Tenders Are Invited For Research And Development Services And Related Consultancy Services – Holistische Evaluation Generativer Foundation Models Im Sicherheitskontext (Hegemon) Holistic Evaluation Of Generative Foundation Models In A Security Context (Hegemon)

Research And Development Services And Related Consultancy Services – Holistische Evaluation Generativer Foundation Models Im Sicherheitskontext (Hegemon) Holistic Evaluation Of Generative Foundation Models In A Security Context (Hegemon). The Term Foundation Model (Less Often: Basic Model) Describes Complex Neuronal Networks That Have Been Trained On Large Amounts Of Data And Are Characterized By A High General Public And Versatility Of Their Possible Uses. For This Purpose, The Models Have Many (Several Hundred Billion) Parameters That Are Learned As Part Of The Training. In The Parameters, The Implicit Knowledge Acquired From The Data Is Represented And Thus Represents The Basis (Foundation) For Further Applications. Generative Foundation Models Differ From Previous Ai Models In The Fact That They Can Be Created On The Basis Of Free Input (Prompt) Complex Text, Image Or Audio Content (Or A Combination Of It) And Therefore Very Versatile For Complex Tasks. The Initial Training Effort Of Such Models Is Extremely High And Therefore Time And Resource-Intensive. This Also Makes Evaluation Much More Complex. While Measurements Are Possible In Classic Regression, Classification And Structural Recognition Tasks Using Quantitative Approaches, Qualitative Approaches Are Necessary For Complex Generative Tasks With The Cover Of Multiple Property Dimensions. Problems Such As Inventing Facts And Sources In Large Voice Models (Hallucinating), The Possibility Of Manipulative Causation Of Non-Intended Model Behavior (Through Direct Or Indirect Prompt Injection), The Perpetuation Of Stereotypes From The Training Data In The Model Output (Bias) As Well As The Necessary High Computing Power For Infection Also Make A Holistic Evaluation All The More Necessary, If A Used Is Considered In Partial Domains Of Internal And External Security. Previous Approaches Are Not Sufficiently Suitable For An Application In The German-Language Context And Taking Into Account Security Domestic-Specific Tasks. Against This Background, The Aim Of The Procedure Is To Provide Innovative Research And Development Services In A Competitive Framework (Challenge) For The Holistic Evaluation Of Generative Foundation Models In The Security Context. This Procurement Procedure Aims At Stimulating Innovative Research And Development Services In A Competitive Framework (“Challenge”) For The Holistic Evaluation Of Generative Foundation Models In A Security Context. Foundation Models Are Complex Neural Networks Trained On Large Amounts Of Data With A High Degree Of Generality And Versatility. Their Implicit Knowledge Is Represented In Billions Of Parameters And Represents The Base (Foundation) For Further Potential Applications. Generative Foundation Models Differ From Previous Ai Models In That They Generate Complex Text, Image Or Audio Content (Or A Combination Of Thesis) On The Base Of Free Input (Prompt) And Can There In A Variety Of Ways For Complex Tasks. The Initial Training Effect For Search Models Is Extremely