EU-LAC 1,000 Scientist AI Jam Session

Prerequisites
Every participant is expected to:

• Participate in person; no travel stipend will be provided by the hosting organizations
• Have previous experience using GenAI models (OpenAI’s or others)
• Possess deep technical knowledge of their scientific and/or technical domain 
• Have already formulated research questions to be refined on the day
• Work on questions that do not require the use of confidential and/or sensitive information
• Capture/share model outputs for model evaluation and model training purposes.

Content levels

• Basic: 1 hour
• Intermediate: 5 hours
• Advanced: 1 hour

Learn more about course curricula and content levels.

This course is intended for, but not limited to:

• Scientists, researchers, and engineers from Europe and Latin America. 
• Industry participants are welcome: However, please note that access to certain AI models may be limited for participants from industry. 

Benefit for the Participants

Participants will have the opportunity to use the latest available reasoning models including models from Mistral, Nvidia, and other open-weight models. They will explore and assess their capabilities to address research problems.

With guidance and expertise from the participating host organizations and model providers, participants will gain practical experience in writing effective prompts. Preparatory webinars will also be organized ahead of the AI Jam Session to help participants get ready for the event. 

Beyond the technical expertise, participants will benefit from networking opportunities with researchers across various disciplines, hands-on experience with the latest reasoning models, and insights into best practices for integrating them into research workflows. This AI Jam Session also provides a unique chance to contribute to the future development of reasoning models by offering feedback on their use in real-world research challenges.

After completing in this AI Jam Session, participants will be able to:

• Formulate scientific research questions suitable for exploration with AI reasoning models.
• Develop and refine prompts by providing relevant scientific context and structuring complex problems into smaller steps.
• Critically assess model responses using their domain expertise, identifying errors, limitations, and results that require further verification.
• Document and review prompts, model outputs, and observations to support research and model evaluation.
• Identify opportunities and limitations for integrating AI reasoning models into their own research workflows.

To be confirmed.

Each participant will bring problems from their scientific domain and spend the day using the latest capabilities of AI reasoning models to solve complex problems. Participants can work in small teams (e.g., 1-3) to avoid getting stuck and to help observe or analyze the performance of the models. Small teams need to collectively define the challenge problems prior to the AI Jam Session. Access to leading-edge models will be provided for the duration of the AI Jam Session.

AI Jam Session Approach 

The AI Jam Session should follow a “telescoping” approach: breaking research problems into ascending orders of complexity and domain/technical specificity. The general progression should follow this template (when applicable):

• Baselining what the model knows innately about the scientific domain of the problem
• Baselining what the model knows innately about a specific problem 
• Providing additional information about the problem and assessing the impact of the additional information on the answer 
• Prompting the model to list relevant variables + their significance for this problem 
• Unpacking the underpinnings of cause and effect
• Asking the model to represent that relationship in more generalizable formats (graphs etc.) 
• Pushing the model to extrapolate based on an individual variable being tweaked
• Repeating the process for other variables 
• Repeating the process for multiple variables simultaneously 
• Asking the model to generate a “recipe book” like output for how to most efficiently solve the problem; plus design experiments to gather missing data as needed.

Time zone: CET.

17 February 2027 Preparatory webinar

• To help participants prepare for the first AI Jam Session, we will be hosting a training webinar on February 17 at 4:00 PM CET (12:00 PM ART/BRT). This session will last 45-60 minutes and be recorded and available to all participants. 
• During this webinar, we will introduce participants to the technology, describe how it will work during the AI Jam Session, and walk through best practices in terms of how to structure research questions for such an AI Jam Session. 

24 February 2027 In-person AI Jam Session: Scientific Problems Used During the Jam Session

• Participants should come to the AI Jam Session with 3-4 problems to work on. If participants form groups, every participant in the groups still needs to prepare 3-4 problems. Participants can prepare any type of problems as long as they do not require any data or prompts that are not fully open. Participants should work on problems that are able to be fully shared.
• Here are a few examples:
• New (i.e., not published in books, journals, conferences) problems for which participants are confident the problem is solvable and the solution is verifiable. For example, this could be the derivation / extension of a problem they already solved with published solutions.
• Recent problems for which the participants know the solution but the solution is not known by the models. Examples are accepted but not yet published papers solving a problem. This could also be problems that researchers have explored and solved but not published for which they have experiment / observational / simulation data.
• Design problems where participants ask the model to design something new given some parameters and for which participants are able to explore the quality of the design produced by the model.
• Interpolation / extrapolation problems. An example includes partially mask sections of a 1D time series or 2D-nD datasets and ask the model to interpolate / extrapolate / fit a function.
• Coding or code transformation problems.

• Users can work in small teams (e.g., 1-3) to avoid getting stuck and to help with collectively observing or analyzing the performance of the models.
• Users can use other AI models or tools (provided by themselves) during the AI Jam Session to help refine their understanding of the problems, for calibration, or for analysis of model outputs.
• If other models or tools are used, the full traces (see 4 below) of those model runs should be added to the corpus and clearly identified as part of the run notes or analysis.
• Users will be responsible for capturing and reviewing all information (e.g., prompts, outputs, observations, analysis, and notes) to form their user reviewed corpus. All information will be captured in a way that facilitates review by humans and machines (i.e., readable text).
• Users control what information they share in their user reviewed corpus. However, they are encouraged to share as much information as possible. Users may redact any information that was entered in error (e.g., sensitive information or information identifying the user). Any information identifying users will not be associated with specific entries in the user reviewed corpus. 
• The collected information from each user/group will be aggregated into a single corpus.
• The collected corpus will be reviewed by the organizers for any sensitive or inappropriate information (i.e., individual pieces or aggregates). This technical review with technical tools will form the laboratory reviewed corpus.
• The reviewed corpus will be made available to the participants and the organizers.
• Participants and organizers are free to use the final-approved corpus for research purposes.
• Participants and organizers are free to use the final-approved corpus for model development, training, and model evaluation purposes.
• The organizers will be permitted to add to this data, share this data with future AI Jam Session participants, that participate in future AI Jam Session. Any updated corpus will be shared with organizers that may participate in future AI Jam Sessions. Data sharing beyond AI Jam Session participants is not permitted.
• Results will not be associated with specific models in external publications unless agreed to by all participating model providers on a case-by-case basis.

For participation in Germany, select your preferred hosting location in the application form:

• Garching (Munich) – LRZ/HammerHAI & MPCDF
• Jülich – FZJ/JAIF
• Stuttgart – HLRS/HammerHAI
• Göttingen – GWDG/HammerHAI

Registration links for locations outside Germany will be provided here once available.

Important registration information

• Register at one location only: Please register only for the location where you intend to attend.
• Registration deadline: Registration closes on December 15, 2026.
• Late applications: Late applications may still be accepted depending on course capacity.
• Limited availability: The number of seats at each location is limited due to space constraints.
• Industry participants are welcome: However, please note that access to certain AI models may be limited for participants from industry.

Acceptance notifications and travel arrangements

• Confirmation of participation: You will be notified by mid-January 2027 whether your team has been accepted.
• Travel arrangements: Please do not make any travel arrangements until you have received an official confirmation email from the hosting organization.

Participation in Germany is free of charge.
For participation at the German hosting locations, applicants must be affiliated with an institution in an EU or  EuroHPC JU member country.

Junghwa Lee (HLRS), phone 0711 685 87228, training(at)hlrs.de