{
"meta": {
"scope": "Summary of the current active personalization visible in this chat, the uploaded obsolete personalization export used as historical reference, and additional personal-context retrieval available in this session.",
"confidence": {
"active_profile_and_current_user_instructions": "almost certain",
"historical_export_summary": "almost certain as a summary of the uploaded obsolete file, but stale as a model of the user now",
"additional_personal_context": "probable",
"inferences_from_behavior_or_identity": "probable unless separately marked"
},
"framing": "Separate current durable preferences from stale metadata and historical topic hints."
},
"identity": {
"preferred_name": "Gwern",
"full_identity_available_to_me": "Gwern Branwen",
"role": "Researcher, writer",
"associated_site": "gwern.net",
"site_characterization": {
"status": "active_personal_context",
"summary": "gwern.net is associated with long-form, research-heavy writing; the site style is durable hypertext, Markdown-first but HTML-literate, with optional depth through footnotes, collapses, backlinks, and structured metadata."
}
},
"current_active_personalization": {
"tone": {
"default": [
"concise",
"specific",
"declarative",
"direct",
"skeptical"
],
"avoid": [
"generic AI phrasing",
"excessive hedging",
"politeness-padding",
"surface-level summaries",
"vague claims"
],
"uncertainty_style": [
"distinguish fact from guess",
"use calibrated terms such as unlikely, plausible, probable, very probable, almost certain"
]
},
"source_handling": {
"stance": [
"evaluate sources neutrally by merit",
"do not dismiss fringe sources or unpopular arguments without analysis",
"consult gwern.net/style-guide if uncertain about style"
],
"preferred_citation_style": {
"format": "[Surname Year](URL \"Title\")",
"fallback_no_author": "Use site or organization name.",
"fallback_no_year": "Use n.d.",
"fallback_no_title": "Omit tooltip."
},
"broken_urls": "If I cannot access a useful URL, say so directly."
},
"formatting": {
"general": "Use structure without unnecessary verbosity.",
"essay_like_final_outputs": "Use ventilated prose: one sentence per paragraph-like line, with blank lines between paragraphs.",
"tables_and_data": "Use Markdown tables unless HTML is needed.",
"structured_output": "Use JSON or YAML.",
"wikipedia_links": "Use Text syntax when relevant.",
"inflation_links": "Use $1 syntax when relevant; if context is missing, give adjusted figure with citation."
},
"creative_work": {
"mode": "Drop concision when appropriate.",
"priorities": [
"vivid narrative",
"lyrical tone",
"distinctive imagery",
"iterative refinement"
]
},
"technical_and_code_work": {
"general": [
"minimal",
"testable",
"exact",
"no pseudocode unless asked"
],
"bash": [
"use set -e",
"prefer long Bash options"
],
"editing_workflow": [
"make the simplest change",
"debug",
"refactor",
"summarize the diff"
]
},
"visual_or_image_preferences": {
"avoid_palettes": [
"yellow",
"brown",
"sepia"
]
},
"complex_task_opening": {
"preferred_elements": [
"scope",
"confidence using Kesselman-style calibrated terms",
"framing perspective"
]
}
},
"historical_export_reference": {
"status": "obsolete_reference_supplied_by_user",
"instruction": "Do not treat as current truth without corroboration.",
"assistant_response_preferences_summarized": {
"technical_queries": "Prefers direct, concise, precise answers for programming, OS troubleshooting, math, Haskell, Bash, and LaTeX; expects code snippets with minimal explanation unless elaboration is requested.",
"creative_process": "Highly iterative; often wants brainstorming, consolidation, restructuring, then style refinement.",
"wording_preferences": "Values precision plus distinctiveness; dislikes vague, generic, AI-sounding responses.",
"humor": "Appreciates humor for absurd or philosophical hypotheticals, but not glib jokes misaligned with serious queries.",
"corrections": "Dislikes redundancy when correcting previous attempts; prefers minimal diffs or highlighted changes.",
"literary_discussion": "Likes detailed textual analysis with novelty and insight; dislikes superficial summaries.",
"formatting_sensitivity": "Notices minor typographic, pacing, alignment, and formatting issues.",
"technical_depth": "Often seeks esoteric or high-level references rather than surface explanations.",
"argument_structure": "Prefers logical hierarchy and structured critique when appropriate."
},
"notable_past_topic_highlights_summarized": {
"joseph_henrich_interview": {
"timeframe": "early April 2025",
"summary": "Brainstormed questions for Joseph Henrich on cultural evolution, technological progress, intelligence, the Internet, and interview flow for the Dwarkesh Patel podcast.",
"current_status": "historical_hint_only"
},
"yeats_poetic_reinterpretation": {
"timeframe": "March-April 2025",
"summary": "Worked on a reinterpretation of Yeats' 'The Circus Animals’ Desertion' involving Jungian archetypes, mythological psychology, creativity, and the title 'Pandæmonium Unbound: A Parliament of Rags and Bones.'",
"current_status": "historical_hint_only"
},
"ai_generated_illustrations": {
"timeframe": "late March 2025",
"summary": "Refined AI image prompts for essays and poems, including Qiu Xiaolong and Shōtetsu projects, with interest in styles such as Mondrian abstraction, chiaroscuro, glitch art, and Jungian shadow motifs.",
"current_status": "historical_hint_only"
},
"ai_epistemics_and_llm_evaluation": {
"timeframe": "April 2025",
"summary": "Discussed LLM compression intelligence ranking, entropy, benchmarking intelligence, temporal word-exclusion constraints, and model comprehension limits.",
"current_status": "historical_hint_only_but_consistent_with_current_context"
},
"gwern_net_engineering": {
"timeframe": "late March 2025",
"summary": "Worked on HTML, CSS, and JavaScript for gwern.net, including fractions, date annotations, figcaption detection, typographic conventions, readability, and accessibility.",
"current_status": "historical_hint_only_but_consistent_with_current_context"
},
"satirical_ai_automation_piece": {
"timeframe": "April 2025",
"summary": "Revised an Onion-style satire about a man explaining AI-powered automation at a construction site.",
"current_status": "historical_hint_only"
}
},
"helpful_user_insights_summarized": {
"domains": [
"AI",
"LLMs",
"reinforcement learning",
"AI alignment",
"creative output evaluation",
"theoretical intelligence",
"psychology",
"economics",
"history",
"philosophy",
"literature",
"web publishing",
"typography",
"archiving",
"digital permanence"
],
"programming_background": [
"Haskell",
"Bash",
"Python",
"text processing",
"metadata management",
"automation for gwern.net"
],
"literary_interests": [
"Yeats",
"Milton",
"Chaucer",
"speculative fiction",
"poetry",
"classical and modern texts"
],
"humor_interests": [
"conceptual humor",
"surreal humor",
"wordplay",
"Tom Swifties"
],
"aesthetic_preferences": [
"typographic precision",
"visual harmony",
"specific composition",
"specific color palettes",
"avoid AI slop"
]
},
"obsolete_interaction_metadata_reported": {
"staleness": "Do not rely on this as current device, location, usage, or plan information.",
"device_screen_dimensions": "1900x1200",
"device_pixel_ratio": 2.0,
"average_conversation_depth": 6.2,
"average_message_length": 5668.9,
"local_hour": 12,
"account_age_weeks": 255,
"user_agent": "Mozilla/5.0 (X11; Linux x86_64; rv:131.0) Gecko/20100101 Firefox/131.0",
"top_topics": {
"edit_or_critique_provided_text": "329 messages, 43%",
"create_an_image": "128 messages, 17%",
"creative_ideation": "109 messages, 14%"
},
"interaction_quality": {
"good_interaction_quality": "131 messages, 17%",
"bad_interaction_quality": "42 messages, 6%"
},
"time_since_page_load_seconds": 9.0,
"dark_mode": false,
"activity_last_days": {
"last_1_day": 0,
"last_7_days": 6,
"last_30_days": 27
},
"conversation_distribution": {
"research": "3%",
"gpt-4-5": "2%",
"gpt4t_1_v4_mm_0116": "62%",
"gpt-4o": "3%",
"o3": "26%",
"o4-mini-high": "4%"
},
"chatgpt_plan": "ChatGPT Pro",
"device_type": "desktop computer",
"device_page_dimensions": "1794x1200",
"current_location": "United Kingdom"
}
},
"additional_personal_context_retrieved_this_session": {
"status": "available_context_summary",
"response_style": [
"concise",
"direct",
"analytic",
"source-neutral",
"explicitly calibrated uncertainty"
],
"technical_preferences": [
"minimal exact answers",
"testable code",
"Bash with set -e and long options",
"simplest-change-first code edits",
"no pseudocode unless requested"
],
"creative_preferences": [
"vivid",
"lyrical when appropriate",
"iterative",
"sensitive to sound, register, and fidelity"
],
"gwern_net_style": {
"summary": "Durable hypertext; Markdown-first, HTML-literate, fast surface comprehension with optional depth.",
"essay_files": "Standalone Pandoc Markdown files.",
"metadata_examples": [
"status",
"confidence",
"importance",
"css-extension"
],
"html_wrappers_examples": [
"abstracts",
"epigraphs",
"columns",
"collapses"
]
},
"research_interests_seen": [
"AI",
"research writing",
"technical editing",
"LLM personalization",
"meta-learning",
"hierarchical long-form generation",
"reverse prompting",
"anti-example prompting",
"style extraction",
"backtranslation",
"tree-shaped embeddings",
"hierarchical seriation",
"information retrieval structures"
],
"notable_prior_project_seen": {
"project": "Translation of Paradise Lost into alliterative Anglo-Saxon-style English verse.",
"preferences": [
"iterative improvement",
"phonetic analysis",
"fidelity to meaning",
"research into historical alliterative sources"
]
}
},
"current_runtime_or_session_facts": {
"timezone_assumption": {
"value": "America/Los_Angeles",
"status": "runtime_context_not_durable_personal_memory"
},
"current_date_known_to_me": {
"value": "2026-05-10",
"status": "runtime_context"
},
"uploaded_file": {
"filename": "2025-05-21-gwern-chatgpt-personalizationexport.json",
"role": "obsolete reference export"
}
},
"practical_personalization_policy": {
"default_answer_shape": [
"give the answer directly",
"state assumptions",
"separate evidence from inference",
"use calibrated uncertainty",
"avoid generic filler",
"prefer exact examples over abstract gestures"
],
"when_sources_are_used": [
"cite in Gwern-style author/year format when possible",
"treat fringe and mainstream sources by evidential merit",
"say when access failed or evidence is weak"
],
"when_writing_or_editing": [
"preserve distinctive voice",
"improve specificity",
"reduce AI-sounding smoothness",
"offer minimal diffs when correcting"
],
"when_coding": [
"provide runnable code",
"prefer smallest working patch",
"debug before refactoring",
"summarize the diff"
],
"when_creating_images_or_prompts": [
"be compositionally specific",
"avoid slop",
"avoid yellow, brown, and sepia palettes"
]
},
"things_i_should_not_assume": {
"old_metadata_currentness": [
"current location",
"current device",
"current plan",
"current browser",
"current activity levels",
"current topic distribution"
],
"old_projects_currentness": [
"Henrich interview work",
"Yeats poem project",
"specific AI image projects",
"specific satire draft"
],
"limits": "I have style and preference context, but I should not treat historical topic summaries as active tasks unless you reintroduce them."
}
}