Hudson Golino
@hudsongolino.bsky.social
Associate Professor of Quantitative Methods at the Department of Psychology - University of Virginia. NLP/LLMs, Network Science, Info Theory, Psychometrics. Web: tinyurl.com/HudsonGolino
Web 2: https://r-ega.net
Web 2: https://r-ega.net
Pre-print: osf.io/preprints/ps...
OSF
osf.io
April 22, 2025 at 12:30 PM
Pre-print: osf.io/preprints/ps...
This work exemplifies how innovative psychometric methods can reshape our understanding of fundamental psychological constructs. Its implications extend beyond personality assessment to potentially revolutionize how we conceptualize and measure human traits across psychology.
April 22, 2025 at 12:30 PM
This work exemplifies how innovative psychometric methods can reshape our understanding of fundamental psychological constructs. Its implications extend beyond personality assessment to potentially revolutionize how we conceptualize and measure human traits across psychology.
Our research effectively integrates empirical findings scattered across personality literature into a coherent hierarchical structure. This data-driven framework demonstrates TGA's value for investigating complex psychological constructs and offers a rigorous new perspective on personality taxonomy.
April 22, 2025 at 12:30 PM
Our research effectively integrates empirical findings scattered across personality literature into a coherent hierarchical structure. This data-driven framework demonstrates TGA's value for investigating complex psychological constructs and offers a rigorous new perspective on personality taxonomy.
a "Disinhibition" meta-trait at the third level represents a major departure from the traditional five-factor model. 7/n
April 22, 2025 at 12:30 PM
a "Disinhibition" meta-trait at the third level represents a major departure from the traditional five-factor model. 7/n
While some dimensions aligned with traditional IPIP-NEO structure, we found significant deviations. The emergence of novel dimensions like "Sociability," "Integrity," and "Impulsivity" at the second level and 6/n
April 22, 2025 at 12:30 PM
While some dimensions aligned with traditional IPIP-NEO structure, we found significant deviations. The emergence of novel dimensions like "Sociability," "Integrity," and "Impulsivity" at the second level and 6/n
What makes this approach very interesting is how it addresses longstanding methodological challenges in personality assessment:
Local independence violations
Wording effects
Dimensionality assessment
Structural robustness 5/n
Local independence violations
Wording effects
Dimensionality assessment
Structural robustness 5/n
April 22, 2025 at 12:30 PM
What makes this approach very interesting is how it addresses longstanding methodological challenges in personality assessment:
Local independence violations
Wording effects
Dimensionality assessment
Structural robustness 5/n
Local independence violations
Wording effects
Dimensionality assessment
Structural robustness 5/n
28 first-level dimensions (facets)
6 second-level dimensions (traits)
3 third-level dimensions (meta-traits) 4/n
6 second-level dimensions (traits)
3 third-level dimensions (meta-traits) 4/n
April 22, 2025 at 12:30 PM
28 first-level dimensions (facets)
6 second-level dimensions (traits)
3 third-level dimensions (meta-traits) 4/n
6 second-level dimensions (traits)
3 third-level dimensions (meta-traits) 4/n
imposing existing theoretical models. We applied TGA to the open-source 300-item IPIP-NEO dataset with over 149,000 participants and reveal a three-level structure of personality:
April 22, 2025 at 12:30 PM
imposing existing theoretical models. We applied TGA to the open-source 300-item IPIP-NEO dataset with over 149,000 participants and reveal a three-level structure of personality:
In our preprint (link in the comments) we introduce a new approach to understanding personality structure: Taxonomic Graph Analysis (TGA), is a comprehensive network psychometrics approach that identifies hierarchical personality structures from the bottom up rather than 2/n
April 22, 2025 at 12:30 PM
In our preprint (link in the comments) we introduce a new approach to understanding personality structure: Taxonomic Graph Analysis (TGA), is a comprehensive network psychometrics approach that identifies hierarchical personality structures from the bottom up rather than 2/n
and
Develop interactive dashboards to explore these findings dynamically.
Develop interactive dashboards to explore these findings dynamically.
April 10, 2025 at 3:11 PM
and
Develop interactive dashboards to explore these findings dynamically.
Develop interactive dashboards to explore these findings dynamically.
Looking ahead, I plan to extend this approach:
Implement a large-scale simulation to see the effects of varying the number of items per dimension
Investigate how adversarial inputs or fine-tuning might shift these representational landscapes,
Implement a large-scale simulation to see the effects of varying the number of items per dimension
Investigate how adversarial inputs or fine-tuning might shift these representational landscapes,
April 10, 2025 at 3:11 PM
Looking ahead, I plan to extend this approach:
Implement a large-scale simulation to see the effects of varying the number of items per dimension
Investigate how adversarial inputs or fine-tuning might shift these representational landscapes,
Implement a large-scale simulation to see the effects of varying the number of items per dimension
Investigate how adversarial inputs or fine-tuning might shift these representational landscapes,
By linking internal representations to cognitive functions—ranging from shallow syntactic analysis to deep conceptual abstraction—we can better understand and eventually improve LLM performance, interpretability, and safety.
April 10, 2025 at 3:11 PM
By linking internal representations to cognitive functions—ranging from shallow syntactic analysis to deep conceptual abstraction—we can better understand and eventually improve LLM performance, interpretability, and safety.
Implications and Future Directions
This investigation not only advances our understanding of how GPT-2 (and by extension, other large language models) processes language internally but also opens up new avenues for interpretability research.
This investigation not only advances our understanding of how GPT-2 (and by extension, other large language models) processes language internally but also opens up new avenues for interpretability research.
April 10, 2025 at 3:11 PM
Implications and Future Directions
This investigation not only advances our understanding of how GPT-2 (and by extension, other large language models) processes language internally but also opens up new avenues for interpretability research.
This investigation not only advances our understanding of how GPT-2 (and by extension, other large language models) processes language internally but also opens up new avenues for interpretability research.
Global peaks in NMI values,
Functional labels such as “Token Feature Extraction,” “Contextual Integration,” and “Optimal Abstraction Peak,” and
A horizontal transition boundary, precisely indicating where the network’s representational strategy shifts.
Functional labels such as “Token Feature Extraction,” “Contextual Integration,” and “Optimal Abstraction Peak,” and
A horizontal transition boundary, precisely indicating where the network’s representational strategy shifts.
April 10, 2025 at 3:11 PM
Global peaks in NMI values,
Functional labels such as “Token Feature Extraction,” “Contextual Integration,” and “Optimal Abstraction Peak,” and
A horizontal transition boundary, precisely indicating where the network’s representational strategy shifts.
Functional labels such as “Token Feature Extraction,” “Contextual Integration,” and “Optimal Abstraction Peak,” and
A horizontal transition boundary, precisely indicating where the network’s representational strategy shifts.
Visualization Innovations
One of the project’s central achievements is the development of a mind-blowing NMI heatmap that visually narrates these transitions. The heatmap uses a non-linear Viridis color scale to highlight regions of high semantic coherence and overlays annotations that mark:
One of the project’s central achievements is the development of a mind-blowing NMI heatmap that visually narrates these transitions. The heatmap uses a non-linear Viridis color scale to highlight regions of high semantic coherence and overlays annotations that mark:
April 10, 2025 at 3:11 PM
Visualization Innovations
One of the project’s central achievements is the development of a mind-blowing NMI heatmap that visually narrates these transitions. The heatmap uses a non-linear Viridis color scale to highlight regions of high semantic coherence and overlays annotations that mark:
One of the project’s central achievements is the development of a mind-blowing NMI heatmap that visually narrates these transitions. The heatmap uses a non-linear Viridis color scale to highlight regions of high semantic coherence and overlays annotations that mark:
Layer 13: Pre-Generation Compression
The final layer reverts to lower NMI values. This “compression” is indicative of the model preparing its internal state for next-token prediction rather than maintaining a rich semantic structure.
The final layer reverts to lower NMI values. This “compression” is indicative of the model preparing its internal state for next-token prediction rather than maintaining a rich semantic structure.
April 10, 2025 at 3:11 PM
Layer 13: Pre-Generation Compression
The final layer reverts to lower NMI values. This “compression” is indicative of the model preparing its internal state for next-token prediction rather than maintaining a rich semantic structure.
The final layer reverts to lower NMI values. This “compression” is indicative of the model preparing its internal state for next-token prediction rather than maintaining a rich semantic structure.
Layers 10–12: Conceptual Refinement Layers
The network continues to process and refine these representations. Although NMI values remain high, they reflect a fine-tuning process in conceptual understanding rather than the initial burst of abstraction seen in Layer 9.
The network continues to process and refine these representations. Although NMI values remain high, they reflect a fine-tuning process in conceptual understanding rather than the initial burst of abstraction seen in Layer 9.
April 10, 2025 at 3:11 PM
Layers 10–12: Conceptual Refinement Layers
The network continues to process and refine these representations. Although NMI values remain high, they reflect a fine-tuning process in conceptual understanding rather than the initial burst of abstraction seen in Layer 9.
The network continues to process and refine these representations. Although NMI values remain high, they reflect a fine-tuning process in conceptual understanding rather than the initial burst of abstraction seen in Layer 9.
The robust, disentangled representations here capture personality constructs in a way that aligns with theoretical expectations of deep semantic abstraction.
April 10, 2025 at 3:11 PM
The robust, disentangled representations here capture personality constructs in a way that aligns with theoretical expectations of deep semantic abstraction.
Layers 7–9: Optimal Abstraction Layers
This zone is where the magic happens—Layer 9, in particular, consistently shows the highest semantic coherence (bright yellow on the heatmap).
This zone is where the magic happens—Layer 9, in particular, consistently shows the highest semantic coherence (bright yellow on the heatmap).
April 10, 2025 at 3:11 PM
Layers 7–9: Optimal Abstraction Layers
This zone is where the magic happens—Layer 9, in particular, consistently shows the highest semantic coherence (bright yellow on the heatmap).
This zone is where the magic happens—Layer 9, in particular, consistently shows the highest semantic coherence (bright yellow on the heatmap).
Layers 4–6: Contextual Integration Zone
A transition begins in these layers as context starts to shape token representations. Notably, Layer 6 marks the first significant rise in NMI (approaching 0.91), suggesting a shift toward meaningful semantic integration.
A transition begins in these layers as context starts to shape token representations. Notably, Layer 6 marks the first significant rise in NMI (approaching 0.91), suggesting a shift toward meaningful semantic integration.
April 10, 2025 at 3:11 PM
Layers 4–6: Contextual Integration Zone
A transition begins in these layers as context starts to shape token representations. Notably, Layer 6 marks the first significant rise in NMI (approaching 0.91), suggesting a shift toward meaningful semantic integration.
A transition begins in these layers as context starts to shape token representations. Notably, Layer 6 marks the first significant rise in NMI (approaching 0.91), suggesting a shift toward meaningful semantic integration.
Layers 1–3: Token Feature Extraction
The earliest layers capture raw token-level properties. Here, the low NMI values (deep blue on the heatmap) indicate that the model primarily encodes lexical-syntactic details without forming robust abstract representations.
The earliest layers capture raw token-level properties. Here, the low NMI values (deep blue on the heatmap) indicate that the model primarily encodes lexical-syntactic details without forming robust abstract representations.
April 10, 2025 at 3:11 PM
Layers 1–3: Token Feature Extraction
The earliest layers capture raw token-level properties. Here, the low NMI values (deep blue on the heatmap) indicate that the model primarily encodes lexical-syntactic details without forming robust abstract representations.
The earliest layers capture raw token-level properties. Here, the low NMI values (deep blue on the heatmap) indicate that the model primarily encodes lexical-syntactic details without forming robust abstract representations.
By processing personality-related items through GPT-2 and applying DynEGA to the activation patterns, I discovered a striking organization:
April 10, 2025 at 3:11 PM
By processing personality-related items through GPT-2 and applying DynEGA to the activation patterns, I discovered a striking organization:
and: TEFI (Total Entropy Fit Index): model fit
New Interpretability Roadmap
The result? A Roadmap with embedding dimensions on the x-axis, GPT-2 layers on the y-axis, and NMI as the color gradient — showing where and how semantic structure emerges.
New Interpretability Roadmap
The result? A Roadmap with embedding dimensions on the x-axis, GPT-2 layers on the y-axis, and NMI as the color gradient — showing where and how semantic structure emerges.
April 10, 2025 at 3:11 PM
and: TEFI (Total Entropy Fit Index): model fit
New Interpretability Roadmap
The result? A Roadmap with embedding dimensions on the x-axis, GPT-2 layers on the y-axis, and NMI as the color gradient — showing where and how semantic structure emerges.
New Interpretability Roadmap
The result? A Roadmap with embedding dimensions on the x-axis, GPT-2 layers on the y-axis, and NMI as the color gradient — showing where and how semantic structure emerges.