Essays on neuro-symbolic AI, evaluation, and building a company around a research thesis.https://www.dinu.at/enhttps://www.dinu.at/writing/cortex-a-fixed-point-theory-of-governed-coding-agents/https://www.dinu.at/writing/cortex-a-fixed-point-theory-of-governed-coding-agents/Treating an agent's validate-repair loop as a monotone operator on a lattice of requirements, so that 'the task is done' becomes a least fixed point you can prove it reaches.Mon, 01 Jun 2026 00:00:00 GMTMarius-Constantin Dinuresearch-paperai-agentsformal-methodsai-safetyhttps://www.dinu.at/writing/ringdown-bounds-on-uv-regularized-black-hole-cores/https://www.dinu.at/writing/ringdown-bounds-on-uv-regularized-black-hole-cores/Using the ringdown of merging black holes to bound the size of a non-singular quantum core, and the scaling-law regression framework that turns a whole catalogue of events into one test.Tue, 30 Sep 2025 00:00:00 GMTMarius-Constantin Dinuresearch-papergravitational-wavesquantum-gravityblack-holesai-for-sciencehttps://www.dinu.at/writing/the-inflection-point-the-story-behind-extensityai/https://www.dinu.at/writing/the-inflection-point-the-story-behind-extensityai/Why I wrote The Inflection Point, and how the neuro-symbolic architecture it argues for became a working product with story.one.Sat, 27 Sep 2025 00:00:00 GMTMarius-Constantin Dinuneuro-symbolicsoftware-3-0story-onebookhttps://www.dinu.at/writing/recap-of-the-2nd-international-rl-bootcamp-2025/https://www.dinu.at/writing/recap-of-the-2nd-international-rl-bootcamp-2025/A look back at the RL Bootcamp 2025 in Salzburg: highlights, speakers, and the recordings, slides and community links the organizers published.Thu, 25 Sep 2025 00:00:00 GMTMarius-Constantin Dinureinforcement-learningcommunityeventshttps://www.dinu.at/writing/a-dialogue-on-ai-market-power-and-a-sustainable-future/https://www.dinu.at/writing/a-dialogue-on-ai-market-power-and-a-sustainable-future/Notes on a cross-disciplinary collaboration with Indian legal scholars and the resulting joint report on AI market power and India in a multipolar world.Tue, 19 Aug 2025 00:00:00 GMTMarius-Constantin Dinuai-policymarket-powerneuro-symboliccollaborationhttps://www.dinu.at/writing/hydra-verifiable-knowledge-graphs/https://www.dinu.at/writing/hydra-verifiable-knowledge-graphs/Design-by-Contract applied to LLM-driven ontology and knowledge-graph generation, and an honest negative result about the benchmark used to measure it.Mon, 21 Jul 2025 00:00:00 GMTMarius-Constantin Dinuresearch-paperknowledge-graphsneuro-symbolicllmhttps://www.dinu.at/writing/three-body-problem/https://www.dinu.at/writing/three-body-problem/Inspired by sci-fi's The Three-Body Problem, this post explores chaos, system dynamics, and how AI research automation can master complexity.Wed, 09 Jul 2025 00:00:00 GMTMarius-Constantin Dinuthree-body-problemchaos-theorysystem-dynamicsai-researchhttps://www.dinu.at/writing/primality-testing-via-circulant-matrices/https://www.dinu.at/writing/primality-testing-via-circulant-matrices/A prime is exactly an integer whose circulant matrix has a minimal polynomial with two irreducible factors over the rationals. A clean characterisation, and a slow algorithm.Mon, 28 Apr 2025 00:00:00 GMTMarius-Constantin Dinuresearch-papernumber-theorygalois-theoryai-for-sciencehttps://www.dinu.at/writing/empowering-research-innovation/https://www.dinu.at/writing/empowering-research-innovation/How the neuro-symbolic stack behind SymbolicAI became a research automation platform, and what happened when we pointed it at an open question in primality testing.Wed, 23 Apr 2025 00:00:00 GMTMarius-Constantin Dinuresearch-automationneuro-symbolicsymbolic-aiverificationhttps://www.dinu.at/writing/ai-benchmark-illusion/https://www.dinu.at/writing/ai-benchmark-illusion/AI models ace benchmarks yet fail real-world tests — a critical assessment in evaluations and why we need better frameworks to measure intelligence.Mon, 07 Apr 2025 00:00:00 GMTMarius-Constantin Dinubenchmarksevaluationneuro-symbolicllmhttps://www.dinu.at/writing/pyflow-ts-bridging-ml-research-with-production/https://www.dinu.at/writing/pyflow-ts-bridging-ml-research-with-production/How PyFlow.ts exposes Python ML code to TypeScript with one decorator, removing the API, typing and client boilerplate of the last-mile deployment step.Fri, 07 Mar 2025 00:00:00 GMTMarius-Constantin Dinupyflow-tspythontypescriptml-deploymenthttps://www.dinu.at/writing/year-1-recap/https://www.dinu.at/writing/year-1-recap/Recap of ExtensityAI's first year: origins as a neuro-symbolic research framework, the team, milestones, publications, and plans for 2025.Mon, 16 Dec 2024 00:00:00 GMTMarius-Constantin Dinuextensityaineuro-symbolicresearch-automationyear-in-reviewhttps://www.dinu.at/writing/parameter-choice-and-neuro-symbolic-domain-invariant-learning/https://www.dinu.at/writing/parameter-choice-and-neuro-symbolic-domain-invariant-learning/My doctoral thesis at JKU Linz: how to pick hyperparameters for domain adaptation when the target has no labels, and what to do when you cannot train at all.Mon, 01 Jul 2024 00:00:00 GMTMarius-Constantin Dinuresearch-paperdomain-adaptationneuro-symbolicphd-thesishttps://www.dinu.at/writing/llms-can-self-improve-at-web-agent-tasks/https://www.dinu.at/writing/llms-can-self-improve-at-web-agent-tasks/Harvesting an agent's own WebArena trajectories, filtering them without labels, and fine-tuning on the result — plus two metrics for what a benchmark score hides.Thu, 30 May 2024 00:00:00 GMTMarius-Constantin Dinuresearch-paperai-agentsself-improvementevaluationhttps://www.dinu.at/writing/symbolicai-framework-industry-recognition-and-community-impact/https://www.dinu.at/writing/symbolicai-framework-industry-recognition-and-community-impact/Recognition that the SymbolicAI framework and its research paper received from AI researchers, industry figures, media outlets, and the community.Wed, 20 Mar 2024 00:00:00 GMTMarius-Constantin Dinusymbolic-aineuro-symbolicopen-sourcecommunityhttps://www.dinu.at/writing/extensityai-announces-symbolicai-framework-a-new-paradigm-for-logic-based-ai/https://www.dinu.at/writing/extensityai-announces-symbolicai-framework-a-new-paradigm-for-logic-based-ai/The research paper behind SymbolicAI: a modular framework for logic-based approaches that compose generative models with solvers, plus a benchmark for evaluating LLMs in AI-centric workflows.Sat, 03 Feb 2024 00:00:00 GMTMarius-Constantin Dinusymbolic-aineuro-symbolicresearch-paperopen-sourcehttps://www.dinu.at/writing/a-neuro-symbolic-perspective-on-large-language-models/https://www.dinu.at/writing/a-neuro-symbolic-perspective-on-large-language-models/The Symbolic API: using LLMs as the core of a neuro-symbolic stack, decomposing hard tasks into simple zero-shot operations and recombining them.Sun, 22 Jan 2023 00:00:00 GMTMarius-Constantin Dinuresearch-paperneuro-symbolicsymbolic-aillmopen-sourcehttps://www.dinu.at/writing/infodist-online-distillation-with-informative-rewards/https://www.dinu.at/writing/infodist-online-distillation-with-informative-rewards/Why curriculum-learning agents generalise badly to the next task, and why the culprit is the states they explore rather than the non-stationarity they endure.Fri, 09 Dec 2022 00:00:00 GMTMarius-Constantin Dinuresearch-paperreinforcement-learningcurriculum-learninggeneralizationhttps://www.dinu.at/writing/parameter-choice-in-unsupervised-domain-adaptation-by-aggregation/https://www.dinu.at/writing/parameter-choice-in-unsupervised-domain-adaptation-by-aggregation/Extending weighted least squares to vector-valued functions, with a target error asymptotically no worse than twice the unknown optimal aggregation.Sat, 13 Aug 2022 00:00:00 GMTMarius-Constantin Dinudomain-adaptationunsupervised-learningresearch-papertheoryhttps://www.dinu.at/writing/a-dataset-perspective-on-offline-reinforcement-learning/https://www.dinu.at/writing/a-dataset-perspective-on-offline-reinforcement-learning/Two measures — SACo for exploration, TQ for exploitation — that characterise a behavioural policy's dataset and predict which offline RL algorithms will work on it.Mon, 01 Aug 2022 00:00:00 GMTMarius-Constantin Dinureinforcement-learningoffline-rldatasetsresearch-paperhttps://www.dinu.at/writing/reactive-exploration-in-lifelong-reinforcement-learning/https://www.dinu.at/writing/reactive-exploration-in-lifelong-reinforcement-learning/Tracking continual domain shifts as they happen: why policy-gradient methods adapt faster than Q-learning when the environment keeps moving.Mon, 01 Aug 2022 00:00:00 GMTMarius-Constantin Dinureinforcement-learninglifelong-learningexplorationresearch-paperhttps://www.dinu.at/writing/align-rudder-learning-from-few-demonstrations/https://www.dinu.at/writing/align-rudder-learning-from-few-demonstrations/Replacing RUDDER's LSTM with a profile model from multiple sequence alignment, so reward redistribution works from a handful of demonstrations.Thu, 10 Mar 2022 00:00:00 GMTMarius-Constantin Dinureinforcement-learningreward-redistributionlearning-from-demonstrationsresearch-paperhttps://www.dinu.at/writing/the-balancing-principle-for-parameter-choice-in-domain-adaptation/https://www.dinu.at/writing/the-balancing-principle-for-parameter-choice-in-domain-adaptation/Borrowing the balancing principle from ill-posed inverse problems to justify the regularization parameter when the target domain has no labels.Fri, 24 Sep 2021 00:00:00 GMTMarius-Constantin Dinudomain-adaptationregularizationinverse-problemsresearch-paperhttps://www.dinu.at/writing/lighter-dependency-injection-for-pytorch/https://www.dinu.at/writing/lighter-dependency-injection-for-pytorch/Python borrowed little from the aspect-oriented practices that matured in Java and C#. Lighter brings dependency injection to PyTorch projects.Wed, 09 Jun 2021 00:00:00 GMTMarius-Constantin Dinupythonpytorchdependency-injectionopen-sourcehttps://www.dinu.at/writing/jku-ai-overview-video/https://www.dinu.at/writing/jku-ai-overview-video/A short overview of what AI makes possible, produced at Johannes Kepler University Linz.Thu, 18 Feb 2021 00:00:00 GMTMarius-Constantin Dinujkuartificial-intelligencevideohttps://www.dinu.at/writing/xai-and-strategy-extraction-via-reward-redistribution/https://www.dinu.at/writing/xai-and-strategy-extraction-via-reward-redistribution/Using Align-RUDDER as an interpretability method: multiple sequence alignment surfaces the key events an agent relies on, and those events are usually human-readable sub-tasks.Sat, 17 Oct 2020 00:00:00 GMTMarius-Constantin Dinureinforcement-learningexplainable-aireward-redistributionresearch-paperhttps://www.dinu.at/writing/align-rudder-reward-redistribution-from-few-demonstrations/https://www.dinu.at/writing/align-rudder-reward-redistribution-from-few-demonstrations/The first write-up of Align-RUDDER: two modifications to RUDDER that let profile models built from as few as two demonstrations carry the reward redistribution.Wed, 30 Sep 2020 00:00:00 GMTMarius-Constantin Dinuresearch-paperreinforcement-learningreward-redistributionlearning-from-demonstrationscredit-assignmenthttps://www.dinu.at/writing/overcoming-catastrophic-forgetting-with-context-dependent-activations/https://www.dinu.at/writing/overcoming-catastrophic-forgetting-with-context-dependent-activations/Master's thesis: context-based gating that switches pathways through a network, cutting destructive interference on long task sequences.Mon, 25 Nov 2019 00:00:00 GMTMarius-Constantin Dinucontinual-learningcatastrophic-forgettingreinforcement-learningthesishttps://www.dinu.at/writing/imagine-kara/https://www.dinu.at/writing/imagine-kara/A thank-you to the team behind Imagine Kara as the project wound down.Wed, 09 May 2018 00:00:00 GMTMarius-Constantin Dinuprojectsteamshttps://www.dinu.at/writing/seamlessly-entering-the-crypto-world-with-apollon/https://www.dinu.at/writing/seamlessly-entering-the-crypto-world-with-apollon/A short pointer to a longer write-up on the Apollon project and getting started with masternodes.Wed, 25 Apr 2018 00:00:00 GMTMarius-Constantin Dinublockchainapollonmasternodeshttps://www.dinu.at/writing/earning-bitcoin-with-masternodes/https://www.dinu.at/writing/earning-bitcoin-with-masternodes/How hosting Apollon masternodes worked out over three weeks in 2018, and how the project's search interest compared to the top ten coins.Fri, 20 Apr 2018 00:00:00 GMTMarius-Constantin Dinublockchainapollonmasternodeshttps://www.dinu.at/writing/uni-swift-project/https://www.dinu.at/writing/uni-swift-project/A university iOS app for managing photos, built around actually finding the right image again.Mon, 15 May 2017 00:00:00 GMTMarius-Constantin Dinuswiftiosuniversity-projecthttps://www.dinu.at/writing/deep-learning-script/https://www.dinu.at/writing/deep-learning-script/A DSL for Caffe that cuts the line count of prototyping a network architecture, with a VS Code extension that transpiles to Caffe Script.Wed, 19 Apr 2017 00:00:00 GMTMarius-Constantin Dinudslcaffedeep-learningopen-sourcehttps://www.dinu.at/writing/deep-learning-primer/https://www.dinu.at/writing/deep-learning-primer/Two reference notes from 2016 — an overview of the field, and a practical guide to running deep learning on Ubuntu with Docker.Sat, 26 Nov 2016 00:00:00 GMTMarius-Constantin Dinudeep-learningdockernoteshttps://www.dinu.at/writing/operation-phrike/https://www.dinu.at/writing/operation-phrike/A VR combat-simulation study: Oculus Rift plus Unreal Engine, with sensors reading the test subject's stress level.Wed, 07 Sep 2016 00:00:00 GMTMarius-Constantin Dinuvirtual-realityunreal-engineuniversity-projecthttps://www.dinu.at/writing/internship-report/https://www.dinu.at/writing/internship-report/A Xamarin app that classifies handwritten characters with a neural network, compared against a support vector machine. Written at Siemens Corporate Technology.Wed, 31 Aug 2016 00:00:00 GMTMarius-Constantin Dinuneural-networksxamarininternshipsiemenshttps://www.dinu.at/writing/bachelor-thesis/https://www.dinu.at/writing/bachelor-thesis/Compiling JavaScript to .NET IL by pairing a Coco/R-generated parser with Roslyn's syntax trees.Tue, 21 Jul 2015 00:00:00 GMTMarius-Constantin Dinucompilersroslyndotnetthesis