arXiv subjects

Hector Zenil

Publications and source records attributed to Hector Zenil.

At least 19 recordsLinked to original sources

arXiv2026

Enhancing Clinical Decision Support and Differential Diagnosis with Knowledge Graphs, and Retrieval Augmented Generation in Generative AI

Diagnostic error carries a burden, while unconstrained large language models (LLMs) remain vulnerable to hallucination and weak integration of quantitative laboratory dynamics. We developed a decision-support pipeline combining disease-specific biomarker correlation graphs, ordinary differential equations (ODEs), deep sequence classification, and retrieval-augmented generation (RAG). For 103 disease classes from a full blood count (FBC) repository, biomarker networks were used as coupling matrices to generate 30 trajectories per disease (3,090 total). A one-dimensional convolutional neural network (CNN) and long short-term memory (LSTM) network classified disease trajectories and six dynamical clusters. A constrained GPT-4o-mini RAG layer used a 19-pattern BMJ Best Practice/NICE corpus to generate differential diagnoses evaluated for diagnostic suitability, evidential grounding, and clinical plausibility. Across five random-seed runs, disease-level accuracy was $0.940 \pm 0.006$ for the CNN (95\% CI 0.933--0.948) and $0.852 \pm 0.019$ for the LSTM (95\% CI 0.828--0.875); the CNN advantage was 8.87 percentage points (95\% CI 6.47--11.27; $t(4)=10.26$, $p=5.1\times10^{-4}$; Hedges' $g=3.67$). Among 100 sampled RAG cases, 96 parsed successfully; evidence was cited in 97.9\%, the true diagnosis was mentioned in 71.9\%, and the composite score was 3.82/5 with a 47.9\% strict pass rate. The central finding was a decoupling between grounding and diagnostic correctness: classifier-correct versus classifier-wrong outputs differed in diagnostic suitability but not evidential grounding. Post-hoc analysis confirmed a 1.02-point diagnostic-score difference (Mann--Whitney $p=0.0024$; Hedges' $g=0.72$), whereas grounding differed by only $-0.02$ points ($p=0.839$; $g=-0.04$).

q-bio.OT↗

arXiv2026

Measuring in-context algorithmic reasoning in language models against an exact Bayes-optimal reference

Whether large language models perform algorithmic inference or pattern completion is hard to test, because most benchmarks supply answers but no distributional reference for what the shown evidence licenses. F-ICL supplies one exactly: we exhaustively enumerate the 86 million valid programs of length at most 13 on a Turing-complete machine F, complement-symmetrised to remove output-polarity bias, and compute the exact posterior under a declared bounded Levin--Solomonoff prior. It is Bayes-optimal for that stated prior rather than universal, and models are never told it exists, so the score reads the inductive prior their served distribution already encodes. Across 105 serving configurations spanning open models from 0.8B to 675B and frontier systems, models answer up to 92% of queries correctly, yet 45 of the 46 exposing distributions sit farther from the F reference than a keystroke reference. This is not an artefact of task selection: on the bit coordinate, the half the length quota cannot distort, 69 of 80 runs stay below the anchor. Fidelity is inert to scale, which accuracy tracks; continuation improves late without converging; and models un-solve a solved task once per two gains, where the F reference does so once per nine and always repairs it. Because absolute distances are reference-dependent, we prove sequential bounds holding for rival priors: any predictor whose prior gives the reference positive weight has bounded cumulative excess loss, and, in a loss never invoking the reference, any Bayesian mixture giving the realised truth positive mass has a bounded truth-loss budget. On 23,998 trajectories, 86.7% already spend over 10 bits of it. Sequences ending by position nine cannot exclude an arbitrarily large finite constant, so these are lower bounds on what a rival prior must already pay. F-ICL is an open benchmark and toolkit.

cs.LG↗

arXiv2026

Tighter Bounds for Algorithmic Complexity Estimation Using a Reusable Code-Based Block Decomposition Method

The Block Decomposition Method (BDM) was introduced as an alternative to popular lossless compression methods such as LZW for estimating algorithmic complexity from the principles of algorithmic probability and classical information theory. It extends the Coding Theorem Method (CTM) from small objects to larger ones by combining local estimates of algorithmic complexity with a global account of repetition based on Shannon entropy. Here, we introduce a version of BDM in which dependencies between blocks are utilized to reduce the length of the description based on reusable program code in the decomposition of an object, and on conditional descriptions capable of accounting for shared structure between observations. We formalize this allocation of descriptive resources as algorithmic attention. Repeated or related components need not be described independently, and the resulting reduction in description length is governed by the amount of shared algorithmic information. We formulate this extension as a reuse optimization problem, show that exact optimization is NP-hard, derive conditions under which it improves upon independent descriptions, relate the achievable gains to algorithmic mutual information, prove the relationship with the previous BDM version, and provide a roadmap for its implementation using CTM-derived complexity and conditional complexity estimates.

cs.IT↗

arXiv2026

Patterns in Individual Blood Count Trajectories in the UK Biobank Characterise Disease-Specific Signatures and Anticipate Pan-Cancer Risk

We investigate the longitudinal behaviour of blood markers from common haematological tests as a marker of disease and as a function of disease progression in a variety of conditions including cancer, cardiovascular disease, and infections. We study confounding and non-confounding factors to allow for the earlier detection of disease and conditions based on their longitudinal signatures from biomarker patterns commonly measured in popular and scalable common blood tests across routine clinical tests, in particular the Complete Blood Count (CBC or FBC). Our analysis with normalised temporal profiles and machine learning techniques even before any symptoms appear demonstrates that analyte-group patterns found in blood testing are disease sensitive and disease specific. We demonstrate that CBC markers contribute to the majority of the predictive signal, while biochemistry and other blood panels provide only a modest additional gain mostly associated to very the individual disease for which the test was designed (e.g. CRP, liver enzymes, blood sugar). Our results demonstrate how regular monitoring, computational intelligence, and machine learning applied to longitudinal CBC data can converge to uncover disease patterns, advancing the potential for precision healthcare and predictive medicine on a mass scale leveraging an existing and pervasive blood test.

q-bio.QM↗

arXiv2026

Quantum LEGO Learning: A Modular Design Principle for Hybrid Artificial Intelligence

Hybrid quantum-classical learning models increasingly integrate neural networks with variational quantum circuits (VQCs) to exploit complementary inductive biases. However, many existing approaches rely on tightly coupled architectures or task-specific encoders, limiting conceptual clarity, generality, and transferability across learning settings. In this work, we introduce Quantum LEGO Learning, a modular and architecture-agnostic learning framework that treats classical and quantum components as reusable, composable learning blocks with well-defined roles. Within this framework, a pre-trained classical neural network serves as a frozen feature block, while a VQC acts as a trainable adaptive module that operates on structured representations rather than raw inputs. This separation enables efficient learning under constrained quantum resources and provides a principled abstraction for analyzing hybrid models. We develop a block-wise generalization theory that decomposes learning error into approximation and estimation components, explicitly characterizing how the complexity and training status of each block influence overall performance. Our analysis generalizes prior tensor-network-specific results and identifies conditions under which quantum modules provide representational advantages over comparably sized classical heads. Empirically, we validate the framework through systematic block-swap experiments across frozen feature extractors and both quantum and classical adaptive heads. Experiments on quantum dot classification demonstrate stable optimization, reduced sensitivity to qubit count, and robustness to realistic noise.

cs.LG↗

arXiv2026

On Solomonoff Induction in Large Language Models and the Limits of Self-Improving: The Singularity Is Not Near Without Symbolic Model Synthesis

On the one hand, the question of whether large language models (LLMs) are Solomonoff induction estimators has become an explicit question at the intersection of Algorithmic Information Theory (AIT) and Machine Learning (ML) of great interest. On the other hand, the now old idea of an AI Singularity that requires a reliable positive-feedback process in which a system can generate, evaluate and retain genuine improvements to itself continues to come up and is a recurrent concept in the discussion of AGI. We connect and provide some answers to these issues based on current assumptions and future developments of neurosymbolic ML. We will demonstrate that cross-entropy, negative log-likelihood and cognate next-token objectives do not or cannot, by themselves, implement Solomonoff induction: they optimise fit to a supplied conditional distribution rather than a program-weighted universal mixture. While more compute within a fixed objective can improve fit without changing the inductive principle, additional computational resources do not intrinsically without external hyper-parameter or architectural changes, behave as optimal predictors in the Solomonoff and Levin sense. While the data-processing inequality (DPI) and Levin non-growth remain valid, we will show that for finite learners and finite observers, theoretical boundaries have less relevancy and generate a drift between possible approaches. To this end, we interpret different resource-bounded estimators as finite tools for mechanism search that show divergence, not violation, of (algorithmic) information conservation laws. A neurosymbolic direction taken by current frontier-model developers points towards the adoption of model synthesis and no longer purely statistical approaches to LLMs but where Solomonoff estimators are possible.

cs.IT↗

arXiv2025

XGBoost-Powered Digital Twins Leverage Routine Blood Tests for Early Detection of Cancer and Cardiovascular Disease

Early detection of cancer and cardiovascular diseases is fundamental to improving patient outcomes and reducing healthcare expenditure. Current cancer screening programs are targeted towards specific cancers and are often inaccessible to large parts of the population, particularly in remote regions. This project aimed to develop digital blood twins: machine learning models that leverage routinely collected blood test data, demographics, comorbidities, and prescribed medications, for scalable and cost-effective disease screening. Digital blood twins were constructed using the UK Biobank dataset (n = 373,269). Using age, sex, comorbidities, medication profiles, and blood test z-scores, three iterations of XGBoost classifiers were trained for broad cancer, colorectal cancer, and cardiovascular disease prediction. Model interpretability was achieved through SHAP and dimensionality reduction analyses (UMAP, t-SNE). Broad-category cancer models achieved ROC-AUC = 0.607-0.706. Colorectal cancer prediction demonstrated excellent discrimination (ROC-AUC = 0.816-0.993), and cardiovascular models showed clinical utility, notably for hypertension (ROC-AUC = 0.813, F1 = 0.861). SHAP revealed consistent importance of age, sex, basophil count, and cystatin C. Immune digital blood twins as an agnostic tool demonstrate proof-of-concept feasibility for accessible, low-cost, and scalable screening of cancer and cardiovascular diseases, supporting future integration into predictive and preventive healthcare.

q-bio.OT↗

arXiv2025

Systematic Reconstruction of Disease Networks from Longitudinal Blood Data for Causal Discovery and Intervention Analysis

We explore the hyperparameters and introduce a methodological framework to convert disease patterns from time series data of blood test results into correlation graphs for causal hypothesis exploration. The networks represent hypotheses that can then be validated or rejected both for causal discovery and causal analysis (under intervention). We synthetically recreated a repository of 105 typical disease longitudinal patterns extracted from medical guidance and research literature of common blood markers to build a systematic pipeline to translate multidimensional clinical data into intervenable disease networks for causal discovery and causal analysis. This study demonstrates that knowledge graphical models reconstructed from longitudinal data can transform routine medical data into clinically interpretable structures. By integrating multiple thresholding strategies and causal graph design, the framework has the purpose to move beyond statistical correlation toward clinically and testable inference networks. These results highlight a practical pathway for more transparent, explainable, and scalable tools in clinical decision support for AI training, precision healthcare and predictive medicine, offering interpretable, clinically actionable outputs that support safer use of AI in differential diagnosis.

q-bio.OT↗

arXiv2025

Multi-omic Enriched Blood-Derived Digital Signatures Reveal Mechanistic and Confounding Disease Clusters for Differential Diagnosis

Understanding disease relationships through blood biomarkers offers a pathway toward data-driven taxonomy and precision medicine. In this study, we constructed a digital blood twin, a computational model derived from 103 disease signatures comprising longitudinal hematological and biochemical analytes. Profiles were standardized into a unified disease-analyte matrix, and pairwise Pearson correlations were computed to assess similarity across conditions. Hierarchical clustering revealed consistent grouping of hematopoietic disorders, while metabolic, endocrine, and respiratory diseases were more heterogeneous, reflecting weaker internal cohesion. To evaluate cluster structure, the tree was partitioned at a stringent distance threshold, yielding 16 groups. Enrichment analysis of the largest and most heterogeneous cluster demonstrated convergence on cytokine-signaling pathways, indicating shared inflammatory mechanisms that transcend conventional clinical boundaries. PCA and UMAP corroborated the correlation-based results, consistently separating hematological diseases as a distinct cluster. Random Forest feature selection identified neutrophils, mean corpuscular volume, red blood cell count, and platelet count as the most discriminative analytes, reinforcing the role of hematopoietic markers as key drivers of disease stratification. Collectively, these findings show that blood-derived digital signatures can recover clinically meaningful disease clusters while uncovering mechanistic overlaps across categories. This network physiology framework highlights the potential of integrating routine laboratory data with computational methods to refine disease ontology, map comorbidities, and advance precision diagnostics.

q-bio.OT↗

arXiv2025

Exhaustive Investigation of CBC-Derived Biomarker Ratios for Clinical Outcome Prediction: The RDW-to-MCHC Ratio as a Novel Mortality Predictor in Critical Care

Ratios of common biomarkers and blood analytes are well established for early detection and predictive purposes. Early risk stratification in critical care is often limited by the delayed availability of complex severity scores. Complete blood count (CBC) parameters, available within hours of admission, may enable rapid prognostication. We conducted an exhaustive and systematic evaluation of CBC-derived ratios for mortality prediction to identify robust, accessible, and generalizable biomarkers. We generated all feasible two-parameter CBC ratios with unit checks and plausibility filters on more than 90,000 ICU admissions (MIMIC-IV). Discrimination was assessed via cross-validated and external AUC, calibration via isotonic regression, and clinical utility with decision-curve analysis. Retrospective validation was performed on eICU-CRD (n = 156530) participants. The ratio of Red Cell Distribution Width (RDW) to Mean Corpuscular Hemoglobin Concentration (MCHC), denoted RDW:MCHC, emerged as the top biomarker (AUC = 0.699 discovery; 0.662 validation), outperforming RDW and NLR. It achieved near-universal availability (99.9\% vs.\ 35.0\% for NLR), excellent calibration (Hosmer--Lemeshow $p = 1.0$; $\mathrm{ECE} < 0.001$), and preserved performance across diagnostic groups, with only modest attenuation in respiratory cases. Expressed as a logistic odds ratio, each one standard deviation increase in RDW:MCHC nearly quadrupled 30-day mortality odds (OR = 3.81, 95\% CI [3.70, 3.95]). Decision-curve analysis showed positive net benefit at high-risk triage thresholds. A simple, widely available CBC-derived feature (RDW:MCHC) provides consistent, externally validated signal for early mortality risk. While not a substitute for multivariable scores, it offers a pragmatic adjunct for rapid triage when full scoring is impractical.

q-bio.QM↗

arXiv2025

Similarity Analysis of Blood Count Reference Intervals Across Continents Reveals No Reproducible Population or Geography-Linked Structure and Supports Personalised Values

Blood reference intervals (RIs) underpin diagnostic interpretation and therapeutic monitoring worldwide. However, many widely used RI systems originate from limited historical cohorts and have been propagated across health systems without harmonised derivation protocols, shared metadata, or cross-population validation. Consequently, the global RI landscape reflects a heterogeneous mixture of legacy standards and local laboratory practices rather than a biologically grounded framework. Here we examine published Complete Blood Count (CBC) reference intervals, one of the most commonly used laboratory panels worldwide. We compiled CBC RI data from 28 countries and analysed their similarity using variability mapping, hierarchical clustering, information-theoretic distances, cohesion benchmarking, and nonlinear manifold visualisation. Body mass index (BMI) served as a methodological positive benchmark and exhibited clear continent-level clustering (mean cohesion approximately 0.78-0.81). In contrast, CBC reference intervals showed no reproducible geography-linked clustering across methods, with uniformly high cohesion scores (mean approximately 1.27-1.30). Weak signals in red-cell indices (MCV, HGB) were unstable across sexes and distance metrics. This absence of structure should not be interpreted as evidence that current CBC reference intervals represent universal biological standards. Rather, it is more consistent with the fragmented and historically inherited nature of the global RI landscape. These findings indicate that published CBC reference intervals do not encode coherent global structure and provide limited support for universal population-based diagnostic thresholds. Instead, they support a transition toward recalibrated and personalised reference frameworks based on longitudinal individual baselines and harmonised derivation standards.

q-bio.OT↗

arXiv2025

A Quantitative Approach to Estimating Bias, Favouritism and Distortion in Scientific Journalism

While traditionally not considered part of the scientific method, science communication is increasingly playing a pivotal role in shaping scientific practice. Researchers are now frequently compelled to publicise their findings in response to institutional impact metrics and competitive grant environments. This shift underscores the growing influence of media narratives on both scientific priorities and public perception. In a current trend of personality-driven reporting, we examine patterns in science communication that may indicate biases of different types, towards topics and researchers. We focused and applied our methodology to a corpus of media coverage from three of the most prominent scientific media outlets: Wired, Quanta, and The New Scientist -- spanning the past 5 to 10 years. By mapping linguistic patterns, citation flows, and topical convergence, our objective was to quantify the dimensions and degree of bias that influence the credibility of scientific journalism. In doing so, we seek to illuminate the systemic features that shape science communication today and to interrogate their broader implications for epistemic integrity and public accountability in science. We present our results with anonymised journalist names but conclude that personality-driven media coverage distorts science and the practice of science flattening rather than expanding scientific coverage perception. Keywords : selective sourcing, bias, scientific journalism, Quanta, Wired, New Scientist, fairness, balance, neutrality, standard practices, distortion, personal promotion, communication, media outlets.

cs.DL↗

arXiv2025

Complexity-Informed Causal Modeling of Neurodevelopmental Trajectories in Pediatric High-Grade Gliomas: Divergences from Neural Stem Cell Signatures

Pediatric high grade gliomas are lethal evolutionary disorders with stalled developmental trajectories and disrupted differentiation hierarchies. We integrate transcriptional and algorithmic network complexity based perturbation analysis to elucidate gene expression patterns and molecular divergence between Diffuse Midline Gliomas and glioblastoma, revealing shared developmental programs steering cell fate decision making. Our complex systems approach supports the emerging paradigm that pediatric high grade gliomas are neurodevelopmental disorders with hybrid lineage identities and disrupted patterning. We identify dysregulated neurodevelopmental and morphogenetic signatures, alongside bioelectric and neurotransmitter signaling programs that alter synaptic organization, neuronal fate commitment, and phenotypic plasticity, regulating glioma phenotypic switching. Causal drivers and regulators of plasticity were predicted as both biomarkers and therapeutic targets, reinforcing the view that pediatric gliomas are disorders of cell fate decisions and collective cellular identity. Decoded plasticity signatures indicate a teleonomic bias toward neural progenitor or neuron like identities, while synaptic transmission gene enrichment supports neuron glioma interactions shaping fate trajectories. These findings advance differentiation therapy as a systems medicine strategy, discovering plasticity regulators to reprogram malignant fates toward stable lineages, offering complex systems based targets for cancer reversion and predictive, preventive, precision oncology.

q-bio.QM↗

arXiv2025

Neurosymbolic Learning for Predicting Cell Fate Decisions from Longitudinal Single Cell Transcriptomics in Paediatric Acute Myeloid Leukemia

Paediatric Acute Myeloid Leukemia is a complex adaptive ecosystem with high morbidity. Current trajectory inference algorithms struggle to predict causal dynamics in AML progression, including relapse and recurrence risk. We propose a symbolic AI and deep learning framework grounded in complexity science, integrating Recurrent Neural Networks, Transformers, and Algorithmic Information Dynamics to model longitudinal single cell transcriptomics and infer complex state transitions in paediatric AML. We identify key plasticity markers as predictive signatures regulating developmental trajectories. These were derived by integrating deep learning with complex systems based network perturbation analysis and dynamical systems theory to infer high dimensional state space attractors steering AML evolution. Findings reveal dysregulated epigenetic and developmental patterning, with AML cells in maladaptive, reprogrammable plastic states, i.e., developmental arrest blocking terminal differentiation. Predictions forecast neurodevelopmental and morphogenetic signatures guiding AML cell fate bifurcations, suggesting ectoderm mesoderm crosstalk during disrupted differentiation. Neuroplasticity and neurotransmission related transcriptional signals implicate a brain immune hematopoietic axis in AML cell fate cybernetics. This is the first study combining RNNs and AID to predict and decipher longitudinal patterns of cell fate transition trajectories in AML. Our complex systems approach enables causal discovery of predictive biomarkers and therapeutic targets for ecosystem engineering, cancer reversion, precision gene editing, and differentiation therapy, with strong translational potential for precision oncology and patient centered care.

q-bio.QM↗

arXiv2025

Binarized Neural Networks Converge Toward Algorithmic Simplicity: Empirical Support for the Learning-as-Compression Hypothesis

Understanding and controlling the informational complexity of neural networks is a central challenge in machine learning, with implications for generalization, optimization, and model capacity. While most approaches rely on entropy-based loss functions and statistical metrics, these measures often fail to capture deeper, causally relevant algorithmic regularities embedded in network structure. We propose a shift toward algorithmic information theory, using Binarized Neural Networks (BNNs) as a first proxy. Grounded in algorithmic probability (AP) and the universal distribution it defines, our approach characterizes learning dynamics through a formal, causally grounded lens. We apply the Block Decomposition Method (BDM) -- a scalable approximation of algorithmic complexity based on AP -- and demonstrate that it more closely tracks structural changes during training than entropy, consistently exhibiting stronger correlations with training loss across varying model sizes and randomized training runs. These results support the view of training as a process of algorithmic compression, where learning corresponds to the progressive internalization of structured regularities. In doing so, our work offers a principled estimate of learning progression and suggests a framework for complexity-aware learning and regularization, grounded in first principles from information theory, complexity, and computability.

cs.LG↗

arXiv2025

Advancing the Scientific Method with Large Language Models: From Hypothesis to Discovery

With recent Nobel Prizes recognising AI contributions to science, Large Language Models (LLMs) are transforming scientific research by enhancing productivity and reshaping the scientific method. LLMs are now involved in experimental design, data analysis, and workflows, particularly in chemistry and biology. However, challenges such as hallucinations and reliability persist. In this contribution, we review how Large Language Models (LLMs) are redefining the scientific method and explore their potential applications across different stages of the scientific cycle, from hypothesis testing to discovery. We conclude that, for LLMs to serve as relevant and effective creative engines and productivity enhancers, their deep integration into all steps of the scientific process should be pursued in collaboration and alignment with human scientific goals, with clear evaluation metrics. The transition to AI-driven science raises ethical questions about creativity, oversight, and responsibility. With careful guidance, LLMs could evolve into creative engines, driving transformative breakthroughs across scientific disciplines responsibly and effectively. However, the scientific community must also decide how much it leaves to LLMs to drive science, even when associations with 'reasoning', mostly currently undeserved, are made in exchange for the potential to explore hypothesis and solution regions that might otherwise remain unexplored by human exploration alone.

cs.AI↗

arXiv2025

Neurodivergent Influenceability as a Contingent Solution to the AI Alignment Problem

The AI alignment problem, which focusses on ensuring that artificial intelligence (AI), including AGI and ASI, systems act according to human values, presents profound challenges. With the progression from narrow AI to Artificial General Intelligence (AGI) and Superintelligence, fears about control and existential risk have escalated. Here, we investigate whether embracing inevitable AI misalignment can be a contingent strategy to foster a dynamic ecosystem of competing agents as a viable path to steer them in more human-aligned trends and mitigate risks. We explore how misalignment may serve and should be promoted as a counterbalancing mechanism to team up with whichever agents are most aligned to human interests, ensuring that no single system dominates destructively. The main premise of our contribution is that misalignment is inevitable because full AI-human alignment is a mathematical impossibility from Turing-complete systems, which we also offer as a proof in this contribution, a feature then inherited to AGI and ASI systems. We introduce a change-of-opinion attack test based on perturbation and intervention analysis to study how humans and agents may change or neutralise friendly and unfriendly AIs through cooperation and competition. We show that open models are more diverse and that most likely guardrails implemented in proprietary models are successful at controlling some of the agents' range of behaviour with positive and negative consequences while closed systems are more steerable and can also be used against proprietary AI systems. We also show that human and AI intervention has different effects hence suggesting multiple strategies.

cs.AI↗

arXiv2025

Can Complexity and Uncomputability Explain Intelligence? SuperARC: A Test for Artificial Super Intelligence Based on Recursive Compression

We introduce an increasing-complexity, open-ended, and human-agnostic metric to evaluate foundational and frontier AI models in the context of Artificial General Intelligence (AGI) and Artificial Super Intelligence (ASI) claims. Unlike other tests that rely on human-centric questions and expected answers, or on pattern-matching methods, the test here introduced is grounded on fundamental mathematical areas of randomness and optimal inference. We argue that human-agnostic metrics based on the universal principles established by Algorithmic Information Theory (AIT) formally framing the concepts of model abstraction and prediction offer a powerful metrological framework. When applied to frontiers models, the leading LLMs outperform most others in multiple tasks, but they do not always do so with their latest model versions, which often regress and appear far from any global maximum or target estimated using the principles of AIT defining a Universal Intelligence (UAI) point and trend in the benchmarking. Conversely, a hybrid neuro-symbolic approach to UAI based on the same principles is shown to outperform frontier specialised prediction models in a simplified but relevant example related to compression-based model abstraction and sequence prediction. Finally, we prove and conclude that predictive power through arbitrary formal theories is directly proportional to compression over the algorithmic space, not the statistical space, and so further AI models' progress can only be achieved in combination with symbolic approaches that LLMs developers are adopting often without acknowledgement or realisation.

cs.AI↗