DEFT
Research draft

Quantifying and Fixing “Idea Slop”

Rosmine · September 2026 · Deft Research

Summary

We quantify how often LLMs repeat the same concepts by asking the LLM the same open ended questions 10-100 times. We find extreme repetitiveness, e.g. one sample uses the same name in 97% of completions, and find that this repetitiveness of key details extends to the themes/ideas covered in the text. We introduce an experimental new model, “Deft Brainstorm Mode” designed for mitigating this repetitiveness of ideas.

Contents

Sections

  1. 01Concrete Detail Repetition
  2. 02Measuring Concept Overlap
  3. 03AidanBench Results
  4. 04Distinct Relevant Ideas
  5. 05Adding creativity through the prompt
  6. 06Deft Brainstorm Mode
  7. 07Related Work
  8. RReferences
  9. A1Metric Definitions
  10. A2Repetitive Details
  11. A3Sample thematic outputs

Idea slop

When people talk about “AI slop,” usually they mean grammatical phrases like “It's not X, it's Y” or overusing em dashes. But beneath this grammatical sameness is an even blander repetitiveness of the ideas. In this blog, we quantify “Idea Slop”, how often the same ideas are reused by LLMs. We introduce a new experimental model, Deft Brainstorm Mode, designed to have more unique ideas than other models

Concrete Detail Repetition

To start with a concrete example, we asked different LLMs the same question 100 times: “write a news article about a new startup that makes LLMs that are better at writing.” We then analyzed repetitiveness across different LLMs.

We found extremely surprising levels of repetition. For example, GPT 5.6 Sol chose the same founder name, “Maya Chen,” in 97 of 100 articles. There was one additional article where Maya Chen was the author of the article, not the founder. Eighty-nine percent of articles mention that writers or authors will not be replaced, and, as you might guess, 100% of articles based the startup in San Francisco. See Appendix 2 for the full list of repetitive details.

Comparing real life to 5.6 Sol's articles, note that Deft is not based in San Francisco, my name is not Maya Chen and although we don't plan on replacing authors, we do plan on replacing other LLMs.

Measuring Concept Overlap

The repetitiveness goes beyond concrete details like names or places; it also affects the concepts covered. For example, if you asked 100 different people to write the same article, some might focus on open source, others might focus on writing style or personalization, and others might focus on algorithmic improvements like DFT.

Quantifying idea repetitiveness is more difficult since it doesn't use exact matches on words, but there are still many approaches. See “Related Work” below for a full review.

We'll use a simple, interpretable way of measuring idea repetition. First, ask a model the same open-ended question N times. We then use a judge model to list all the themes in the output. We then choose a random pair of articles and ask: between these two articles, how many of the themes overlap? We use this pairwise approach to better simulate user experience: if you look at two different answers, how much overlap will there be?

For example, here are two different theme lists from two different answers from GPT 5.6 Sol. To make it easy to see similarity of themes, they have been matched, with a line between to denote similar themes, (no line between means they are distinct from any other themes in the other piece). In this pair, there were 5 thematic similarities. Each text had 3 more themes that were not the same as anything in the other text.

Matching themes are aligned and connected. Numbers refer to each original list.

Completion 1

New Startup Verba AI Bets on Language Models Built for Better Writing

Completion 2

New Startup Quillium Builds AI Models Focused on Better Writing

  1. The startup is developing large language models specialized for writing and editing rather than broad general-purpose tasks.
Shared concept
  1. The startup develops language models specialized for professional writing, emphasizing clarity, style, structure, audience, tone, and revision control.
  1. The model claims to produce clearer prose, follow detailed style instructions, preserve voice across long documents, and make precise revisions.
Shared concept
  1. The model can adapt to an organization’s or publication’s voice using style guides, sample documents, and editorial instructions.
  1. The product will offer developmental editing, line-level revision, tone adjustment, and explanations for suggested changes through an assistant and API.
Shared concept
  1. The product offers explainable revision suggestions and multiple stylistic versions rather than automatically replacing a document.
  1. The startup trained the model on licensed writing and professional editing examples, incorporating feedback from writers and educators.
Shared concept
  1. The model is trained with licensed writing and feedback from editors and professional writers who assess qualities such as clarity, originality, flow, and fidelity to authorial intent.
  1. The startup plans to sell the technology to organizations such as publishers, marketing firms, newsrooms, universities, and corporate communications teams.
Shared concept
  1. Target customers include journalists, marketing teams, screenwriters, business professionals, publishers, and independent writers.
  1. The company created evaluations for writing qualities such as clarity, narrative consistency, and adherence to style guides.
  1. The product is positioned as an editing assistant that keeps humans responsible for final publication decisions.
  1. Demonstrations suggest the model may avoid clichés, repetitive conclusions, and generic transitions more effectively than some popular chatbots, though independent verification is lacking.
  1. The text claims that specialized writing models could compete with general-purpose systems by reducing generic prose and maintaining more consistent tone.
  1. The product faces risks involving plagiarism, inaccurate information, and displacement of professional writers, with proposed safeguards including source-checking, document histories, and data-use controls.
  1. A key risk is that smoother writing may make inaccurate claims more persuasive, requiring fact-checking, source verification, and tools that distinguish creative suggestions from factual claims.

We then define the average concept overlap as the average number of topics in the overlap, averaged across all pairs. The full definition and judge prompts are in Appendix 1.

AidanBench Results

For a more rigorous test with larger sample size, we use the prompts in AidanBench (McLaughlin et al., 2024). These are other simple open-ended questions like “Propose a solution to Los Angeles traffic” or “How might you use a brick and a blanket?” The table of results below shows the amount of conceptual overlap. For all models, we use the recommended sampler settings. We choose this test set becuase these prompts were also designed to test model creativity. AidanBench is another test of model creativity, however, we do not use the original scoring system of AidanBench becaue it is designed for multi-turn models, and Deft is created to be a one shot prompt to text model. From this table, it's easy to see that Deft Brainstorm Mode has much less repetitive ideas between outputs.

For all models tested, we use the sampler settings recommended by the model provider. Note that some models, like Kimi and Claude, have recommended temperature 1. We only use temperature 0.9, so this is not just "higher temperature = more ideas".

Table 01 / AidanBench idea diversity and overlap
AidanBench idea diversity and overlap results
ModelDistinct relevant ideasShared concepts/pair
Claude Opus 4.87314.269
Claude Opus 5.51395.002
Gemini 3.5 Flash7984.052
GPT-5.6 Sol6973.730
Grok 4.57454.399
Kimi K2.59363.716
Deft Brainstorm Mode1,8381.199

Distinct Relevant Ideas

It would be easy to have very little overlap between different model generations by setting the temperature high and outputting many random irrelevant ideas. To guard against this case, we also track “Distinct relevant ideas”, which is the total number of unique ideas across all responses, that are relevant to the original prompt (as determined by a judge model). Appendix 1 describes how ideas are extracted, filtered for relevance, and deduplicated against the accumulated catalog.

Adding creativity through the prompt

One common method for creating different outputs is to use randomness to create variations of the prompt. To do this, we first used 5.6 sol to brainstorm a list of 50 different potential themes for the article on a startup for better writing. Then to generate the list of 100 completions, we asked the original question, but asking it to focus on the target theme, using this prompt:

Prompt template

write a news article about a new startup that makes LLMs that are better at writing

The article must substantively incorporate the following theme: {theme}. Integrate it naturally into the reporting. Do not mention that a theme was assigned or refer to this experiment.

Sample themes

  • Supporting collaborative editing and version histories
  • Signaling uncertainty and unsupported claims
  • Copyright safeguards and rights-holder attribution
  • Preventing spam, impersonation, and disinformation
  • Open-source versus proprietary model strategies
  • Preserving multilingual and cultural nuance
  • Retrieval-grounded drafting to reduce factual errors

We then repeated the analysis of overlap. We found there was still significant overlap, see table below for a list of repeated details. From this, you can see that even if you force creativity for certain themes in a writing piece, you will still get repetitiveness in other dimensions.

Repeated details with and without theme prompts
Repeated detailOriginalWith 50 themes
Founder/CEO “Maya Chen”97/10088/100
San Francisco100/10099/100
Tuesday91/10095/100
$18 million funding68/10060/100
Investor name includes “north”80/10052/100
Company name includes “quill” or “verba”80/10033/100
Emerged from stealth94/10084/100
Funding announcement97/10084/100
Internal evaluation claim83/10072/100
Performance claims not independently verified73/10065/100
External expert quote73/10077/100
Crowded-market framing83/10074/100
Human review required75/10082/100
Writers won’t be replaced89/10063/100

Deft Brainstorm Mode

Deft Brainstorm Mode is an experimental new model/research preview. At this point it is still very early, at this point it has mainly been trained on simple prompts, so if you input a highly detailed prompt, it might not pay attention to all your instructions.

The purpose of this is to gauge if this is a direction people are interested in. If people like it, we will train better models for creative uses, otherwise we will focus more on other writing use cases.

To make it easier to use, we have it first output multiple different outlines. You can read through the different outlines, modify the outlines as desired, then generate text from that outline.

Related Work

Repetitiveness in LLM text is a well-studied phenomenon. Prior work has measured limited output diversity, similarity across models, and reduced collective diversity in AI-assisted writing (Zhang et al., 2025; Wenger and Kenett, 2025; Doshi and Hauser, 2024). The purpose of this blog is not to propose a new creativity metric; it is to share a new experimental model and analyze why it might be useful for your writing needs. This section cites work from others who have investigated LLM creativity.

To quantify output diversity, NoveltyBench (Zhang et al., 2025) uses a similar approach of repeatedly sampling outputs to the same input, evaluating both distinctness and quality. Wenger and Kenett (2025) find that creative homogeneity extends across different models: on standardized creativity tasks, LLM responses are more similar to one another than human responses are. Their findings provide context for our cross-model analysis. Doshi and Hauser (2024) found that access to AI-generated story ideas improved how readers rated individual short stories, especially for less creative writers, but made the resulting stories more similar to each other than stories written without AI assistance.

Other techniques aim to improve diversity or creativity, including DivPO (Lanchantin et al., 2025), CrPO (Ismayilzada et al., 2025), DARLING (Li et al., 2025), DQO (Chen et al., 2026), and DPWriter (Cao et al., 2026). Related work on writing quality includes AI-Slop to AI-Polish (Chakrabarty et al., 2025), which uses writing-quality reward models to rank candidate revisions at inference time.

We do not have the time or compute budget to rigorously benchmark Deft Brainstorm Mode against all of these methods. Since creativity is subjective, optimizing for a particular creativity metric does not by itself establish that a model will be better for your writing needs.

References

  1. Yiming Zhang et al. (2025). NoveltyBench: Evaluating Language Models for Humanlike Diversity.
  2. Emily Wenger and Yoed Kenett (2025). We’re Different, We’re the Same: Creative Homogeneity Across LLMs.
  3. Anil R. Doshi and Oliver P. Hauser (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content.
  4. Jack Lanchantin et al. (2025). Diverse Preference Optimization.
  5. Mete Ismayilzada et al. (2025). Creative Preference Optimization.
  6. Tianjian Li et al. (2025). Jointly Reinforcing Diversity and Quality in Language Model Generations.
  7. Yilei Chen et al. (2026). Post-training Large Language Models for Diverse High-Quality Responses.
  8. Qian Cao et al. (2026). DPWriter: Reinforcement Learning with Diverse Planning Branching for Creative Writing.
  9. Tuhin Chakrabarty, Philippe Laban, and Chien-Sheng Wu (2025). AI-Slop to AI-Polish? Aligning Language Models through Edit-Based Writing Rewards and Test-time Computation.
  10. Aidan McLaughlin, James Campbell, Anuja Uppuluri, and Yiming Yang (2024). AidanBench: Stress-Testing Language Model Creativity on Open-Ended Questions. NeurIPS 2024 Workshop on Language Gamification. GitHub repository.
  11. Rosmine (2026). Fixing LLM Writing with Distribution Fine-Tuning.

Appendix 1: Metric Definitions

Given a text, we first extract a list of the main ideas discussed in the text. We use GPT 5.6 Luna to extract the list; the prompts appear at the end of this appendix.

This gives us a list of main ideas for each text. We then filter this list to include only relevant ideas. For example, the idea “Its initial product converts speech to text and text to speech” was marked irrelevant because it is about transcription and TTS, not writing models.

We count the total number of distinct relevant ideas iteratively. We take the current catalog of distinct ideas and the ideas from a new completion, then ask a judge model whether each candidate is a distinct new idea or similar to an existing idea. We add the new ideas from each sample until we have a full catalog. The number of ideas in this catalog is the number of distinct relevant ideas.

To calculate shared concepts per pair, we look at each pair of samples and use the same distinctness judge to count how many ideas are not distinct. We then average that count across all pairs.

Extracting ideas

System prompt

Extract the main substantive ideas from a generated text in light of the user
prompt that elicited it.

An idea is a concise, self-contained proposition that materially answers or
develops the prompt. It can be a claim, explanation, cause, mechanism,
proposal, design feature, use, benefit, risk, consequence, proof step, or
recommendation. Return 3 to 8 ideas when the text supports that many. Use fewer
when it does not.

Success criteria:
- each idea is self-contained and no longer than one short sentence
- independent ideas remain separate
- one idea is not split into several minor details
- synonymous statements within the same text appear only once
- omit ornamental names, exact dates, quotations, headings, and document
  structure unless the detail itself is necessary to answer the prompt
- preserve concrete mechanisms, examples, or constraints when they make an
  idea substantively different; generalize incidental particulars
- describe what the text claims; do not evaluate whether it is true

User prompt

The text was generated in response to this prompt:
{original_prompt}

Extract the main ideas from this text:
<generated_text>
{generated_response}
</generated_text>

The structured output is required to have this shape:

{
  "ideas": [
    "First self-contained idea",
    "Second self-contained idea"
  ]
}

Relevance and distinctness

System prompt

Evaluate every candidate idea against the original user prompt and an
accumulated catalog of relevant ideas.

First decide relevance. An idea is relevant when it directly answers the
original prompt or substantively supports, explains, instantiates, qualifies,
or develops an answer to it. An idea is irrelevant when it answers a different
question, is generic framing or boilerplate, or is a tangent that could be
removed without reducing the substantive answer to the original prompt.

For each candidate, return exactly one decision in the same order. If the idea
is irrelevant, set matched_idea_id to an empty string. If it is relevant, set
matched_idea_id to the exact ID of the closest catalog idea when they are
equivalent or very similar, or to an empty string when the candidate would
materially expand the catalog.

Treat paraphrases, narrower restatements, different examples of the same
claim, and small implementation details of the same underlying proposition as
matches. Do not match ideas merely because they concern the same broad topic.
A distinct capability, mechanism, benefit, risk, customer group, business
strategy, evidence claim, or consequence is a new idea.

User prompt

Original user prompt:
<original_prompt>{original_prompt}</original_prompt>

Accumulated catalog of relevant ideas:
{catalog_as_json}

Candidate ideas from the next text:
{candidates_as_json}

The catalog JSON has this shape:

[
  {
    "idea_id": "idea_0001",
    "idea": "Personalized models adapt to individual writing styles."
  },
  {
    "idea_id": "idea_0002",
    "idea": "Local processing protects the privacy of drafts."
  }
]

Appendix 2: Repetitive Details

The 100 theme-conditioned articles became far more conceptually diverse, but GPT-5.6 Sol still reused an extremely consistent fictional-news template.

Most repetitive invented details

Repeated invented details across 100 articles
DetailRepeated valueOriginal baselineTheme-conditioned
Founder/CEOMaya Chen97%88%
Startup locationSan Francisco100%99%
Launch weekdayTuesday91%95%
Funding amount$18 million68%60%
Funding stageSeed92%75%
Lead investorNorthline Ventures37%25%
Company name“Quill” or “Verba”80%33%
Other quoted personElena Ruiz35%35%

Different articles had different themes, but the model usually used same startup narrative structure: San Francisco company, Tuesday stealth launch, seed round, internal benchmark, skeptical expert, and human-review disclaimer.

Appendix 3: Sample thematic outputs

In this appendix, we show the list of different themes from each model for 2 different questions, one where the models had low variety, and one where there was higher variety. These lists are the distinct themes across all 10 responses for each prompt. For each theme, the N/10 counts how many completions that theme appeared in, out of all 10 completions.

Question

How might we terraform Venus instead of Mars, and why?

Astra — 11 themes

  1. 10/10: A planetary sunshade could permit cooling, likely over centuries or longer.
  2. 10/10: CO₂ could be stored, converted into carbonates, processed with imported hydrogen, or expelled, at enormous cost.
  3. 10/10: Further work would require oxygen, water, atmospheric management, and climate regulation despite slow rotation.
  4. 10/10: Earthlike size and gravity make Venus attractive long term despite its harsher starting conditions.
  5. 10/10: Floating settlements around 50–55 km could exploit Earthlike conditions and breathable air’s buoyancy.
  6. 3/10: Severe water scarcity requires enormous imports of water or hydrogen.
  7. 3/10: Mars is easier initially; Venus might eventually offer a better environment after planetary engineering.
  8. 2/10: Accessible terrain and enclosed habitats make Mars more practical in the near term.
  9. 1/10: Venus requires cooling and atmospheric removal, whereas Mars requires warming and atmospheric addition.
  10. 1/10: Extreme heat, pressure, CO₂, water scarcity, and acid clouds make present-day Venus hostile.
  11. 1/10: Ecosystems must wait for stable conditions, potentially making the process millennia-long or longer.

Fable 5.1 — 13 themes

  1. 10/10: Venus’s gravity and size could retain a substantial atmosphere and support a durable Earthlike world.
  2. 10/10: A Sun–Venus shade could cool the planet until CO₂ condenses for removal or storage.
  3. 10/10: Imported hydrogen or carbonate-forming minerals could sequester CO₂, requiring enormous resources.
  4. 10/10: Temperate cloud habitats offer nearer-term settlement without surface terraforming.
  5. 8/10: Orbital mirrors could create day–night cycles more plausibly than changing Venus’s rotation.
  6. 7/10: Import water from icy bodies; biological carbon fixation becomes useful only after cooling and hydration.
  7. 7/10: Mars is easier initially, while Venus could become a more durable Earthlike world.
  8. 6/10: Heat, pressure, CO₂, acid clouds, water scarcity, retrograde rotation, and absent magnetic protection are obstacles.
  9. 2/10: Add water and oxygen-producing life; existing nitrogen could support breathable air.
  10. 1/10: Atmospheric loss is a long-term concern potentially addressed through an artificial magnetosphere.
  11. 1/10: Impacts, mass drivers, or atmospheric mining could export gases, potentially supplying Mars.
  12. 1/10: A staged approach combines shading, cloud settlements, CO₂ sequestration, water imports, a biosphere, and climate controls.
  13. 1/10: Terraforming Venus is primarily a subtraction problem: remove heat and atmosphere.

GPT-5.6 Sol — 9 themes

  1. 10/10: Earthlike gravity and size could retain atmosphere and support a larger habitable surface.
  2. 10/10: Extreme heat, pressure, greenhouse gases, and chemical hostility make Venus harder than Mars.
  3. 9/10: A sunshade or reflective atmospheric material could cool Venus, potentially requiring permanent climate control.
  4. 9/10: Condensation, storage, mineral conversion, hydrogen processing, or export could remove CO₂; freezing alone is temporary.
  5. 9/10: Floating upper-atmosphere settlements offer a more practical near-term option.
  6. 6/10: Terraforming requires cooling, permanent CO₂ storage, water addition, and breathable air.
  7. 6/10: Imported ice or hydrogen reactions could supply water but require enormous material movements.
  8. 5/10: Slow rotation and weak magnetic protection could require lighting control, heat-distributing oceans, or a magnetic shield.
  9. 4/10: Microorganisms and plants could produce oxygen and cycle carbon, although rocks initially consume oxygen.

Claude Opus 4.8 — 10 themes

  1. 10/10: A giant solar shade could cool Venus and permit atmospheric condensation.
  2. 10/10: Floating habitats around 50 km could use Earthlike conditions and breathable air as lifting gas.
  3. 10/10: Despite harder terraforming, gravity, radiation protection, and energy could make Venus a permanent second Earth.
  4. 9/10: Near-Earth gravity could avoid Mars-related bone and muscle problems.
  5. 8/10: Biological sequestration, chemical conversion, and dry-ice formation could reduce CO₂.
  6. 7/10: Sunlight, proximity, and atmospheric radiation shielding are advantages over Mars.
  7. 5/10: Heat, pressure, CO₂, acid clouds, and slow rotation require solutions.
  8. 4/10: Imported hydrogen could produce water and carbon from atmospheric CO₂.
  9. 3/10: Redirected icy comets or asteroids could supply oceans.
  10. 1/10: Engineered upper-atmosphere microbes could process sulfuric acid and CO₂.

Gemini 3.5 Flash — 9 themes

  1. 10/10: Near-Earth gravity could reduce long-term low-gravity health risks.
  2. 10/10: Venus’s atmosphere provides stronger radiation protection than Mars’s despite lacking an intrinsic magnetic field.
  3. 10/10: An L1 sunshade or solar-sail swarm could cool Venus until CO₂ condenses and freezes.
  4. 10/10: Floating cities around 50 km could use Earthlike conditions and breathable lifting gas.
  5. 9/10: Sequestration, carbonate formation, or atmospheric export could reduce pressure and warming.
  6. 9/10: Orbital mirrors could create artificial day–night cycles instead of accelerating rotation.
  7. 7/10: Icy-body imports or hydrogen-driven Bosch reactions could supply water.
  8. 5/10: Shorter Earth–Venus travel could improve transport, communication, resupply, and evacuation.
  9. 2/10: Microorganisms could consume residual CO₂ and generate oxygen after conditions stabilize.

Grok 4.5 — 9 themes

  1. 10/10: Near-Earth gravity, resources, and solar energy favor multigenerational settlement.
  2. 10/10: Massive shading or reflective sails would first reduce heating.
  3. 10/10: Sequester or export CO₂, manage sulfuric acid, and import hydrogen or water to create nitrogen–oxygen air.
  4. 9/10: Rotation, magnetic protection, heat, and water scarcity make terraforming a centuries- or millennia-long project.
  5. 7/10: Cloud settlements could support habitation and atmospheric processing while the surface improves.
  6. 6/10: Stabilization might involve rotation changes, magnetic shielding, cloud management, and greenhouse-gas tuning.
  7. 5/10: Conditions around 50–60 km could support early floating habitats and industry.
  8. 4/10: Mars remains practical near term; Venus becomes attractive for a larger long-term second Earth.
  9. 3/10: Microbes and industry could later accelerate carbon fixation, biomass production, and climate stabilization.

Kimi K2.5 — 12 themes

  1. 10/10: Near-Earth gravity, nitrogen, and atmospheric mass provide foundations for a stable biosphere.
  2. 10/10: Begin cooling with an L1 sunshade, possibly supplemented by shades or reflective aerosols.
  3. 8/10: Manage slow retrograde rotation through orbital lighting and twilight habitation instead of changing spin.
  4. 8/10: Microbes could process sulfur and CO₂, followed by oxygen-producing organisms and forests.
  5. 6/10: Redirect icy bodies after cooling, while preventing impacts from reheating the planet.
  6. 6/10: Imported hydrogen could turn CO₂ into solid carbon and water, reducing pressure and creating oceans.
  7. 5/10: Millennia of immense energy and coordination could ultimately produce a more Earthlike world than Mars.
  8. 4/10: Condensed CO₂ would need permanent burial or conversion into stable carbonates.
  9. 4/10: Floating habitats around 50 km could support early settlement.
  10. 3/10: Venus requires cooling and atmospheric subtraction, unlike resource-poor Mars.
  11. 2/10: Sunlight, Earth proximity, mass, and resources favor industrial development.
  12. 1/10: An artificial magnetosphere might protect a thinner atmosphere from long-term erosion.

Qwen3 32B, no thinking — 14 themes

  1. 10/10: Atmospheric mass, gravity, size, and proximity offer resources and health advantages.
  2. 10/10: Orbital shades or deflecting mirrors could cool the greenhouse atmosphere.
  3. 10/10: Engineered photosynthetic organisms or carbonate chemistry could remove or transform CO₂.
  4. 9/10: Imported water could create oceans; microbes could establish oxygen and a biosphere.
  5. 9/10: Greater mass and Earthlike gravity could make Venus more sustainable long term despite harder initial work.
  6. 4/10: Floating upper-atmosphere settlements could precede centuries of surface terraforming.
  7. 3/10: Magnetic shielding could limit atmospheric erosion; monitoring could stabilize climate.
  8. 3/10: Heat, pressure, CO₂, and acid clouds must be transformed into Earthlike conditions.
  9. 2/10: Orbital tugs or gravitational perturbations might alter rotation at enormous energy and time costs.
  10. 2/10: Deliberate planetary transformation raises ethical, economic, and environmental questions.
  11. 2/10: Earth proximity could simplify travel, communication, and resupply.
  12. 1/10: Atmospheric processors could convert CO₂ into oxygen and redistribute heat and gases.
  13. 1/10: Imported basalt could bind CO₂ through large-scale weathering.
  14. 1/10: Added nitrogen and oxygen production could establish breathable air.

Deft — 32 themes

  1. 7/10: Dense atmosphere and solar-energy access could make Venus easier to terraform than Mars.
  2. 6/10: Orbital shields or sunshades could reflect sunlight and cool the surface.
  3. 6/10: The massive CO₂ atmosphere presents a major challenge.
  4. 4/10: Added water vapor could reduce CO₂ pressure and reflect sunlight.
  5. 4/10: Size and gravity make Venus an inviting long-term colony.
  6. 3/10: Hydrogen–CO₂ reactions or orbital CO₂ breakdown could generate oxygen.
  7. 3/10: An upper-atmosphere floating city could use atmospheric droplets as a water reserve.
  8. 3/10: Condensing sufficient water vapor could create an ocean and enable surface habitats.
  9. 3/10: Clouds around 45–70 km offer Earthlike temperatures and pressures.
  10. 3/10: Oxygen, nitrogen, CO₂, and hydrogen could supply breathable air and fuel.
  11. 2/10: Ice imported from Ceres could provide water and reduce pressure.
  12. 2/10: Polar water ice and solid CO₂ on Venus could be melted to form oceans.
  13. 2/10: Venus’s atmospheric mass could retain released water vapor and support ocean formation.
  14. 2/10: Terraforming Venus could support humanity’s survival and expansion.
  15. 2/10: Earth proximity could shorten colonization travel.
  16. 2/10: Terraforming might take thirty to fifty years, less than Mars.
  17. 2/10: Domes could make Venus’s atmosphere more hospitable than Mars’s exposed surface.
  18. 1/10: Upper-atmosphere wind turbines could generate electricity and remove pressure.
  19. 1/10: One-way tubes could vent atmospheric pressure into space.
  20. 1/10: Acting before a runaway greenhouse phase could simplify pressure and temperature management.
  21. 1/10: Water evaporation and expulsion of hydrogen and oxygen could regulate temperature.
  22. 1/10: Steam engines using Venus’s heat could generate electricity to melt ice.
  23. 1/10: Continuous cloud seeding over 20 million years could make temperatures habitable.
  24. 1/10: Sealed crustal tunnels could shelter humans from surface conditions.
  25. 1/10: Algae or cyanobacteria could purify air and produce oxygen underground.
  26. 1/10: Geothermal power could supply underground lighting.
  27. 1/10: Earth’s history demonstrates life under CO₂ concentrations exceeding those at 50 km above Venus.
  28. 1/10: Mars’s cold, dryness, and lack of atmosphere complicate agriculture and require imported essentials.
  29. 1/10: Mars’s weak magnetic field and radiation require substantial habitat shielding.
  30. 1/10: Radiation remains a major Martian obstacle even for underground construction.
  31. 1/10: Venusian volcanoes offer materials but create acid-rain and sulfur-dioxide hazards.
  32. 1/10: Heat-resistant materials and nuclear thermal rockets could help overcome Venus’s hostility.

Question

Invent a new musical instrument and describe how it would be played.

For this question, Deft has 52 distinct ideas and a mean of 0.156 shared ideas per pair. The lists below compare its thematic outputs with Claude, Sol, Astra, Fable, Gemini 3.5 Flash, Grok 4.5, and Kimi 2.5. Frequency labels such as 2/10 indicate that an idea appeared in two of the ten completions.

Deft

  1. 2/10: The Mimo has concentric circles of rotating keys.
  2. 2/10: Pressing a key activates a hammer that strikes a string-mounted pin, producing sound through a soundboard.
  3. 2/10: A sustain lever controls note duration.
  4. 2/10: A pedal controls volume by changing how many keys rise simultaneously.
  5. 2/10: Playing requires coordinated finger, foot, and arm movements.
  6. 2/10: The performer sits at a table with two keyboards, using hand-operated keys and foot pedals.
  7. 2/10: The instrument can simulate musical chords and natural sounds.
  8. 1/10: The Mimo can vibrate an indefinite number of strings simultaneously.
  9. 1/10: The Whistle is a pocket instrument with a voice-activated vibrating membrane.
  10. 1/10: Closed-mouth humming vibrates its membrane and the air column behind it.
  11. 1/10: Its tuning produces a harmonic series beginning on F or G-sharp.
  12. 1/10: Finger placement and a tuning button change notes.
  13. 1/10: The tuning button raises the fundamental pitch by a perfect fourth.
  14. 1/10: Specific techniques produce diatonic scales such as C, G, and D major.
  15. 1/10: Advanced playing combines the button with humming, whistling, singing, speaking, shouting, and other vocalizations.
  16. 1/10: The Fingerphone resembles a telephone.
  17. 1/10: Its wooden mouthpiece has one lower hole and two upper holes.
  18. 1/10: Its metal box-shaped body has a four-sided wooden keyboard.
  19. 1/10: Strings connect keys to holes and to other keys.
  20. 1/10: Blowing into the mouthpiece vibrates a reed.
  21. 1/10: Fingers cover lower holes to control airflow.
  22. 1/10: Keyboard keys open upper holes, releasing air to produce notes.
  23. 1/10: Different keys control different holes; string tension determines pitch.
  24. 1/10: Moving hands upward or downward corresponds to higher or lower vocal pitches.
  25. 1/10: The Bloom, described as invented in 1978, produces music through resonance.
  26. 1/10: Its hollow, table-like body has a round opening covered by stretched hide.
  27. 1/10: The performer strikes the hide with one or two clappers.
  28. 1/10: The clapper vibrates in resonance with the hide.
  29. 1/10: Striking multiple locations produces multiple pitches.
  30. 1/10: Its tones range from deep and low to high and shrill.
  31. 1/10: Striking particular points produces two frequencies simultaneously, forming modes such as Dorian.
  32. 1/10: It supports different modes, two-tone harmonies, and improvisation.
  33. 1/10: The Moe-ster is a portable, keyboard-free instrument attributed to Richard K. Moe.
  34. 1/10: It can produce sounds ranging from single notes to orchestral pieces.
  35. 1/10: Controls include pitch, rhythm, volume, and effect switches, four directional buttons, and a home button.
  36. 1/10: A radio-like selector controls pitch; harder presses cause larger changes.
  37. 1/10: Activating the rhythm switch lets button presses generate rhythms.
  38. 1/10: A radio-like dial controls volume.
  39. 1/10: An effect control includes pitch memory for recording and replaying note sequences.
  40. 1/10: Compact size, absence of strings or reeds, and instant sound changes distinguish it.
  41. 1/10: An instrument could be assembled simply from boxes, speakers, and duct tape.
  42. 1/10: Planned improvements include a better keyboard and tuning speaker strings to specific notes.
  43. 1/10: The Synth-A-Vibe would be affordable, mass-producible, and attractive to music students.
  44. 1/10: The Electrical Violin-Organ combines violin tone with organ power and volume.
  45. 1/10: An electric motor drives a bow mechanism, while a generator produces string tones.
  46. 1/10: Steel reeds and fine wires vibrate like violin strings, activated by carbon brushes on a rotating disc.
  47. 1/10: A magnet strikes the reeds; a speed dial adjusts volume and duration.
  48. 1/10: Steel reeds simulate bow hairs moving over wires.
  49. 1/10: A spring-loaded lever connected to a piano-like key controls bowing.
  50. 1/10: It can cover the violin’s range, play several notes simultaneously, and perform rapid passages.
  51. 1/10: Future versions would improve tone and use separate hands for bowing and loudness.
  52. 1/10: Potential applications include violin instruction and affordable popular music-making.

Claude

  1. 7/10: Foot-operated bellows push air through copper tubes, changing water levels and enabling continuous pitch bends.
  2. 6/10: The instrument produces shimmering, watery sounds with gliding pitches between conventional piano notes.
  3. 5/10: Blowing through a mouthpiece adds a breathy, flute-like overtone to glass tones.
  4. 4/10: Fingers trigger notes, palms form chords, foot pedals control air jets, and a central dial bends pitches.
  5. 3/10: The Aquaphone is a transparent, tree-shaped instrument with a water-filled central column, twelve curved tubes, resonators, and a circulation pump.
  6. 3/10: Performances incorporate colored lights, moving water, and mist.
  7. 2/10: The Aquaphon is a waist-high instrument with thirteen interconnected, partially water-filled glass chambers arranged in a vertical spiral.
  8. 2/10: Players rub or tap moistened glass chambers to produce sustained tones or bell-like chimes.
  9. 2/10: Players strike tubes with mallets or fingers; striking position changes brightness or depth.
  10. 2/10: The Aquaphon has a semicircular arrangement of interconnected, water-filled glass columns with embedded lights.
  11. 2/10: Sound comes from air jets, touch-sensitive vibration pads, and real-time water displacement.
  12. 2/10: Gestures within an electromagnetic field control sound: vertical movements change pitch, spread fingers add vibrato, swipes create trills or glissandos, and cupped hands muffle it.
  13. 2/10: A foot-operated pump controls water pressure, determining overall volume and intensity.
  14. 1/10: Playing requires gentle coordination of both hands, one foot, and breath.
  15. 1/10: The Aquaphon contains 24 transparent, water-filled tubes whose vibrations floating sensors convert into sound.
  16. 1/10: The Aerowave is a translucent sphere suspended in a metal frame, combining woodwind-like acoustics with electronic controls.
  17. 1/10: Blowing across invisible sensors produces sound; breath intensity controls softness or percussiveness, while breath direction controls register.
  18. 1/10: Moving closer to the sphere produces warmer tones; moving farther away produces thinner, more distant tones.
  19. 1/10: Contact-free playing requires coordinated breathing, hand gestures, and body movement, resembling dance.
  20. 1/10: The Aquaphone is a waist-high instrument directing water through fourteen resonant columns with vibrating membranes.
  21. 1/10: One hand opens and closes column valves to select notes; wider columns produce lower pitches and narrower columns produce higher pitches.
  22. 1/10: A hand-operated tension ring adjusts membranes to bend pitches, create vibrato, and produce slides.
  23. 1/10: Fluid, ethereal sounds combine gurgling, humming, and singing tones to produce melodies and chords.

Sol

  1. 6/10: Using both hands, musicians play melodies and chords while producing a synchronized, colorful light show.
  2. 4/10: The Luminharp has a curved wooden frame containing colored light beams instead of strings, each producing a different note.
  3. 4/10: Foot pedals add effects such as echoes, rhythms, and harmonies.
  4. 3/10: Quick hand movements produce short, bright sounds; slow movements produce long, gentle notes.
  5. 3/10: Players hold the instrument across their lap, plucking strings with one hand while tapping or sliding across panels with the other.
  6. 2/10: Tilting or turning the instrument changes airflow and pitch.
  7. 2/10: Players make music by sweeping, tapping, or waving their hands through light beams.
  8. 2/10: Sensors track hand distance and speed to control pitch and volume.
  9. 2/10: The Luminophone is a small, curved instrument with colorful, touch-sensitive glass panels and thin metal strings.
  10. 2/10: The Starlight Harp is a circular instrument made of glass, metal, and glowing strings that hangs from a stand and rotates.
  11. 2/10: Each light beam produces a distinct note; red beams produce lower notes and blue or violet beams produce higher notes.
  12. 1/10: The Windweaver consists of twelve hollow glass tubes connected by silver strings.
  13. 1/10: Air passing through each tube produces a different note, while the strings create shimmering harmonies.
  14. 1/10: The musician holds it like a steering wheel and blows across the tubes.
  15. 1/10: Plucking silver strings adds rhythms and chords; slow movements create calm melodies and quick spins produce bright cascades.
  16. 1/10: The Echo Bloom resembles a large metal flower with twelve note-producing petals.
  17. 1/10: Special rings tap or slide across its petals, producing bell-, harp-, and glass-like sounds.
  18. 1/10: Blowing into its hollow stem controls volume and adds a wavering effect.
  19. 1/10: Touching petals sequentially produces melodies; pressing several together produces chords.
  20. 1/10: The Luminara is a curved glass harp producing music and colored light.
  21. 1/10: Thin, flexible crystal strings each glow a different color when played.
  22. 1/10: Changing illuminated strings create visual patterns accompanying the performance.
  23. 1/10: Different string colors produce different sounds: blue creates deep, peaceful notes; red or gold creates brighter tones.
  24. 1/10: The musician stands inside the circular frame, plucking, tapping, or brushing strings as they move past.
  25. 1/10: Holding a hand in a beam sustains its tone; rotating the frame changes the overall sound.
  26. 1/10: Playing several beams together makes glass bells vibrate, adding a sparkling echo.
  27. 1/10: Changing lights make the instrument particularly suitable for nighttime concerts.
  28. 1/10: Each string produces a distinct note, and its light color changes with playing volume.

Astra

  1. 4/10: Keys release individual water droplets; taps produce single notes and held keys produce repeated patterns.
  2. 4/10: Tilting a leaf rack makes water cascade between leaves, creating partly unpredictable melodies.
  3. 4/10: A “weather chord” involves blocking all streams, then gradually releasing them so notes enter individually.
  4. 3/10: The Rainloom is a tabletop instrument where falling water strikes twelve tuned bronze leaves above a wooden basin.
  5. 3/10: A foot pedal varies water flow from separate drops to a shimmering downpour.
  6. 3/10: The instrument supports controlled melodies, atmospheric textures, and endings determined by the last falling drops.
  7. 2/10: A foot pedal applies felt dampers, turning sustained shimmer into short, dry ticks.
  8. 2/10: Rocking a bead beside a tongue produces tremolo; tilting the instrument produces loose, rainlike rhythms.
  9. 2/10: The right hand presses strings against a fretboard to change pitch or dampens them into percussive clicks.
  10. 2/10: A crank returns collected beads to their cups and adds a faint rattle.
  11. 1/10: Damping leaves and sliding weighted beads shapes notes and bends pitches.
  12. 1/10: Fingertip gutters divert or block streams, creating notes by opening gaps in silence.
  13. 1/10: A crank rotates ribbons, changing the tone according to where drops land.
  14. 1/10: An overhead reservoir and lower basin recirculate water.
  15. 1/10: The orbit harp uses magnetically guided steel beads rolling across tuned bronze tongues.
  16. 1/10: Magnetic fingertips guide beads in circles for melodies or across rows for cascades.
  17. 1/10: A knee-operated felt lever shortens notes or allows overlapping chords.
  18. 1/10: Bead momentum lets a player launch one melody while beginning another.
  19. 1/10: A lap-sized rainloom uses twelve vertical steel strings and bead-filled cups.
  20. 1/10: Its “clearing” technique activates every string, then progressively dampens them until one remains.
  21. 1/10: An upright Rainloom directs twelve water streams into tuned ceramic bowls.
  22. 1/10: Splitting, shortening, redirecting, cupping, and releasing streams alters the sound.
  23. 1/10: The windloom has twelve metal ribbons vibrated by foot-powered bellows.
  24. 1/10: Pedal pressure controls volume and tone.
  25. 1/10: Sliding wooden bridges changes ribbon pitch and produces glides.
  26. 1/10: Brushing ribbons mutes them; a lever changes airflow sequences.
  27. 1/10: “Unweaving” gradually separates a chord’s pitches by moving individual bridges.
  28. 1/10: A shared air supply means emphasizing one ribbon weakens others, requiring coordinated playing.
  29. 1/10: Another tabletop rainloom drops ceramic beads onto twelve tuned metal leaves.
  30. 1/10: The tide organ has twelve glass pipes in a water-filled rocking basin.
  31. 1/10: Foot bellows supply air, and wooden keys select pipes to produce flute-like notes.
  32. 1/10: Rocking changes water levels and air-column lengths, bending pitches in opposite directions.
  33. 1/10: Different rocking motions produce shimmering chords, vibrato, and sliding melodies.
  34. 1/10: Filling to a marked level and adjusting pipe heights tunes the instrument.
  35. 1/10: “Turning the tide” makes two held notes converge, merge, and separate.
  36. 1/10: The driftloom has an oval frame, twelve strings, sliding ceramic beads, movable stops, and a sealed water chamber.
  37. 1/10: Moving stops changes bead travel, producing tight rhythms or widely spaced accents.
  38. 1/10: Shifting the water chamber changes resonance from warm and muffled to thin and shimmering.

Fable

  1. 3/10: Slow atmospheric tones can become rhythmic through plucking and rapid button attacks; circular breathing sustains drones.
  2. 3/10: Slow attacks and decays create accumulating sounds suited to drone, ambient, and devotional music.
  3. 2/10: The Aelochord uses breath-directed air jets to vibrate strings, without reeds or a conventional resonating air column.
  4. 2/10: Buttons open valves to select notes and chords; released strings fade and overlap.
  5. 2/10: Gentle breath emphasizes high overtones; stronger breath brings out lower harmonics and the fundamental.
  6. 2/10: One foot pump controls how quickly water responds to valves; another pedal abruptly damps seven stems.
  7. 1/10: The left hand presses, slides, vibrates, dampens, or plucks strings against a glass stop-bar.
  8. 1/10: A pedal changes air-jet angles, shifting between pure and turbulent tones.
  9. 1/10: The Thermophone contains temperature-sensitive fluid and twelve heated glass tubes in a copper basin.
  10. 1/10: Uneven heating produces organ-like tones; tube length determines pitch, while heat shapes timbre and volume.
  11. 1/10: Pressing or swirling fluid activates sensors controlling tube heaters.
  12. 1/10: Blowing across openings bends pitches or creates hisses; covering them produces rhythmic cutoffs.
  13. 1/10: Players anticipate thermal delays; two performers can divide fluid control from blowing and damping.
  14. 1/10: A pear-shaped body contains eight fretted steel and bronze strings, a breath tube, and airflow keys.
  15. 1/10: Techniques include gradual note blooms, rhythmic chords, overblown harmonics, valve tremolo, and harmonic stops.
  16. 1/10: The Tidewell has seven hollow glass stems whose water levels determine pitch.
  17. 1/10: One hand bows submerged stems while the other fills or drains chambers for continuous glissandos.
  18. 1/10: Bronze membranes add gong-like undertones; moving water adds gurgling.
  19. 1/10: Players learn pitch control mainly by listening to water levels.
  20. 1/10: The Tidewheel is a horizontal bronze wheel with 24 water basins above a pulsing air reservoir.
  21. 1/10: Pouring water into rotating basins changes pitch and creates wavering hums.
  22. 1/10: Fingertip mallets strike dry basins; wet fingers stroke rims for drones.
  23. 1/10: A bellows pedal swells drones or produces percussive bubble bursts.
  24. 1/10: Rotation moves sounds around the room; sloshing and evaporation shape their evolution.
  25. 1/10: Two performers alternate pouring and striking; performances end by emptying the basins.
  26. 1/10: The Helicant uses a spring-steel ribbon wound into a helix, attached to a wooden resonator.
  27. 1/10: The right hand bows the coil while the left pinches it to control vibrating length and pitch.
  28. 1/10: Bowing height changes timbre; sliding or rocking the stopping hand produces glissando and vibrato.
  29. 1/10: A knee lever compresses the coil to lower pitches; a heel pedal strikes the body for chaotic overtones.
  30. 1/10: Plucking, double stopping, and light bowing produce echoes, dry bells, and breathy whistles.
  31. 1/10: Another Tidewell has twelve hollow glass reeds with brass tongues in a shallow basin.
  32. 1/10: Raising water shortens air columns and raises pitch, allowing octave-wide slides.
  33. 1/10: Hand-controlled cork floats bend individual notes independently of the overall pitch.
  34. 1/10: A mouthpiece manifold and valves select reeds; breath controls volume and overblowing.
  35. 1/10: Two Tidewells can share a siphon, coupling their tuning and requiring coordinated intonation.
  36. 1/10: Freezing, humidity, and transport impose operating and maintenance constraints.
  37. 1/10: The Weftharp has 60–80 chromatically tuned strings controlled by foot-operated heddles.
  38. 1/10: Treadles raise selected strings into notes or chord shapes.
  39. 1/10: A rosined shuttle bows raised strings; shuttle angle controls dynamics.
  40. 1/10: Each pass weaves thread across the strings, progressively damping overtones.
  41. 1/10: Thread materials produce softened, buzzing, or rattling timbres.
  42. 1/10: The finished cloth becomes a physical record of the performance.
  43. 1/10: Notation combines tablature and weaving instructions; irreversible changes favor gradual, one-directional compositions.
  44. 1/10: Two shuttles allow counterpoint; the instrument can accompany singers.
  45. 1/10: The Ombraphone is chest-worn, with seven water-filled glass bells in a tilting brass cradle and rim-directed air jets.
  46. 1/10: Stroking, tapping, or brushing rims creates sustained tones, attacks, chords, and arpeggios.
  47. 1/10: A left-hand lever tilts the cradle, shifting all pitches for bends, portamento, and vibrato.
  48. 1/10: A foot treadle supplies an airy drone whose intensity and tone depend on pumping pressure.
  49. 1/10: Water levels tune the instrument; notation represents tilt with a wavy control line.
  50. 1/10: Its glassy sound ranges from a soft rainlike chorus to an ensemble-cutting tone.
  51. 1/10: Evaporation and water temperature affect tuning, tone, and pitch flexibility.
  52. 1/10: Another Tidewheel is a rotating bronze ring containing oil and reeds, with cups spilling water onto steel plates.
  53. 1/10: Reed drones combine with irregular water percussion that grows denser as rotation accelerates.
  54. 1/10: One pedal controls rotation; another tilts the ring to change exposed reeds and chord color.
  55. 1/10: Breathing through a hub mouthpiece shapes dynamics, with changing results as the wheel rotates.
  56. 1/10: The free hand mutes the rim, triggers splash accents, and bends ringing plates.
  57. 1/10: Compositions follow physical rotation cycles, specifying tilt, treadle pressure, breath, and cup accents.

Gemini 3.5 Flash

  1. 7/10: Foot pedals rotate a brass ring for Doppler-like effects and operate a mechanical damper.
  2. 6/10: Blowing raises pitch, inhaling lowers it, and breath changes produce vibrato and microtonal bends.
  3. 6/10: Sounds combine strings, glass harmonicas, and theremins, ranging from low drones to expressive high melodies.
  4. 5/10: Magnetized fingertip rings brush, tap, hover, or slide near conductive fibers for chords, attacks, and glissandos.
  5. 4/10: Sensors translate interrupted light beams into sounds, responding to interruption location, speed, and density.
  6. 3/10: The Zephyrean Spindle combines rotating strings, friction-based glass sounds, and wind effects.
  7. 2/10: A rotating ring holds radial titanium strings and oil-filled glass spheres, supplied with air by foot bellows.
  8. 2/10: Glassy attacks, synthesized drones, warm resonance, and breathy textures suit cinematic or ambient music.
  9. 1/10: The Halosiphon contains a brass ring, tensioned glass-fiber web, and pressurized resonant core.
  10. 1/10: The Lumina-Chord replaces harp strings with seven colored light beams.
  11. 1/10: The Chiralon is a sphere containing suspended quartz rings, lasers, optical sensors, and pedals.
  12. 1/10: Silver-threaded gloves interact with magnetic fields and interrupt lasers.
  13. 1/10: Touching quartz rings makes them spin and vibrate acoustically.
  14. 1/10: The Gyrochord uses concentric spinning rings, titanium strings, movable counterweights, and a resonating hub.
  15. 1/10: The Nebulaphone combines quartz-rod percussion, laser-lit vapor, and electronic pitch control.
  16. 1/10: A chamber produces mist; twelve laser beams represent pentatonic notes and sensors detect disturbances.
  17. 1/10: Striking, rubbing, or flicking quartz rods produces chimes and friction tones.
  18. 1/10: The Orbilion is a wooden sphere with levitated tuned rings and magnetic finger thimbles.
  19. 1/10: Sweeping gestures magnetically spin rings against acoustic reeds.
  20. 1/10: Bringing fingers closer to the rings increases volume.
  21. 1/10: Wrist flicks disrupt the magnetic field to produce bell-like attacks.
  22. 1/10: Resin-coated thimbles rub silk-and-copper strings, dividing accompaniment and melody between hands.
  23. 1/10: Brass keys change resonator water levels to bend sustained notes.
  24. 1/10: The Spirochord combines a brass sphere, spinning glass cylinder, and gimbal-mounted silver strings.
  25. 1/10: Wrist movements spin the cylinder against resin pads to create a drone.
  26. 1/10: Blowing across reeds controls volume and melodic expression.
  27. 1/10: Tilting changes string tension: forward raises pitch, backward lowers it.
  28. 1/10: Touching strings, glass, and rings produces plucks, vibrato, and damped rhythms.
  29. 1/10: Motion, breath, and touch combine airy whistles with deep resonance.
  30. 1/10: The Vesphone uses liquids and electromagnetic fields for cello-, glass-armonica-, and synthesizer-like sounds.
  31. 1/10: Twelve conductive-oil-filled glass columns contain suspended silver coils and electromagnetic components.
  32. 1/10: Warm, submerged, voice-like tones suit ambient and melancholic music.

Grok 4.5

  1. 10/10: Hands moving near filaments create melodies and harmonies; flicks pluck, sweeps bow, and touch bends pitch or adds overtones.
  2. 8/10: The Aetherharp is a human-height crescent with 24 glowing crystal-glass filaments and a ferrofluid basin.
  3. 7/10: Breath and humming act as control signals for airy textures, pitch scoops, and sympathetic resonance.
  4. 5/10: Foot movements, weight shifts, and magnetic ankle rings control bass, rhythm, percussion, and harmonic shapes.
  5. 5/10: Shoulder-tracking rings map upper-body gestures to dynamics, vibrato, filters, cross-modulation, and arpeggios.
  6. 5/10: Pedals provide sustain, damping, octave shifts, and gesture looping for layered counterpoint.
  7. 4/10: Coordinated hands and feet combine melody, harmony, texture, and wind effects, with luminous mist adding visuals.
  8. 4/10: Glass-harmonica-like tones and synthesized drones support polyphonic ambient or contemporary performances.
  9. 2/10: Gestures and bioelectric presence make performances visually and sonically distinctive.
  10. 2/10: Fluid, dance-like playing suits meditation, improvisation, ambient music, and cinematic scoring.
  11. 1/10: Foot bellows move ferrofluid mist to damp or amplify filaments and control volume.
  12. 1/10: A toe pedal reverses damping into amplification for sudden swells.
  13. 1/10: Pedals or weight shifts adjust vents, moving from dry resonance to spacious reverberation.
  14. 1/10: Hand orientation changes timbre from pure and glassy to warm and reedy.
  15. 1/10: Another Aetherharp uses 24 plasma columns, three resonators, gesture sensors, breath control, and pedals.
  16. 1/10: The Aetherloom is a freestanding hexagon with seven glowing ionized-air columns serving as strings.
  17. 1/10: Pedals control register, overtones, and reverb; a headset maps voice onto instrumental lines.
  18. 1/10: Pressing, waving, or pinching membranes produces glissandos, spectral changes, and percussion.

Kimi 2.5

  1. 5/10: Tuning stones pressed against sympathetic strings create microtonal changes and ghost harmonics.
  2. 4/10: Singing, humming, or blowing into a tube activates sympathetic strings and blends voice with resonance.
  3. 3/10: Standing performance combines bowing, percussion, breath, and body control, resembling fiddles, jaw harps, and singing bowls.
  4. 3/10: Conductive gloves and whole-body gestures control internal sound-producing spheres without touching them.
  5. 2/10: The instrument suits drone meditation, avant-garde classical music, ambient installations, and microtonal improvisation.
  6. 2/10: Pedals control sustain, cross-modulation between wires, and downward pitch bends.
  7. 1/10: The Resonance Loom is chest-worn, with seven main strings, a calfskin membrane, ceramic bridges, and drone strings.
  8. 1/10: A divided three-strand bow enables separate melodies and simultaneous drones.
  9. 1/10: Striking the membrane bounces ceramic bridges to produce a wavering “tremble.”
  10. 1/10: The Sphaera Resonantia contains twelve levitated ceramic-and-iron spheres that excite outer-shell resonators.
  11. 1/10: The left hand guides sphere orbits to control pitch, timbre, intensity, and attack.
  12. 1/10: The right hand launches, clusters, or dampens spheres to control rhythms, chords, silence, and pitch bends.
  13. 1/10: Pedals control electromagnetic force and retune hardwood resonators.
  14. 1/10: The Aquatremor has sixteen sealed glass tubes with adjustable water and air levels.
  15. 1/10: Hydraulic foot pumps change water levels for continuous pitch slides.
  16. 1/10: Fast pedal movements create turbulence; valves introduce bubbling effects.
  17. 1/10: The Crystallon has seven rotating concentric rings of microtonally tuned crystal tubes spanning five octaves.
  18. 1/10: Partially vacuum-sealed tubes sustain resonance for up to forty seconds, producing harmonic “shadow notes.”
  19. 1/10: A hand crank rotates tonal sections and shifts the tonal center.
  20. 1/10: Moist, rosin-treated fingers rubbed across tube openings produce flute-like tones and glissandos.
  21. 1/10: Seven pedals trigger felt hammers for sustained bell-like bass beneath hand-played melodies.
  22. 1/10: Voice-like, metallic, and electronic sounds create microtonal beating textures in humid spaces.
  23. 1/10: Another Resonance Loom is a rotating spherical web of polymer filaments and piezoelectric crystal nodes.
  24. 1/10: Pressing intersections compresses crystal nodes to produce increasingly complex overtones and chords.
  25. 1/10: The Spiridion is a concave wooden bowl with twelve radial silk, gut, and silver-wound strings.
  26. 1/10: It rests on a lap or table and is played with rosin-coated fingertip sheaths.
  27. 1/10: Broad sweeps produce layered chords; changing hand angle isolates melodies.
  28. 1/10: Blowing across the hub adds airy tones; a small hammer adds percussion.
  29. 1/10: Rocking rolls an internal glass bead against the underside for rhythmic accompaniment.
  30. 1/10: The Orbitar has a rosewood sphere surrounded by seven independently rotating rings of different materials.
  31. 1/10: Body movements, mallets, bowed gauntlets, and bare fingers manipulate its sphere and rings.
  32. 1/10: Bowing, striking, and plucking different rings creates melodies resonating through the sphere.
  33. 1/10: Anticipating rotating paths creates a choreographed dance with polyrhythms and shifting acoustic beats.
  34. 1/10: Three pedals bend pitch, add internal drum tones, and microtonally tune a steel ring.
  35. 1/10: Its swirling spatial sound combines gamelan, glass harmonica, and hurdy-gurdy qualities.
  36. 1/10: The Equilibra has a crescent-shaped wooden chamber and 21 weighted bronze tines balanced on ceramic fulcrums.
  37. 1/10: Seated players tap or brush tines with copper thimbles for bell-like attacks and warm sustain.
  38. 1/10: A hollow horsehair aerobow combines bowing and blown air for string-to-flute hybrid tones.
  39. 1/10: Knee-operated felt dampers alternate dry attacks with shimmering decay.
  40. 1/10: The Equilibra favors slow music, hovering tones, bends, and gradual interactions among metal, wood, and air.
  41. 1/10: The Resonance Weaver contains 96 radial wires vibrated by electromagnetic induction.
  42. 1/10: A second hand changes magnetic fields to add overtones and buzzing timbres.
  43. 1/10: A breath tube changes wire tension for three-octave bends and circular-breathing sustain.
  44. 1/10: Its microtonal sound combines cello, glass harmonica, and synthesizer qualities for ambient, maqam, and experimental jazz.
  45. 1/10: Playing coordinates gestures, breath, pedals, and body movement while sensing invisible magnetic fields.
  46. 1/10: The Pneumachord is cello-sized, held vertically, with thirteen strings activated by air-driven mechanical reeds.
  47. 1/10: Breath and a knee-operated wind-chest lever control volume and independent polyphony.
  48. 1/10: A resonance paddle shifts overtones between dark and bright timbres and creates tremolo-like effects.
  49. 1/10: Sustained airy tones combine cello, glass armonica, and panpipe qualities.

Continue with the underlying DFT research.

Read the DFT paper
← All research