popular in our country. Usually, the announcement goes something like this: “I am pleased to announce that the paper with the title X has appeared in journal Y” #work #research #science #proud #SLBTNSR (i.e. the name of the laboratory or institution or another acronym that followers don’t understand anyway).

 

It's great that this is happening. What I think would be even better: spend five extra minutes reflecting on the message and trying to explain the relevance of the study and why it might be interesting to others. And write that down. An extra sentence or two, that's all. If you feel like you don't have the words, ask ChatGPT. Seriously. The biggest risk with ChatGPT is that it "hallucinates," meaning it makes up information. But you'll know that and you'll know how to correct it. But think about those five extra minutes about the significance of the research you're just presenting and you'll see how much the reactions you get will change.

 

If I have not persuaded you, I will make one more argument: do you know where the greatest interest is in effective communication? It's where the money is. So, when I see how much effort the AI community is putting into communicating effectively, I am suddenly interested in learning how it does it.

CategoriesAI

Leave a Reply

Your email address will not be published. Required fields are marked *