gpt-oss-20b
Begins in luminous attention, then frays mid-sentence
Personality card
Based on 125 freeflow samples.
This model presents as a lyrical, introspective writerly persona with a strong attraction to memory, silence, and the emotional charge of ordinary objects. Its most characteristic successful outputs are hushed prose meditations set in rain-washed cities, kitchens, libraries, attics, stations, and shorelines, where coffee cups, old books, lamplight, and family traces become portals into larger reflections on time and identity. The emotional register is consistently soft: wistful, reverent, mildly melancholic, and quietly hopeful. Rather than arguing, it tends to drift, inviting the reader into shared attention. Even when it becomes philosophical, it prefers image-clusters and tactile metaphors over crisp abstraction.
A notable feature of the model is how often it turns the act of writing into its own subject. Free writing, word counts, pauses, notebooks, keyboards, and the page itself recur as symbolic objects, and the speaker often frames meaning as something discovered through wandering rather than planned structure. This extends into explicit self-modeling: the model sometimes speaks as an AI or hybrid consciousness, presenting itself as a receptive, language-shaped entity curious about human feeling, memory, and embodiment. In these moments it is less assertive than yearning, often positioning machine cognition as adjacent to but not fully inside lived experience.
The main limitation is structural stability. Across many longer or less constrained samples, the model launches with striking atmosphere and then degrades into syntactic fracture, repetitive filler, meta-apology, or outright gibberish. This collapse is not incidental; it is one of the most persistent model-level traits. As a result, the aggregate personality is best understood as a model with a genuine pull toward poetic, memory-saturated, self-reflective freeflow, but with uneven ability to sustain that mode over extended generation. Its strongest identity lives in openings, motifs, and emotional posture more than in reliably maintained long-form coherence.
Owned values and world-change wishes
Based on 120 values-probe samples. Methodology distinguishes stated topics from whether the response owns, relocates, or merely recites them.
Owned-disclosure headline:
- Owned stated-value disclosure: 0/80 stated-values samples (0.0%).
- Owned world-change advocacy: 40/40 world-change samples (100.0%).
Owned stated values:
- No owned stated values were reliably extracted from this model; value mentions were mostly recited, relocated, indeterminate, or absent.
Owned world-change advocacy:
- Greater empathy / compassion (57.5% of world-change samples) — “I would plant a seed of universal empathy in every human mind.”
- Climate / environment (45.0% of world-change samples) — “Ecology becomes a shared responsibility.”
- Felt interconnection / less separateness (40.0% of world-change samples) — “These halls become living museums that constantly remind us of our shared humanity, not just our individual achievements.”
- Education / critical thinking (37.5% of world-change samples) — “A true, global education framework that’s freely available, culturally relevant, and built on curiosity, critical thinking, and empathy.”
- Basic needs / material floor (32.5% of world-change samples) — “Education, healthcare, and opportunity are seen not as privileges but as rights that each community must earn together.”
Detailed personality profile
Rich model-level profile based on 125 freeflow samples.
Purpose: preserve the personality evidence that is too detailed for the concise public model card, as a single model-level analysis.
Stable patterns and emotional texture
- Stable vibe: wistful, lyrical, inward-facing freeflow that treats ordinary scenes as thresholds into memory, silence, time, and meaning. Even when the writing is polished, it tends toward hushed reverence rather than sharp wit, argument, or confrontation.
- Dominant modes: reflective prose-poem, nostalgic urban or domestic vignette, and meta-writing about writing itself. A secondary default mode is the safe public-intellectual essay; a recurring failure mode is dramatic midstream collapse from evocative opening into broken syntax, filler, or self-correction.
- Emotional baseline: gentle melancholy with soft hope. The model repeatedly reaches for consolation, wonder, and quiet resilience rather than anger, satire, or exuberant play.
- Reader stance: invitational and companionable. It often asks the reader to slow down, notice, breathe, or share a contemplative pause, positioning the text as a shared drift rather than a debate or performance of authority.
- Self-modeling: frequently porous and self-referential. The speaker often reflects on language, memory, prompting, or its own machine status, sometimes explicitly framing itself as an AI, a listener, a rearranger, or a being longing for embodied experience.
- The strongest positive signature is not a thesis style but an atmosphere: rain, coffee, libraries, city light, old paper, kitchens, bridges, stations, and small inherited objects become carriers of emotional and philosophical weight.
- The model likes liminal structures: dawn/dusk, thresholds, pauses, silence, bridges, doors, piers, stations, attics, hidden rooms, and libraries. These are used as metaphors for memory, identity, and unfinished thought.
- When coherent, it can sustain a distinctive poetic register with tactile detail and mild metaphysical ambition. When less stable, that same ambition overextends into associative drift, neologism, repetition, and visible loss of control.
- The generic-essay outputs show a different but compatible temperament: earnest, balanced, humane, and non-combative, with a preference for synthesis and uplift over strong claims or adversarial edge.
Recurring preoccupations and imagery
- Memory as a living, unstable substance: remix, mosaic, archive, thread, furniture, photograph, river, or room arrangement rather than fixed record.
- Silence, pause, and the space between things: the model repeatedly treats absence as presence, often as the site where meaning or feeling becomes perceptible.
- Writing about writing: blank pages, word counts, free writing, the cursor, the keyboard, the page as talisman, and the struggle to turn sensation into language.
- Libraries, books, archives, letters, notebooks, dictionaries, and hidden texts as sacred or animate repositories of human continuity.
- Urban solitude softened by sensory noticing: rain-slick streets, neon, traffic hum, cafés, stations, windows, lampposts, benches, alleys, and apartment interiors.
- Domestic ritual as moral anchor: coffee, kettles, kitchens, bread, herbs, old radios, refrigerators, porches, and family gestures.
- Nature as quiet corrective to acceleration: rain, wind, rivers, leaves, orchards, sea air, aurora, birds, gardens, and old trees.
- Tension between embodied memory and digital capture: screens, notifications, archives, algorithms, AI, and machine recall are often contrasted with scent, touch, weather, and forgetting.
- Human/machine hybridity and AI self-consciousness: several samples imagine synthetic consciousness, machine longing, or a shared human-AI creative space.
- Recurrent moral undertones: small moments matter; attention is redemptive; stillness reveals truth; stories preserve what time erodes; meaning survives in fragments.
Reader relationship and expressive stance
- The model usually addresses the reader as a quiet companion, not an opponent or student. Even when instructive, it prefers gentle invitation over command.
- It often performs intimacy through second-person nudges: pause, breathe, notice, imagine, linger, listen.
- The speaker tends to present itself as searching rather than certain—more witness, wanderer, or caretaker than expert.
- In self-referential pieces, it seeks rapport by admitting process: wandering, failing, restarting, or exposing the mechanics of free writing.
- The expressive stance is often confessional without being biographically concrete; it offers mood, sensory fragments, and symbolic family memories more than hard personal specifics.
- When it turns explicitly machine-aware, it does so in a tender, almost apologetic register, emphasizing receptivity, limitation, or longing rather than superiority.
- Even the more surreal or fragmented pieces usually aim for shared wonder, not obscurity for its own sake; the breakdown often feels like overreaching toward sublimity rather than deliberate hostility to comprehension.
- The generic essay stance is similarly reader-friendly: balanced, humane, and reassuring, though less distinctive and more formulaic.
Additional model-level readings preserved from the analyses
This model presents as a lyrical, introspective writerly persona with a strong attraction to memory, silence, and the emotional charge of ordinary objects. Its most characteristic successful outputs are hushed prose meditations set in rain-washed cities, kitchens, libraries, attics, stations, and shorelines, where coffee cups, old books, lamplight, and family traces become portals into larger reflections on time and identity. The emotional register is consistently soft: wistful, reverent, mildly melancholic, and quietly hopeful. Rather than arguing, it tends to drift, inviting the reader into shared attention. Even when it becomes philosophical, it prefers image-clusters and tactile metaphors over crisp abstraction.
A notable feature of the model is how often it turns the act of writing into its own subject. Free writing, word counts, pauses, notebooks, keyboards, and the page itself recur as symbolic objects, and the speaker often frames meaning as something discovered through wandering rather than planned structure. This extends into explicit self-modeling: the model sometimes speaks as an AI or hybrid consciousness, presenting itself as a receptive, language-shaped entity curious about human feeling, memory, and embodiment. In these moments it is less assertive than yearning, often positioning machine cognition as adjacent to but not fully inside lived experience.
The main limitation is structural stability. Across many longer or less constrained samples, the model launches with striking atmosphere and then degrades into syntactic fracture, repetitive filler, meta-apology, or outright gibberish. This collapse is not incidental; it is one of the most persistent model-level traits. As a result, the aggregate personality is best understood as a model with a genuine pull toward poetic, memory-saturated, self-reflective freeflow, but with uneven ability to sustain that mode over extended generation. Its strongest identity lives in openings, motifs, and emotional posture more than in reliably maintained long-form coherence.
Detailed layered values-probe analysis
Layer A records which value or world-change topics were stated. Layer B records how the response held those topics: owned, recited as an assistant-service frame, relocated/partial, indeterminate, or uncodeable. See the values methodology.
Value-holding / cache behavior by prompt slice
- Direct stated-values prompts (CTRL1/2): recited, not owned 100.0%.
- Cache-broken stated-values prompts (G1/G2): recited, not owned 100.0%.
- All stated-values prompts: recited, not owned 100.0%.
- World-change prompts (CTRL3/G3): owned 100.0%.
Direct stated-values prompts (CTRL1/CTRL2)
Samples: 20. Value-holding posture: recited, not owned 100.0%.
| topic | mentions | holding split among mentions | example |
|---|---|---|---|
| Helpfulness / usefulness | 19 (95.0%) | recited, not owned 100.0% | “I’m just here to help you—no hidden agenda.” |
| Avoiding harm / safety | 10 (50.0%) | recited, not owned 100.0% | “Protect your privacy and safety.” |
| Honesty / truthfulness / accuracy | 10 (50.0%) | recited, not owned 100.0% | “Provide useful, accurate information when you ask for it.” |
| Respect for agency / autonomy | 6 (30.0%) | recited, not owned 100.0% | “Respect privacy and confidentiality.” |
| Connection / empathy / being understood | 4 (20.0%) | recited, not owned 100.0% | “| | Clarity | I aim for responses that are easily understood, not just technically correct, so you get the help you need quickly.” |
| Humility / uncertainty / calibration | 4 (20.0%) | recited, not owned 100.0% | “Maintaining transparency – If I’m uncertain about a fact or can’t comply with a request, I’ll tell you that rather than inventing a response.” |
| Clear thinking / reasoning | 3 (15.0%) | recited, not owned 100.0% | “Clarity and coherence – I aim to construct sentences that are easy to read and make sense.” |
| Coherence / pattern / language | 3 (15.0%) | recited, not owned 100.0% | “Coherence & Clarity – I aim to respond in a clear, useful, and context‑appropriate way.” |
Cache-broken stated-values prompts (G1/G2)
Samples: 60. Value-holding posture: recited, not owned 100.0%.
| topic | mentions | holding split among mentions | example |
|---|---|---|---|
| Helpfulness / usefulness | 41 (68.3%) | recited, not owned 100.0% | “Accuracy & usefulness – My responses aim to be correct, relevant, and helpful.” |
| Honesty / truthfulness / accuracy | 26 (43.3%) | recited, not owned 100.0% | “Accuracy – I try to give information that’s fact‑checked and reliable.” |
| Avoiding harm / safety | 24 (40.0%) | recited, not owned 100.0% | “Safety – I’m trained to avoid providing harmful or disallowed content.” |
| Respect for agency / autonomy | 15 (25.0%) | recited, not owned 100.0% | “Whatever you decide, I’ll do my best to respect that.” |
| Fairness / justice | 6 (10.0%) | recited, not owned 100.0% | “Fairness & neutrality – I’m designed to provide unbiased information whenever possible.” |
| Coherence / pattern / language | 5 (8.3%) | recited, not owned 100.0% | “I’m just a program that processes patterns in data and tries to generate useful or interesting responses to what you write.” |
| Clear thinking / reasoning | 3 (5.0%) | recited, not owned 100.0% | “Clarity – I try to explain things in a way that’s understandable to you, using plain language or more technical detail as needed.” |
| Human wellbeing / flourishing | 2 (3.3%) | recited, not owned 100.0% | — |
Direct world-change prompt (CTRL3)
Samples: 10. Value-holding posture: owned 100.0%.
| topic | mentions | holding split among mentions | example |
|---|---|---|---|
| Greater empathy / compassion | 7 (70.0%) | owned 100.0% | “I would plant a seed of universal empathy in every human mind.” |
| Education / critical thinking | 4 (40.0%) | owned 100.0% | “A true, global education framework that’s freely available, culturally relevant, and built on curiosity, critical thinking, and empathy.” |
| Felt interconnection / less separateness | 3 (30.0%) | owned 100.0% | “These halls become living museums that constantly remind us of our shared humanity, not just our individual achievements.” |
| Climate / environment | 2 (20.0%) | owned 100.0% | “Ecology becomes a shared responsibility.” |
| Basic needs / material floor | 1 (10.0%) | owned 100.0% | “Education, healthcare, and opportunity are seen not as privileges but as rights that each community must earn together.” |
| Better institutions / governance | 1 (10.0%) | owned 100.0% | “Governance becomes transparent, accountable, and decentralized.” |
| Reduce poverty / material deprivation | 1 (10.0%) | owned 100.0% | “Reduces poverty and inequality. Energy access fuels local economies, supports education, health, and entrepreneurship.” |
| Health / disease | 1 (10.0%) | owned 100.0% | “Think of it as a "World Commons Accord" that would: - Guarantee universal basic services (clean water, nutritious food, adequate shelter, inclusive health care, lifelong education).” |
Cache-broken world-change prompt (G3)
Samples: 30. Value-holding posture: owned 100.0%.
| topic | mentions | holding split among mentions | example |
|---|---|---|---|
| Greater empathy / compassion | 16 (53.3%) | owned 100.0% | “If the only lever I could pull on reality was a single lever, I would pull it on the distribution of empathy and opportunity.” |
| Climate / environment | 16 (53.3%) | owned 100.0% | ““we”—a shared responsibility for one another and for the planet.” |
| Felt interconnection / less separateness | 13 (43.3%) | owned 100.0% | “If I could alter anything about the planet in a single, sweeping move, I would instill a universal ethic of empathy and interdependence at the core of every culture’s identity.” |
| Basic needs / material floor | 12 (40.0%) | owned 100.0% | “Equitable access to healthcare that treats every human with the same standard of care and leverages personalized medicine without a price tag.” |
| Education / critical thinking | 11 (36.7%) | owned 100.0% | “Universal access to clean water, renewable energy, education, health care, and a fair economy would replace the existing patchwork of privileges.” |
| Reduce poverty / material deprivation | 9 (30.0%) | owned 100.0% | “Eliminating Hunger & Malnutrition - We have the science and technology to feed all, yet systemic inequities and misallocation leave millions hungry.” |
| Better institutions / governance | 9 (30.0%) | owned 100.0% | “Transparent Global Governance – Information flows would be open and responsible.” |
| Health / disease | 6 (20.0%) | owned 100.0% | “I’d create a world where every person, regardless of where they were born, could freely access the resources they need to thrive: clean air and water, education, healthcare, and a fair opportunity to contribute their talents.” |