Frequently Asked Questions

About MishMash

What is MishMash?

MishMash is a Norwegian research centre dedicated to exploring AI in and through creative practices. It brings together researchers across disciplines and institutions to develop new knowledge, methods, and applications at the intersection of AI, creativity, and society. Read more in the centre description.

What does the name MishMash mean?

MishMash is not an acronym; it is a description of how the founders perceived of the assembled group of people and institutions when the application process started out. At first, MishMash was only thought of as the working title for the proposal. However, it proved so memorable that it stuck. Among more than 150 competing centre proposals pitched to the Research Council of Norway in 2024, the name was one of a few people remembered. It also captures the centre’s mission: merging a deliberate mix of artistic and scientific disciplines into something new. Read the full story of the name.

What is the story behind the MishMash emblem?

The emblem’s overlapping circles symbolise two things at once: AI as a meeting point between humans and machines, and MishMash as a centre doing both artistic and scientific research (and various combinations of the two approaches). The emblem was sketched by the centre director during one of the first application workshops and was refined by the designers at Appex design studio. Fittingly for a centre studying AI, the emblem itself was made entirely by humans. Read the full story of the emblem.

What is the story behind the MishMash cube?

The cube was proposed during one of the early application workshops to explain the multidimensionality of the centre’s themes, approaches, and perspectives. It shows the seven work packages on the front, which govern the organisational structure of MishMash. One of the other sides shows how we aim to combine perspectives from creating, exploring, and reflecting on AI in all work packages. Equally important is to think about how AI influences both machines, humans, and society at large. The visual look was refined by the designers at Appex design studio. Read the full story of the cube.

How is MishMash organised?

MishMash operates as a distributed, collaborative network rather than a single centralized organisation.

Research activities are organised into work packages, which form the core of the centre’s work
Each work package has its own leadership and research agenda
A central management team and board provide coordination and strategic oversight

This structure enables broad participation across institutions while maintaining a shared direction. See Organisation for the full governance structure, including the Council and Scientific Advisory Board.

What makes MishMash different from other AI initiatives?

MishMash puts creativity at the centre. Art is not a way of communicating research done elsewhere; it is what the centre studies, what it makes, how it expresses what it finds and how it inquires. The centre works from three perspectives at once, humans, machines and society, and pays as much attention to the cultural, ethical, legal and societal sides of AI as to the technology. It is also deliberately a network: more than 200 people at partner institutions across Norway and abroad, rather than one building.

Does MishMash focus only on technology?

No. Building AI systems, tools and artworks is one of the centre’s three approaches. The other two are exploring how AI changes creativity and learning, and reflecting on what AI does to culture, work and society. The description explains the three approaches and the work packages show how they meet in practice.

What topics does MishMash cover?

AI in artistic performance and artistic processes; creative uses of AI for health and well-being and in education; the creative and cultural industries, including law, economics and ethics; cultural heritage, archives and collections; and human-centred AI for creative problem-solving, from control rooms to design. Music, film, games, visual art, literature and design all appear. The results list shows what has been published, and the projects what is under way.

What are work packages?

Work packages (WPs) are the main organisational units for research and collaboration within MishMash.

Each WP focuses on a specific theme or research area
Researchers and partners participate through one or more WPs
Most activities (projects, events, and collaborations) are organised within these units

The seven work packages are:

Joining and participating

Who can join MishMash?

Everyone! However, keep in mind that MishMash’s primary mission is to connect researchers (and their institutions) in Norway. That said, we are also eager to connect to non-academic Norwegian partners and both academic and non-academic international partners (institutions) and members (people). The centre is continuously developing its network and welcomes new collaborations (see below).

What is the difference between partners and members?

Partners are institutions (both research-performing and others) that have signed either the original consortium agreement or an associate partner agreement. Check the institution list for an overview of connected partners.

Members are individuals that have signed up for one or more work packages. Many are connected through institutions, but MishMash is also open for non-affiliated researchers.

How do I become a partner and what do I get?

Institutions can express interest in becoming an associate partner by sending an e-mail to contact@mishmash.no for more information. You will then receive details about roles, expectations, and next steps. Participation gives institutions access to the MishMash network and invitation to events. Being a partner is not an end-goal, it should be seen as a door opener for connecting people in the partner institution with people in the MishMash network.

You will not receive any funding by joining as a partner but you do not have to pay anything either. However, we see that many partners apply for (and get!) external funding together with other MishMash partners.

How do I become a member and what do I get?

People can express interest in becoming a MishMash member by sending an e-mail to contact@mishmash.no for more information. You will then receive details about roles, expectations, and next steps.

Members are usually connected to a MishMash partner institution, although we are open for members without institutional backing or without a current affiliation.

You will not receive any funding by joining as a member, but you will be able to apply for seed funding together with other MishMash partners.

Can I join more than one work package?

Yes. Members choose one home work package and may take part in as many others as their work spans. The directory lists every work package a person belongs to, and work package events are open to their members.

How do I decide which work package to join?

Read the seven work package pages and pick the one closest to your own work as your home. If your interests span several, join those too; the MeshUps are a good way to see what each work package is doing before you choose. If in doubt, write to contact@mishmash.no and say what you work on.

Why am I not listed in the directory, and how are the profiles made?

The directory shows members who have registered through the membership form and agreed to be listed; registering counts as consent for name, role, affiliation, identifiers and the links you gave, while a portrait needs a separate yes. If you are missing, the usual reason is that the registration has not reached us yet, or that an entry is waiting on that consent. Profiles are built from what you submitted, and results, affiliations and identifiers are refreshed nightly from NVA and ORCID; the hand-entered fields are never overwritten. To be added, changed or removed, write to contact@mishmash.no or use “Suggest a change” in the footer of your page; the privacy notice has the details.

How can businesses and public organisations engage with MishMash?

Through concrete collaboration rather than membership alone: joint projects and applications with a work package, workshops and hackathons, industry or public-sector PhDs hosted with a partner university, and the seed funding rounds, which favour collaboration between partners. Organisations start by becoming an associate partner, which gives access to the mailing lists and events; what happens next depends on finding the researchers whose work meets your need, and the work package leaders help with that.

What is expected from partners?

Partners play an active role in shaping MishMash. Typical expectations include:

Participating with members (people) in work package actvities
Contributing to communication and visibility by sending information about relevant events that can be added to the events page.
Supporting collaboration across the network and developing applications for MishMash-related projects.

Many activities rely on shared effort across institutions, rather than central coordination alone.

What are seed funding projects?

MishMash supports smaller research initiatives through seed funding announced internally in the centre regularly. See the project pages for examples of ongoing projects. Keep in mind that MishMash primarily supports research activities in Norway. International partners can apply, but there should be a solid Norwegian collaboration involved.

Computing power and AI tools

Where do I start if I need more computing power?

Start by working out which of three jobs you have, because they need very different machines:

Running a model (inference): using a model someone else trained. A good laptop or a single consumer GPU often does this.
Adapting a model (fine-tuning): a workstation GPU or a few hours on a cluster.
Training from scratch: this is the one that needs a national or European supercomputer.

Then climb one step at a time: your own machine, your department’s machine, your institution’s cluster, the national systems, and Europe. Most creative work never needs to go past the second step, and work that has to respond in real time on stage belongs on a machine in the room rather than in a data centre. Ask your local IT service first; they answer faster than any application round.

What is the difference between training and inference, and why does it matter?

This is the distinction that resolves most confused conversations about computing power.

Inference is running a model that already exists: you send it something and it answers. It takes seconds, and what it needs is enough fast memory to hold the model. Every chat window, every transcription, every generated image is inference. It is well over ninety per cent of what most people ever do.
Fine-tuning is taking an existing model and continuing to train it on your own material, so that it learns your voice, your sound or your footage. Hours to days, on one to eight good GPUs. This is where most creative research actually lives.
Training from scratch is building a model from nothing. Weeks on hundreds of GPUs, and a much larger data problem than compute problem. Very few projects should do this.

Why it matters: the national supercomputers are built for training, and they are poor at the thing most people actually want, which is an answer in a second. Chat services are built for inference, and none of them will fine-tune anything for you. Asking for a supercomputer allocation when what you needed was a transcription service is the most common mistake in these conversations.

The two also cost differently, which is why they get different arrangements. Inference cost rises smoothly with use, so it suits a quota. Training is a block of hours booked in advance, so it suits an allocation.

What are Sigma2, NRIS, Olivia and KI-fabrikken?

These names are often used interchangeably, and they are not the same thing:

Sigma2 is the state-owned company that runs Norway’s national computing and storage services.
NRIS (Norwegian Research Infrastructure Services) is the partnership that delivers them: Sigma2 together with the IT groups at UiO, UiB, UiT and NTNU. It is an organisation, not a machine.
Olivia is the machine: Norway’s most powerful supercomputer, built for AI as well as classical simulation.
KI-fabrikken (the Norwegian AI Factory) is Sigma2’s service for public sector bodies and businesses, and Norway’s node in the European LUMI AI Factory.
NREC (Norwegian Research and Education Cloud) is a separate cloud service run by UiO and UiB, suited to hosting services and demonstrations rather than training models.

See Sigma2 and the NRIS documentation.

How do my institution's AI services relate to the national ones?

Five different things get called “AI services”, and most confusion comes from mixing them up. Your institution may run several of them, and they are not substitutes for one another.

A chat gateway: a chat window, usually with an API, in front of a set of models, behind your institutional login, with each model marked for the kind of data it may carry. UiO has GPT UiO, NTNU has GPT NTNU, UiB has UiBchat, and Sikt KI is available across the sector. This is an inference service.
An applied service: one job done well, with no model to choose. UiO’s Autotekst transcribes speech to text and subtitles video, and is used by a number of other institutions too.
A batch cluster: you write a job script, queue it, and collect the result later. Fox at UiO, Idun at NTNU, local clusters elsewhere. This is where fine-tuning happens.
A secure environment: an approved room for confidential or health data with computing inside it. TSD at UiO, SAFE at UiB, HUNT Cloud at NTNU.
An infrastructure cloud: virtual machines you administer yourself, for hosting a service or a demonstration. NREC, run jointly by UiO and UiB.

The national systems sit above the third of these, not the first. Sigma2’s machines are for training and fine-tuning; they will not give you a chat window. If your institution has no cluster, that changes what you should ask the national systems for, but it does not change your access to a gateway, which your institution decides.

One thing worth knowing: the institutional gateways are not simply wrappers around commercial services. GPT UiO serves a number of open-weight models on Norwegian hardware at UiO and NTNU, alongside the commercial ones, with separate access for different data classifications.

I am a researcher or student at a Norwegian university. What can I use?

Your own institution first. UiO has Fox and Educloud Research, NTNU has the Idun cluster, and most other institutions run something local. Many departments also have their own machine-learning servers that are easier to get onto than anything national.

When you outgrow that, apply for the national systems through NRIS. Students cannot apply directly and go through a supervisor.

The national systems have a reputation for being difficult that they no longer deserve. The documentation is step by step, example scripts are provided, courses run regularly, and the staff answer email. Researchers in MishMash who have climbed the whole ladder report that the hardest part was deciding to start.

How do I apply for national computing resources?

Through Sigma2’s application pages. Three things are worth knowing before you start:

Who may apply. Norwegian universities and university colleges, governmental research centres, bodies financed by the Ministry of Education and Research or the Research Council, Norwegian hospitals, and industrial parties. Organisations outside these categories are invited to contact Sigma2 and ask.
Who signs. The project leader must hold a permanent, postdoctoral or temporary position at an eligible organisation. Master’s and PhD students and guest researchers apply through a supervisor or colleague.
When. There are two calls a year, with deadlines usually in February and August, and allocations starting 1 April and 1 October. Requests outside the calls are also accepted and are normally answered within a couple of working days, subject to available capacity, so a small trial allocation does not mean waiting for the next round.

My institution has no cluster. What then?

You can still apply for the national systems. Eligibility follows the type of institution, not whether it owns hardware, so an art school, a university college or a research institute can hold a national allocation without any local cluster at all. Ask your IT service whether anyone at your institution already has an allocation you can join, since joining an existing project is faster than starting one.

For a lot of creative work, one well-specified GPU workstation in the building is a better answer than a share of a supercomputer, and it costs less than people expect.

I work in a company or a startup. What can I use?

Two routes, and the European one is faster than most people realise.

KI-fabrikken is Sigma2’s service for Norwegian businesses and public sector bodies, with computing capacity, advice and support. Get in touch with Sigma2 to discuss what you need.
EuroHPC AI Factory access is free of charge to AI SMEs and startups. Playground access is permanently open and normally decided within two working days; Fast Lane gives up to 50 000 GPU hours and is normally decided within four. Applications go through the EuroHPC access portal. Norway takes part through the LUMI AI Factory, so Norwegian companies are eligible.

Industrial parties may also apply for the national systems on the terms above, with payment for use.

I am an artist or freelancer, or I work at a cultural institution. What can I use?

Honestly, this is the biggest gap in the Norwegian landscape, and MishMash is working on it.

A cultural institution that is a public body may be able to use KI-fabrikken, and any organisation outside the listed categories is invited to ask Sigma2 whether a national allocation is possible. Nobody has yet tested that route for a Norwegian cultural institution, so an early enquiry is useful for everyone.

An individual artist with no employer has no national route today. This is a question of identity rather than policy: the application systems assume an institutional account, so a freelancer cannot get in even where the rules would otherwise allow it. For now the practical answers are hardware you own, and commercial services you pay for per use.

If you are working with a MishMash partner, ask whether you can be added to their project. Access through a colleague is the one door that is open now.

I work with images, sound, music or video. What can I use?

Your own machine, in almost every case, and this is worth stating plainly: every Norwegian institutional AI service is language and speech only. GPT UiO, GPT NTNU, UiBchat, Sikt KI and Autotekst do text and transcription. None of them serves an image, audio, music or video model. A professor at a large university and a freelance artist have the same access to a music model through Norwegian infrastructure, which is none.

So the route is open-weight models run locally, through an interface such as ComfyUI. For video, the models in common use are Wan, HunyuanVideo, LTX-Video, CogVideoX and Mochi; for music, ACE-Step and Stable Audio Open; for real-time audio, RAVE from IRCAM, which plugs into Max and Pure Data. Most of these run on a single consumer GPU with 16 to 32 GB of memory, which is a purchase rather than an application.

Read the licence before you rely on anything. “Open weights” is not one thing: some of these are Apache 2.0 and free to use commercially, some carry revenue thresholds, and some are non-commercial, which means their output cannot be used in commercial work however you host the model.

Closing this gap for Norwegian creative material is one of the reasons MishMash exists.

I work at a museum, library, archive or theatre. What applies to us?

It depends a great deal on which institution, and it is a mistake to treat the cultural sector as one case.

The National Library is at one end. It runs its own AI-lab, has published hundreds of models and datasets openly, has a dedicated state funding line for Norwegian and Sámi language models with matching money to Sigma2 for the computing underneath, and sits alongside UiO and NTNU in Språkmodellklynge Norge. For most purposes it is a producer of national infrastructure rather than a user looking for access.

Most other cultural institutions are at the other end, with no cluster, no allocation and no AI-lab. Two routes are worth trying. KI-fabrikken serves public sector bodies, and a publicly owned or publicly funded institution may qualify. Separately, Sigma2 invites organisations outside its listed eligibility categories to get in touch and ask. As far as we can establish, neither route has yet been tested by a Norwegian museum or theatre, so an early enquiry helps the whole sector by establishing what the answer is.

If you try either, we would like to hear what happened: contact@mishmash.no.

I need something that runs in real time, on stage or in the studio.

Then the national systems are the wrong tool, however powerful they are. They are batch-scheduled, which means you submit a job and wait, and they sit in a data centre far from the room you are playing in. Latency is the whole point of a system that plays along.

A single consumer GPU in the studio, or a laptop with a lot of unified memory, runs most real-time audio and small models well. Unified-memory desktops are strong for inference and weaker for training, so many people train on a cluster and perform on a local machine.

Should I rent computing power or buy a machine?

A rule of thumb from practitioners in the centre: with continuous use, about six months of rented cloud capacity pays for the equivalent machine. Rent for bursts, buy for daily use.

There is a second argument for owning that has nothing to do with money. Research exploration cannot be scheduled in advance, and a metered budget introduces a fear of running out that quietly changes what you are willing to try.

What about ChatGPT, Claude and other commercial tools? Does MishMash pay?

MishMash does not buy subscriptions for its members. Check what your own institution offers before paying for anything yourself.

Several Norwegian institutions run gateways that put a chat interface and an API in front of commercial and open-weight models, with a quota per user and per data classification, so the institution pays for what is actually used. UiO and NTNU both run one. Sikt KI serves universities, colleges, research institutions and upper secondary schools through Feide login.

Paying per use through a gateway is usually far cheaper than buying a seat for every person, because a seat costs the same for the heavy user and for the person who tried it twice.

Where do I find open Norwegian AI models?

Three groups publish them openly:

The Language Technology Group at UiO, which makes the NorMistral family.
NorwAI at NTNU, which makes the NorLLM family.
The National Library’s AI-lab, which makes NB-Whisper for Norwegian speech and the Borealis language models, trained on the library’s collection.

These cover text and speech. Open Norwegian models for music, image, video and motion barely exist, which is one of the things MishMash exists to change.

Does MishMash have computing resources of its own?

Not yet. The centre is setting up a national project allocation so that members can be given access through MishMash rather than only through their own institution, and this page will say how to get on it once it exists.

Beyond that, the centre is working on the structural problem behind these questions: that access in Norway depends on which institution you happen to sit in, and that artists, freelancers and cultural institutions have the least access to the tools they most need. If you have hit a wall we should know about, write to contact@mishmash.no and tell us what you could not get.

Events and activities

What are MeshUps?

MeshUps are short, weekly online events designed for sharing ongoing work and ideas. During the semesters, they run on Thursdays 12:00–12:30. They begin with a couple of minutes of information from the management team. This is followed by a 15-minute presentation and 10 minutes of Q&A.

MeshUps provide a lightweight format for connecting across disciplines and institutions. Upcoming and past MeshUps are listed on the events page.

To receive meeting links and reminders, subscribe to the MeshUps announcement list or add MeshUps to your calendar from the events page. Use the “Subscribe (webcal)” link to add the calendar directly, or use “Download calendar (ICS)” if your mail client (for example Outlook) does not open webcal links.

For Norwegian: For å motta møte-lenker og påminnelser, abonner på MeshUps’ annonse-liste eller legg MeshUps til i kalenderen fra arrangementsiden. Bruk «Subscribe (webcal)» for å legge kalenderen direkte til, eller bruk «Download calendar (ICS)» hvis e-postprogrammet ditt (for eksempel Outlook) ikke åpner webcal-lenker.

Are MishMash events open to the public?

Many MishMash activities are open, including the weekly MeshUps and conferences. See the event page for details. Work package events are typically reserved for members.

How often does MishMash organise conferences?

Twice a year, hosted in turn by the partner institutions: after the opening conference in Kristiansand in September 2026 come Kristiania University College (spring 2027), the University of Inland Norway (autumn 2027), NTNU (spring 2028), the University of Bergen (autumn 2028) and Østfold University College (spring 2029). Between the conferences there are weekly MeshUps, work package workshops and hackathons, and partner events.

How do I find upcoming events?

The events page lists MishMash events and partner events, with a calendar you can subscribe to. Announcements also go to the announcements mailing list, which anyone can join, and to the work package lists for members.

Communication and content

Who is responsible for content and communication?

Communication in MishMash is a shared responsibility across the network.

Work package leaders are responsible for ensuring that relevant activities and results are communicated
Partners contribute content such as news, events, and project updates
A central team coordinates overall communication and website development

In practice, MishMash operates as a collaborative effort where contributions from partners are essential.

How can I contribute content to the website?

MishMash relies on active contributions from its community. You can contribute by:

Submitting news or events to be automatically linked
Adding content to NVA and Orcid that will be automatically synced daily
Requesting access to contribute to the GitHub repository that the site is built from

The website

How does the MishMash website work?

The site is a Jekyll static website. Source code, content, and tooling live in the open mishmash-web repository on GitHub.

Build and publish: Changes merged to the main branch are built automatically by GitHub Actions and published to mishmash.no via GitHub Pages
Quality checks: Pull requests are checked for broken links, directory consistency, and accessibility before merge
Data sync: Person profiles and research results can be updated nightly from NVA and ORCID where linked
Contributing: Partners with repo access can propose edits via pull requests; see CONTRIBUTING.md for what to edit locally

Each page shows metadata at the bottom (last edit, licence, data sources, and a link to its source file on GitHub).

Why does the about page have reading levels?

Some pages on mishmash.no are adaptive: the same content is written at three complexity levels, Simple, Standard and Advanced, and you choose which one to read with the “Reading level” switcher at the top of the page. Your choice is remembered as you browse. The idea is that a school class, a curious visitor, and a fellow researcher should all be able to read the same page without one being bored or the other lost.

Dotted-underlined words are stretchtext: click them to unfold a short explanation in place. All the explanations are collected in the glossary. Try it on the centre description.

Is this website made with AI?

Partly, and openly, which is fitting for a centre that studies creative uses of AI. AI assistance is used for drafting content, developing code and automation, and translation. Human editors review everything before it is published and remain responsible for it, machine-translated pages are marked, and every AI-assisted change is traceable in the public commit history.

The AI colophon declares the practice in full, with live numbers.

Is the website accessible?

The site is built to meet WCAG 2.1 level AA, as the Norwegian regulation on universal design of ICT requires, and every change is checked with two accessibility engines before it is published. The accessibility page says what the site does, how it is checked, and what its known limitations are, and how to report a barrier.

Can students help develop the website?

Yes. The website doubles as a development playground. Student groups can build alternative frontends (switchable UI themes that cannot break the production site) or backend automation (data sync, validation, visualization). See the student development guide in the wiki, the open issues for project ideas, and the website project for who to talk to.

Still have questions?

Get in touch at contact@mishmash.no, or explore projects, people, institutions, events, and results to learn more.