With the FWF Annual Report, we not only offer you insights into the past funding year, but also invite you to look back on some special moments in research. Read it to find out more about key funding metrics and statistics. In total, the FWF invested around €340 million in its funding programs in 2025.
Knowledge is not created to order. It grows slowly, often in a previously unknown direction, sometimes over generations – and this is precisely where its strength lies. Basic research does not follow a strict timetable, but the logic of knowledge: asking questions, patiently and persistently.
The researchers funded by the FWF at Austria’s universities and non-university research institutions have been asking questions this way for decades. So much rests upon the foundations created by their findings: Tomorrow’s medications are developed from today’s basic biochemical knowledge. The technologies that will drive our economy forward are often still theoretical models or lab experiments today. And the social institutions that support our democracies would be unthinkable without the contributions of the social sciences and humanities. Last but not least, arts-based research opens up knowledge spaces that neither pure science nor pure art alone can occupy, fundamentally expanding our understanding of the human condition.
The past year has once again shown us how crucial a vibrant, independent scientific culture is for a democratic society. At a time when simple answers are louder than complex truths, research that is committed to complexity is more socially relevant than ever. Science without freedom is not science – it is at best a confirmation of preconceived opinions. The FWF works with the entire scientific community to help ensure this freedom: through competitively awarded funding, the international peer review process, and by trusting excellent researchers to know which questions are worth asking.
How are Austria’s researchers doing? We took this question seriously and asked more than 3,300 scientists as part of a survey conducted by Spectra. The results paint a differentiated but generally positive picture: Austria can hold its own as a science location with a high level of internationality and freedom of research, and two thirds of respondents reported being satisfied with their overall professional situation. That’s good to hear. At the same time, many researchers recognize a need for structural changes – particularly when it comes to reliable career prospects for young researchers, dismantling traditional hierarchies, and increasing the social recognition of research. Respondents rated FWF’s funding programs positively for the most part, but expressed criticism of the low approval rates for proposals and a percieved lack of transparency on grounds for rejection. You can find all the details of the survey here.
These results provide us with both confirmation and a mandate. They show that we are doing well in many areas, but also point out where there is room for improvement. Funding requirements are diverse, ranging from work in small, focused teams to large-scale research networks in the Clusters of Excellence, from targeted career support to science communication. Equal opportunities and strictly avoiding discrimination are as important to us as ensuring good scientific practice and making research ecologically sustainable. The FWF is committed to meeting all of these demands, and is aligning the further development of its funding portfolio accordingly. The most recent example is the introduction of the Specialized Research Groups, a funding option that provides flexible and tailor-made support for inter-institutional teams.
We would like to extend our thanks to the researchers who are making society better with their discoveries (which are also part of this Annual Report), the FWF boards and committees, especially the members of the Scientific Board, the international reviewers, and the universities and non-university research institutions for the successful cooperation over the past year. With 749 newly approved projects and a funding amount of €340 million, it was possible to maintain the growth of basic research in 2025.
We would also like to thank the political decision-makers and the public – because science is not an end in itself. It is a promise to make the world more understandable, step by step.
The Executive Board of the Austrian Science Fund (FWF)
Science is facing numerous challenges. This is nothing new, it’s virtually an integral part of science. Or put another way, you could say that without challenges, there would probably be no science at all. Basically, the only thing that changes is the nature of the respective challenges. Some have been looming for years and have recently increased in intensity and urgency, such as man-made climate change. Others are just coming at us like a torrent, like the myriad changes brought about by the ever-improving artificial intelligence. And still others are consequences of political developments, here I am thinking primarily of the freedom of research and access to its findings. And then, at least in Austria, the question of funding is also a current issue and a real challenge.
In 2025, the FWF did its utmost to support and fund the work of excellent researchers, not only securing academic careers, but also helping to make new findings possible. The numbers resulting from these efforts and illustrated in this Annual Report are impressive, especially when you consider how much creativity and energy, how much talent and knowledge the more than 5,000 researchers who have received FWF funding have put into their scientific work.
I am confident that institutions such as the FWF will continue to strengthen research in Austria and give young scientists in particular the opportunity to demonstrate their excellence in an international environment.
Federal President of the Republic of Austria Alexander Van der Bellen
Science is a cornerstone of a defensible democracy: It provides a fact-based foundation for political decisions, strengthens critical thinking among its citizens, and helps to counter disinformation and populist simplification with facts. Especially at a time like today when democratic institutions are under pressure worldwide, independent research, free speech and discourse, and the protection of scientific integrity are crucial for a resilient democracy.
Reliable and quality-assured research funding is needed to strengthen this role of science in the long term. The FWF and its programs ensure the independence of basic research, fund high-risk, innovative projects, and give researchers the freedom they need to pursue long-term work.
Equal opportunities and diversity in the science and research system are particularly important to me. The FWF steps up to its responsibility in this area: through transparent procedures, targeted programs to fund early-stage researchers, measures to improve the compatibility of research careers with family life, and measures to increase the number of female researchers, especially in disciplines where women are often underrepresented. Excellent research can only thrive in an environment where all talents – regardless of gender, ethnic origin, or life situation – have the same fair chances. This is why we introduced the Perspectives Package for Science and Research in 2025, which helps research institutions attract international researchers and strengthens Austria as a research location.
A highlight in 2025 was the Austrian Science Awards, where the FWF ASTRA Awards were presented for the first time – it is especially noteworthy that half of these awards went to women. Elly Tanaka, the 2025 FWF Wittgenstein Award winner, is an outstanding personality whose internationally recognized work in regeneration research stands for excellence, pioneering spirit, and perseverance, making her an important role model for young researchers. I would also like to congratulate the FWF on granting €340 million in funding and facilitating 749 new projects. A total of 2,599 projects employing over 5,000 people are currently being funded by the FWF.
Following the successful establishment of the excellent=austria initiative over the past few years, the Clusters of Excellence with over 1000 researchers are already reporting their first successful results. They are also actively engaged in outreach, dialog, and knowledge transfer activities. I am looking forward to the next award round of the second Emerging Fields program in early 2026 with great anticipation and curiosity.
Finally, I would like to express my sincere appreciation to the FWF and all its employees and congratulate the successful researchers and participating research institutions on their achievements.
Eva-Maria Holzleitner (Federal Minister for Women, Science and Research)
When Wikipedia first became big on the internet, many academics and intellectuals were very suspicious of it. How could a crowd-sourced repository of information about everything on earth really be accurate? It seemed perfectly fine when you used it to find out things about which you were ignorant, but sometimes when you looked for things about your own area of expertise, it was inaccurate, eccentric, random, and messy. It was a bit like reading the newspaper. I generally trust what I read in reputable print newspapers on matters to do with politics and the world, but whenever the papers report on something I know about — philosophy or science, say — their reporting was invariably inaccurate, in small but annoying ways. (Curiously, I never generalised this lesson to the other subject-matters the newspapers reported on.)
I used to tell my philosophy students to avoid philosophy entries on Wikipedia. But things have changed. Wikipedia has improved hugely, and is still improving. The advice I give to students these days is that Wikipedia is usually a good source, though they must be careful when cutting and ‘accidentally’ pasting it into their essays. Wikipedia is now one of the many wonderful online sources available to students, many of them free and most of them well-edited.
However, the widespread adoption of generative AI has taken things in a different direction. In addition to using chatbots like ChatGPT and Claude to create documents, people use them for internet searches, to answer questions and requests for information or facts. Google’s Gemini is one of the most popular AI chatbots and is becoming integrated into the whole Google package. In a very important new development, an ‘AI summary’ is now the default first response in a Google search.
This might all seem fine — after all, don’t we all use Google and Wikipedia interchangeably without thinking, and simply rely on the results? What is the difference between looking up something on Wikipedia and using a Google AI summary? Aren’t they ultimately the same kind of thing?
I want to argue that they are actually very different, and the difference derives from fundamental facts about knowledge. Wikipedia has a right to be called a knowledge store, AI Chatbots do not. Let me explain.
In a recent interview, the Nobel Prize winner and founder of Google DeepMind, Sir Demis Hassabis said that one reason that Large Language Models (LLMs, the computational structures behind today’s AI chatbots) are so successful is that they are ‘starting with all of human knowledge, what we put on the internet.’ This claim is very revealing. On the face of it, it might seem plausible — even allowing for exaggeration — but in fact it is profoundly wrong.
Let us begin with the phenomenon of making a mistake. Everyone knows that chatbots make mistakes. Or more precisely, they convey information that isn’t true. We have all experienced it, and this tendency has not been eliminated even from the most advanced models (Google standardly reminds readers of its AI summaries that ‘all responses may include mistakes’). AI researchers have taken to calling these mistakes ‘hallucinations.’ AI hallucinations range from making up non-existent academic or legal documents, to fabricating stories about real people, to simple mathematical and logical errors.
AI scientists tend to dismiss these hallucinations as teething problems, minor malfunctions in the system which can be eliminated as the models are refined and trained further. But in fact, from the point of view of the AI machine itself — that is, the point of view of what it is designed to do — these are not malfunctions. What the chatbot is designed to do is to produce grammatical sentences in response to a prompt, whose content is relevant to the prompt. Hallucinations are therefore not malfunctions: the machine is doing what it is trained to do.
For this reason, the evocative word ‘hallucination’ is deeply misleading. A visual hallucination is when the visual system produces experiences which radically misrepresent how things really are — for example, the hallucinations of those who take LSD, or more tragically those suffering from Charles Bonnet Syndrome or alcoholism. Hallucinations are a failure, a malfunctioning in the visual system. The system is failing to do what evolution designed it to do: represent the visual world accurately. But this is not the case with the hallucinations of LLMs.
The fact that LLMs are not designed to represent things correctly also explains why it doesn’t matter, from their point of view, that not all the information they are trained on is correct. Some of it is misleading, incomplete, or plainly deceptive. This doesn’t matter if the purpose of the machine is to output grammatical sentences. But it does matter if they are meant to create knowledge.
What is wrong with Hassabis’s claim that LLMs are trained on ‘all of human knowledge,’ then, is not just that what is on the internet is incomplete, and therefore not the whole of knowledge; but more importantly, because a lot of what is on the internet is not knowledge at all. It is in the nature of knowledge — a matter of definition, if you want to put it that way — that knowledge must be true. If you claim to know that the Sun orbits the Earth, you are wrong, and therefore you cannot know it. You only think you know it, but you do not know it. And if you claim to know that MMR vaccines cause autism, you are wrong, and therefore you cannot know it, no matter how strongly you believe it.
Some people may object here that people have different ‘knowledges’ and we should not dismiss these alternative views. This is a misunderstanding. Of course, people have different and incompatible views, and how we behave towards those with different views is an important ethical and political question. But if these views are not correct, then they cannot be knowledge. Knowledge is essentially a true or correct representation of reality.
The importance of this point cannot be overstated. The concept of knowledge is central to all human cultures: not all human languages have distinct words for eating and drinking, or for he and she, but all human languages have a word for knowledge, where this implies truth. Knowledge — no matter what you call it in your language — is the concept that classifies a certain kind of correct or true representation. If it is not true, it can’t be knowledge. There is no false knowledge.
But the distinction between truth and falsehood is invisible to LLMs. They construct sentences based on interactions with billions of texts on the internet. But it is not their role — nor the role of the underpaid data-labelling employees who classify their outputs — to distinguish between true and false texts. Any text is good enough, so long as it is relevant to the prompt.
This, then, is the key to understanding the difference between Wikipedia and LLMs. The aim of Wikipedia is to get the correct information, and it tries to achieve this with its huge collection of volunteers, who are regularly updating entries and correcting the interventions of others, according to the fairly strict Wikipedia protocols. What they are aiming at is accuracy, of which truth is a specific version. Sometimes contributors make mistakes, and they are corrected by others. No doubt mistakes remain. But the point is that Wikipedia, unlike an LLM, aims to be trained on the opinions about the real world provided by the volunteers, with the overall goal of representing the world correctly. This is one of the reasons why some leading scientists who are critical of LLMs — like Gary Marcus and Yann LeCun — have argued that AI machines need ‘world models’ in addition to their text-producing abilities. How this idea will develop is one of the most interesting things in AI at the moment.
Someone may object here that most LLMs have swallowed Wikipedia anyway, so what is the real difference in principle between these two systems?
Now, of course it is good that LLMs are drawing on Wikipedia, rather than just on things like conspiracy theory websites. The difference, however, is about the aim of the systems, not just what they are trained on. Although Wikipedia contains false information, its aim is to produce true or correct information. It is a flaw in a Wikipedia entry if it contains falsehoods; but as we have seen, a hallucination in a chatbot is not a flaw, given the point of the chatbot. Wikipedia contributors are working with the aim of making the entries as correct as possible; LLM data labellers do not have that aim.
Wikipedia gives no absolute guarantee of correctness — there is no such thing. But what it gives you is a reliable way of finding things out about the world. LLMs give you a reliable way of producing coherent texts in a few seconds. But they give you no reliable way of finding out about the world. This is why Wikipedia can be a source of knowledge and LLMs are essentially not a source of knowledge, because reliability is essential to the way we acquire knowledge. Knowledge requires truth, but not just truth — because you can obtain a truth by accident (for example, by a lucky guess). What you need for knowledge are reliable methods for achieving truth — for example, by using your eyesight, appealing to established reliable authorities or by using scientific experiments to confirm theories. The idea of a reliable way of achieving truth is the essence of knowledge.
In the FWF Cluster of Excellence “Knowledge in Crisis,” we are applying these fundamental ideas from the theory of knowledge (epistemology) to all of today’s various crises of knowledge — of which the crisis of AI is only one. We have had workshops and public events discussing the relevance of the philosophy of truth and knowledge to the crises, and our researchers have published academic articles and books on these subjects. We collaborate with researchers from other fields — for example, science communication, social sciences, psychology, and economics — on questions of common interest; and this year we will collaborate with the FWF Cluster of Excellence “Bilateral AI” on public events discussing the meaning of today’s AI.
It is remarkable how deep the philosophical roots of these crises go. Sometimes the epistemological crises are discussed merely in terms of things like science denial (e.g., vaccine denial) or the problem of misinformation and disinformation, where people are being deliberately misled about political matters in social media. These are important problems, and at Knowledge in Crisis we have been working hard to understand them. But the epistemology of AI — in particular of LLMs — is something which is less obvious, much deeper and threatens the heart of our knowledge-producing practices. Unless we have a proper understanding of how the working of these machines relates to facts about truth and knowledge, we will dig ourselves deeper into the pit of ignorance and confusion which the AI companies are creating for us. One of our ambitions in Knowledge in Crisis is to create such an understanding, and distribute it to everyone.
Tim Crane, Director of Research, FWF Cluster of Excellence “Knowledge in Crisis”
Tim Crane is Director of Research of the Cluster of Excellence “Knowledge in Crisis” and professor of philosophy and prorector at Central European University in Vienna. He founded the Institute of Philosophy at the University of London and was Knightbridge Professor at the University of Cambridge.
Austria's cutting-edge research continues to be on the upswing, and this growth is also reflected in the funding going towards third-party-funded research. Last year, the Austrian Science Fund (FWF) was able to finance research projects worth €340 million. €136 million went to projects in the natural sciences and technology, €125 million in biology and medical sciences, and €77 million in the humanities and social sciences. The FWF is funding a total of 5,311 researchers in ongoing projects at Austria's universities and other research institutions – a new record.
You can find a review of the past year and all the performance figures in this year’s FWF Annual Report. (currently available only in German; English version will go online soon)