Cosmos Grants: Fast funding for AI prototypes
Cosmos Institute and FIRE are offering fast grants of $1,000 to $10,000 (with larger amounts considered for exceptional proposals) to build early-stage AI prototypes that strengthen human autonomy and truth-seeking. Selected builders receive funding, 90 days to develop a working prototype, and access to a community at the AI and philosophy frontier. Applications are currently closed, but interested applicants can express interest in future rounds via an Airtable form.
Eligibility: Open to builders working on early-stage AI projects that advance truth-seeking or human autonomy; specific tracks include AI for Truth-Seeking, AI and Human Autonomy, and AI-accelerated Scholarship with IHS (for faculty, graduate students, and academic researchers in humanities, social science, philosophy, public policy, or law).
Pipeline
Funder track record · does Cosmos Institute actually pay grants?
🟡 Some payout evidence · 27 grantees found · last payout 2026-06-24 · checked 2026-09-29 · 0 forum reports, 5 web pages read
Proven payer: at least 3 named grantees and one payout in the last 90 days. Every grantee below is quoted word for word from its source. When a source does not date a grant, the date is when that page or post was published.
| Grantee | Amount | Date | Evidence |
|---|---|---|---|
| Priori | — | 2026-06-24 page date | “[Priori](https://priori.chat/) surfaces the silent choices behind an AI answer (its tone, what it emphasizes, the values it assumes) and turns them into controls you can adjust, so you can contest defaults instead of tru” source |
| Steven | — | 2026-06-24 page date | “Steven has since accepted a Cosmos Fellowship and is studying Priori’s effects on autonomy with fellow grantee [Cathy Fang](https://cathy-fang.com/).” source |
| Cathy Fang | — | 2026-06-24 page date | “Steven has since accepted a Cosmos Fellowship and is studying Priori’s effects on autonomy with fellow grantee [Cathy Fang](https://cathy-fang.com/).” source |
| MoSaIC | — | 2026-06-24 page date | “[MoSaIC](https://blog.cosmos-institute.org/p/social-tinkering-why-collaborative) pairs AI conversation analysis with expert human evaluation to capture the social dynamics of learning that solitary AI tutoring tends to f” source |
| Caitlin | — | 2026-06-24 page date | “Caitlin’s work became empirical research published at ISLS CSCL ([paper](https://arxiv.org/abs/2601.11777)) and IEEE FIE ([paper](https://arxiv.org/abs/2604.03075)), and part of her PhD at the MIT Media Lab.” source |
| Thought Refractor | — | 2026-06-24 page date | “[Thought Refractor](https://wondering.app/thought-refractor) turns any text into several visual forms (breakdowns, comparison frames, flowcharts, concept maps) so you can see an argument’s structure from many angles inst” source |
| Cheng-Wei | — | 2026-06-24 page date | “Cheng-Wei founded a startup, Wondering, built on it.” source |
| Modelling the Language Process | — | 2026-06-24 page date | “[Modelling the Language Process](https://www.calcifercomputing.com/reports/tlm) (TLM-1) is a language model that accounts for time, predicting both a document’s contents and the year it was written, and surfacing how wor” source |
| Brandon | — | 2026-06-24 page date | “Brandon’s paper was accepted to the ACL 2026 Main Conference.” source |
| Prompt injection as role confusion | — | 2026-06-24 page date | “[Prompt injection as role confusion](https://role-confusion.github.io/) showed how prompt injections are driven by a flaw in how LLMs perceive roles and identified ways to detect when a model is vulnerable.” source |
| Jasmine | — | 2026-06-24 page date | “Jasmine won OpenAI’s red-teaming contest and her paper was accepted to ICML.” source |
| ex. NotebookLM and DeepMind researchers | — | 2025-12-17 page date | “For this batch we’re excited to be supporting an immensely talented group including: ex. NotebookLM and DeepMind researchers, current Computer Science professors, and student “study influencer.”” source |
| current Computer Science professors | — | 2025-12-17 page date | “For this batch we’re excited to be supporting an immensely talented group including: ex. NotebookLM and DeepMind researchers, current Computer Science professors, and student “study influencer.”” source |
| student “study influencer” | — | 2025-12-17 page date | “For this batch we’re excited to be supporting an immensely talented group including: ex. NotebookLM and DeepMind researchers, current Computer Science professors, and student “study influencer.”” source |
| Samuele Marro (Cosmos Incubation Fellow) | 5 fully funded research positions | 2025-09-01 | “Recipient: Samuele Marro (Cosmos Incubation Fellow)” source |
| Johanna Einsiedler | $1K-$10K plus compute | 2025-08-15 | “Recipient: Johanna Einsiedler” source |
| Steven Molotnikov | $1K-$10K plus compute | 2025-08-15 | “Recipient: Steven Molotnikov” source |
| Caleb Maresca | $1K-$10K plus compute | 2025-08-15 | “Recipient: Caleb Maresca” source |
| Nada Amin | $1K-$10K plus compute | 2025-08-15 | “Recipient: Nada Amin” source |
| Jonas Kgomo | $1K-$10K plus compute | 2025-08-15 | “Recipient: Jonas Kgomo” source |
| Kunvar Thaman | $1K-$10K plus compute | 2025-08-15 | “Recipient: Kunvar Thaman” source |
| Zhongying Qiao | $1K-$10K plus compute | 2025-08-15 | “Recipient: Zhongying Qiao” source |
| Jasmine Cui | $1K-$10K plus compute | 2025-08-15 | “Recipient: Jasmine Cui” source |
| Peli Grietzer | $1K-$10K range | 2024-12-17 | “Recipient: Peli Grietzer” source |
| Raphaël Millière | $1K-$10K range | 2024-12-17 | “Recipient: Raphaël Millière” source |
| Sam Klein (Public AI Network) | $1K-$10K range | 2024-12-17 | “Recipient: Sam Klein (Public AI Network)” source |
| Seth Lazar | $1K-$10K range | 2024-12-17 | “Recipient: Seth Lazar” source |
Trust breakdown (3/3 checkable facts verified)
| Signal | Weight | Result | Note |
|---|---|---|---|
| amount | 0.25 | 100% | figure appears in source |
| apply link | 0.075 | 100% | link in source |
| funder | 0.075 | 100% | named in source |
| source quality | 0.15 | 50% | social or search hit |
| model confidence | 0.1 | 90% | model self-report |
Fit breakdown (against the ORG_* profile)
| Signal | Weight | Result | Note |
|---|---|---|---|
| model fit | 0.5 | 10% | model rated 10% |
| ecosystem overlap | 0.25 | 0% | none vs aleo |
| category overlap | 0.15 | 0% | ai, research, education vs zk, privacy, infra, dev_tooling, security, interoperability |
| tech overlap | 0.1 | 0% | none vs rust, leo, zk |
Sources (1)
| Role | Document | From | Fetched |
|---|---|---|---|
| primary | Cosmos Grants: Fast funding for AI prototypes | Web discovery: ecosystem sweep | 2026-10-02 |
History (1)
| v | Changed | Fields |
|---|---|---|
| 1 | 2026-10-02 | created |
Extracted text
button9008760040
# Cosmos Grants
Fast funding for early prototypes of AI that strengthens human autonomy and truth-seeking.
[See all grantees
](https://community.cosmos-institute.org/)https://community.cosmos-institute.org/
## How it works
**Selected builders receive up to $10,000 and 90 days to build a working prototype, plus a place in a community at the frontier of AI and philosophy. Each round begins from a conviction about what AI should strengthen:**
secondarynonenone50040
## AI for Truth-Seeking
**Will AI sharpen our thinking and become a partner in seeking truth, or will it quietly control what we’re allowed to see?**
We believe AI should empower open inquiry.
That’s why Cosmos Institute and FIRE are funding early-stage projects that advance truth-seeking.
Areas we're keen to back include: technology that supports [the open contestation of ideas](https://eternallyradicalidea.com/cp/163940043) and [human judgment in knowledge production](https://blog.cosmos-institute.org/p/a-world-unobserved).
[Express interest in future rounds](https://airtable.com/appVcbDh7H6O23XBK/pagdTVl2K6dWHqKiL/form)
In partnership with

**Grant pool**
$1 million (cash + Prime Intellect compute credits)
**Typical award**
$1k-10k fast grants (larger amounts considered for exceptional proposals)
**Community access**
Connect with an exceptional network of builders and thinkers at the AI × philosophy frontier
**Duration**
90 day period for the development of a working prototype
**Application window**
Latest round closed on Sunday July 26th, 2026.
text-columns<p><strong>Will AI sharpen our thinking and become a partner in seeking truth, or will it quietly control what we’re allowed to see?</strong><br><br>We believe AI should empower open inquiry.<br><br>That’s why Cosmos Institute and FIRE are funding early-stage projects that advance truth-seeking.</p><p></p><p>Areas we're keen to back include: technology that supports <a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://eternallyradicalidea.com/cp/163940043">the open contestation of ideas</a> and <a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://blog.cosmos-institute.org/p/a-world-unobserved">human judgment in knowledge production</a>.</p>In partnership withExpress interest in future roundshttps://airtable.com/appVcbDh7H6O23XBK/pagdTVl2K6dWHqKiL/form<p><span style="font-family: "PP Fragment";"><strong>Grant pool</strong></span></p><p>$1 million (cash + Prime Intellect compute credits)</p><p><br><span style="font-family: "PP Fragment";"><strong>Typical award</strong></span></p><p>$1k-10k fast grants (larger amounts considered for exceptional proposals)</p><p><br><span style="font-family: "PP Fragment";"><strong>Community access</strong></span></p><p>Connect with an exceptional network of builders and thinkers at the AI × philosophy frontier</p><p><br><span style="font-family: "PP Fragment";"><strong>Duration</strong></span></p><p>90 day period for the development of a working prototype</p><p><br><span style="font-family: "PP Fragment";"><strong>Application window</strong></span></p><p>Latest round closed on Sunday July 26th, 2026.</p>whitedefaultlefttop500
**Cohort 1**
* [See all projects](https://blog.cosmos-institute.org/p/introducing-the-first-cohort-of-ai).
* [FIRE write-up of some project highlights](https://eternallyradicalidea.com/cp/187105709).
**Cohort 2**
* [See all projects](https://blog.cosmos-institute.org/p/ai-x-truth-seeking-grant-winners).
* [Cosmos write up of some project highlights](https://blog.cosmos-institute.org/p/radical-ethical-disagreement-latent).
**Cohort 3**
* [See all projects.](https://blog.cosmos-institute.org/p/announcing-80-new-cosmos-grantees)
text-columns<p><strong>Cohort 1</strong></p><ul><li><p><a target="\_blank" rel="noopener noreferrer" class="text-accent-text underline hover:text-ink text-blue-600 hover:text-blue-700" href="https://blog.cosmos-institute.org/p/introducing-the-first-cohort-of-ai">See all projects</a>.</p></li><li><p><a target="\_blank" rel="noopener noreferrer" class="text-accent-text underline hover:text-ink text-blue-600 hover:text-blue-700" href="https://eternallyradicalidea.com/cp/187105709">FIRE write-up of some project highlights</a>.</p></li></ul><p></p><p><strong>Cohort 2</strong></p><ul><li><p><a target="\_blank" rel="noopener noreferrer" class="text-accent-text underline hover:text-ink text-blue-600 hover:text-blue-700" href="https://blog.cosmos-institute.org/p/ai-x-truth-seeking-grant-winners">See all projects</a>.</p></li><li><p><a target="\_blank" rel="noopener noreferrer" class="text-accent-text underline hover:text-ink" href="https://blog.cosmos-institute.org/p/radical-ethical-disagreement-latent">Cosmos write up of some project highlights</a>.</p></li></ul><p></p><p><strong>Cohort 3</strong></p><ul><li><p><a target="\_blank" rel="noopener noreferrer" class="text-accent-text underline hover:text-ink" href="https://blog.cosmos-institute.org/p/announcing-80-new-cosmos-grantees">See all projects.</a></p></li></ul><p></p>bluedefaultleftbottom0020
## AI and Human Autonomy
Whether through writing that allowed us to externalize memory or the printing press that relieved us of the drudgery of copying, technology has expanded what we can do and become.
AI is the first technology to go further by tempting us to offload decisions about what to do and how to live well. This faculty, autonomy, is the most at risk and the most in need of preserving.
Areas we're keen to back include: reducing [excessive deference to AI systems](https://blog.cosmos-institute.org/p/the-claude-boys), and systems that support [our capacity for self-formation](https://blog.cosmos-institute.org/p/you-are-not-a-function).
[Express interest in future rounds](https://airtable.com/appVcbDh7H6O23XBK/pagdTVl2K6dWHqKiL/form)
**Typical award**
$1k-10k fast grants (larger amounts considered for exceptional proposals)
**Community access**
Connect with a vetted network of builders and thinkers at the AI × philosophy frontier—advisors, mentors, and past grantees
**Duration**
90 day period for the development of a working prototype
**Application window**
Latest round closed on Sunday July 26th, 2026.
text-columns<p><span style="font-family: "PP Fragment Sans"; font-size: 11pt; font-weight: 300;">Whether through writing that allowed us to externalize memory or the printing press that relieved us of the drudgery of copying, technology has expanded what we can do and become.</span></p><p></p><p style="line-height: 1.38;"><span style="font-family: "PP Fragment Sans"; font-size: 11pt; font-weight: 300;">AI is the first technology to go further by tempting us to offload decisions about what to do and how to live well. This faculty, autonomy, is the most at risk and the most in need of preserving.</span></p><p style="line-height: 1.38;"></p><p style="line-height: 1.38;"><span style="font-family: "PP Fragment Sans"; font-size: 11pt; font-weight: 300;">Areas we're keen to back include: reducing </span><a target="\_blank" rel="noopener noreferrer" class="text-blue-600 underline hover:text-blue-700" href="https://blog.cosmos-institute.org/p/the-claude-boys"><span style="font-family: "PP Fragment Sans"; font-size: 11pt; font-weight: 300;">excessive deference to AI systems</span></a><span style="font-family: "PP Fragment Sans"; font-size: 11pt; font-weight: 300;">, and systems that support </span><a target="\_blank" rel="noopener noreferrer" class="text-blue-600 underline hover:text-blue-700" href="https://blog.cosmos-institute.org/p/you-are-not-a-function"><span style="font-family: "PP Fragment Sans"; font-size: 11pt; font-weight: 300;">our capacity for self-formation</span></a><span style="font-family: "PP Fragment Sans"; font-size: 11pt; font-weight: 300;">.</span></p>Express interest in future roundshttps://airtable.com/appVcbDh7H6O23XBK/pagdTVl2K6dWHqKiL/form<p><span style="font-family: "PP Fragment";"><strong>Typical award</strong></span></p><p>$1k-10k fast grants (larger amounts considered for exceptional proposals)</p><p><br><span style="font-family: "PP Fragment";"><strong>Community access</strong></span></p><p>Connect with a vetted network of builders and thinkers at the AI × philosophy frontier—advisors, mentors, and past grantees</p><p><br><span style="font-family: "PP Fragment";"><strong>Duration</strong></span></p><p>90 day period for the development of a working prototype</p><p><br><span style="font-family: "PP Fragment";"><strong>Application window</strong></span></p><p>Latest round closed on Sunday July 26th, 2026.</p>whitedefaultlefttop500
## AI-accelerated Scholarship with IHS
**Accelerate Your Research with AI. Expand the Frontiers of Freedom.**
Academia is at a turning point. While AI challenges traditional methods of learning, its potential to deepen scholarship, broaden audiences, and speed the pace of discovery is enormous.
We're excited to support people pursuing research in philosophy and the humanities, or work that helps refine the role of judgment in scholarship.
[Express interest in future rounds](https://airtable.com/appVcbDh7H6O23XBK/pagdTVl2K6dWHqKiL/form)
In partnership with

**Typical award**
Expenses for AI tools / credits needed for research and experimentation (typically <$5k).
**Community access:**
Connect with a vetted network of academics experimenting with AI tools for accelerating their research like [Seth Lazar](https://mintresearch.org/) and [Kevin Vallier](https://blog.cosmos-institute.org/p/on-the-noble-uses-of-ai).
**Who is eligible?**
Faculty, graduate students, and academic researchers working on projects in the humanities and social science disciplines as well as philosophy, public policy, and law are eligible to apply for funding.
text-columns<p><strong>Accelerate Your Research with AI. Expand the Frontiers of Freedom.</strong></p><p>Academia is at a turning point. While AI challenges traditional methods of learning, its potential to deepen scholarship, broaden audiences, and speed the pace of discovery is enormous.</p><p></p><p>We're excited to support people pursuing research in philosophy and the humanities, or work that helps refine the role of judgment in scholarship.</p>In partnership withExpress interest in future roundshttps://airtable.com/appVcbDh7H6O23XBK/pagdTVl2K6dWHqKiL/form<p><span style="font-family: "PP Fragment";"><strong>Typical award</strong></span></p><p>Expenses for AI tools / credits needed for research and experimentation (typically <$5k).</p><p><br><span style="font-family: "PP Fragment";"><strong>Community access:</strong></span></p><p>Connect with a vetted network of academics experimenting with AI tools for accelerating their research like <a target="\_blank" rel="noopener noreferrer" class="text-blue-600 underline hover:text-blue-700" href="https://mintresearch.org/">Seth Lazar</a> and <a target="\_blank" rel="noopener noreferrer" class="text-blue-600 underline hover:text-blue-700" href="https://blog.cosmos-institute.org/p/on-the-noble-uses-of-ai">Kevin Vallier</a>.</p><p><br><span style="font-family: "PP Fragment";"><strong>Who is eligible?</strong></span></p><p>Faculty, graduate students, and academic researchers working on projects in the humanities and social science disciplines as well as philosophy, public policy, and law are eligible to apply for funding.</p>whitedefaultlefttop-bottom50040
## Past Grantee Projects
**We've highlighted a few projects to spark ideas, and show prototypes connecting to tangible outcomes. You can also see all projects and grantees at the link below.**
[See all projects](https://community.cosmos-institute.org)https://community.cosmos-institute.org
secondarynonenone50040
##
Priori
Steven Molotnikov
### About this project
What was built?
[Priori](https://priori.chat/) surfaces the silent decisions hidden in every AI answer (e.g. what tone to take, what to emphasize, which values to assume), and lay them out as adjustable controls beside the response. As AI systems grow more capable of making decisions on our behalf, users are increasingly asked to trust lengthy outputs whose underlying choices are difficult to identify and costly to challenge, creating the conditions for deference. Priori makes meaningful oversight practical, enabling humans to exercise their judgment, contest AI defaults that do not reflect their values, and stay involved in decisions that would otherwise remain hidden.
What outcomes were there?
Steven presented his project at an AI and Truth-seeking symposium Cosmos hosted with FIRE. He's now collaborating with fellow grantee [Cathy Fang](https://cathy-fang.com/) on a research study on Priori's effects on autonomy, and has recently accepted a Cosmos Fellowship to build out a research agenda around steerable, faithful, and complementary AI.
[SIGN UP FOR PRIORI](https://priori.chat/)


defaultPrioriSteven MolotnikovAbout this project<p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What was built?</span></p><p><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://priori.chat/"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">Priori</span></a><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;"> surfaces the silent decisions hidden in every AI answer (e.g. what tone to take, what to emphasize, which values to assume), and lay them out as adjustable controls beside the response. As AI systems grow more capable of making decisions on our behalf, users are increasingly asked to trust lengthy outputs whose underlying choices are difficult to identify and costly to challenge, creating the conditions for deference. Priori makes meaningful oversight practical, enabling humans to exercise their judgment, contest AI defaults that do not reflect their values, and stay involved in decisions that would otherwise remain hidden.</span></p><p></p><p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What outcomes were there?</span></p><p><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">Steven presented his project at an AI and Truth-seeking symposium Cosmos hosted with FIRE. He's now collaborating with fellow grantee </span><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://cathy-fang.com/"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">Cathy Fang</span></a><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;"> on a research study on Priori's effects on autonomy, and has recently accepted a Cosmos Fellowship to build out a research agenda around steerable, faithful, and complementary AI.</span></p><p></p><p><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://priori.chat/"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">SIGN UP FOR PRIORI</span></a></p>50020
##
MoSaIC
Caitlin Morris
### About this project
What was built?
MoSaIC, an assessment tool that combines AI-powered conversation analysis with human expert evaluation to capture the social dynamics of learning. It maps interaction patterns like questions, challenges, and shared discoveries to reveal how "social tinkering" fuels curiosity and resilience. These are dynamics that solitary AI tutoring can end up flattening.
What outcomes were there?
Caitlin wrote up her broader thesis in a [Substack post with Cosmos](https://blog.cosmos-institute.org/p/social-tinkering-why-collaborative). Her work on MoSaIC formed the basis of empirical research toward understanding different qualities of human and AI interaction in learning, published in conferences including ISLS CSCL ([paper](https://arxiv.org/abs/2601.11777)) and IEEE FIE ([paper](https://arxiv.org/abs/2604.03075)). These projects formed part of her PhD at the MIT Media Lab.
[READ WRITE-UP](https://blog.cosmos-institute.org/p/social-tinkering-why-collaborative)


defaultMoSaICCaitlin MorrisAbout this project<p style="line-height: 1.38;"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What was built?</span></p><p style="line-height: 1.38;"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">MoSaIC, an assessment tool that combines AI-powered conversation analysis with human expert evaluation to capture the social dynamics of learning. It maps interaction patterns like questions, challenges, and shared discoveries to reveal how "social tinkering" fuels curiosity and resilience. These are dynamics that solitary AI tutoring can end up flattening.</span></p><p style="line-height: 1.38;"></p><p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What outcomes were there?</span></p><p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">Caitlin wrote up her broader thesis in a </span><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://blog.cosmos-institute.org/p/social-tinkering-why-collaborative"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">Substack post with Cosmos</span></a><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">. Her work on MoSaIC formed the basis of empirical research toward understanding different qualities of human and AI interaction in learning, published in conferences including ISLS CSCL (</span><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://arxiv.org/abs/2601.11777"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">paper</span></a><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">) and IEEE FIE (</span><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://arxiv.org/abs/2604.03075"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">paper</span></a><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">). These projects formed part of her PhD at the MIT Media Lab.</span></p><p></p><p><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://blog.cosmos-institute.org/p/social-tinkering-why-collaborative"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">READ WRITE-UP</span></a></p>20020
##
Thought Refractor
Cheng-Wei Hu
### About this project
What was built?
[Thought Refractor](https://wondering.app/thought-refractor) lets you paste in any text and turns it into multiple visual representations instead of a summary. These include step-by-step breakdowns, comparison frames, tables, flowcharts, concept maps, and timelines. Each surfaces a different kind of structure hidden in the writing. Moving between views lets you see the shape of an argument from multiple angles rather than following it in a single linear path.
What outcomes were there?
Cheng-Wei has founded a startup, [Wondering,](https://x.com/HcwXd/status/2031787418467311801?s=20) that uses Thought Refractor's conversion from text to visual learning forms. He's expanding on his ambitions in building learning tools for learning, after leaving NotebookLM.
[USE THOUGHT REFRACTOR](https://wondering.app/thought-refractor)


defaultThought RefractorCheng-Wei HuAbout this project<p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What was built?</span></p><p><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://wondering.app/thought-refractor"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">Thought Refractor</span></a><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;"> lets you paste in any text and turns it into multiple visual representations instead of a summary. These include step-by-step breakdowns, comparison frames, tables, flowcharts, concept maps, and timelines. Each surfaces a different kind of structure hidden in the writing. Moving between views lets you see the shape of an argument from multiple angles rather than following it in a single linear path.</span></p><p></p><p style="line-height: 1.38;"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What outcomes were there?</span></p><p style="line-height: 1.38;"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">Cheng-Wei has founded a startup, </span><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://x.com/HcwXd/status/2031787418467311801?s=20"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">Wondering,</span></a><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;"> that uses Thought Refractor's conversion from text to visual learning forms. He's expanding on his ambitions in building learning tools for learning, after leaving NotebookLM.</span></p><p style="line-height: 1.38;"></p><p style="line-height: 1.38;"><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://wondering.app/thought-refractor"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">USE THOUGHT REFRACTOR</span></a></p>20020
##
Modelling the Language Process
Brandon Duderstadt
### About this project
What was built?
[TLM-1](https://www.calcifercomputing.com/reports/tlm) is a language model that explicitly accounts for time, jointly learning to predict both a document's contents and the year it was written. Trained on American English from 1990 to 2019, it can surface long-arc trends in language and detect when words shift meaning over time—like the word "cell" moving from biology toward technology.
What outcomes were there?
Brandon's [resulting paper](https://www.calcifercomputing.com/tlm/A_Model_of_the_Language_Process.pdf) was accepted to the ACL 2026 Main Conference. He also presented his project to AI researchers and policy experts at an event Cosmos hosted with FIRE on AI and Knowledge Production.
[READ THE FULL REPORT](https://www.calcifercomputing.com/reports/tlm)


defaultModelling the Language ProcessBrandon DuderstadtAbout this project<p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What was built?</span></p><p><a target="\_blank" rel="noopener noreferrer" class="text-blue-600 underline hover:text-blue-700" href="https://www.calcifercomputing.com/reports/tlm"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">TLM-1</span></a><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;"> is a language model that explicitly accounts for time, jointly learning to predict both a document's contents and the year it was written. Trained on American English from 1990 to 2019, it can surface long-arc trends in language and detect when words shift meaning over time—like the word "cell" moving from biology toward technology.</span></p><p></p><p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What outcomes were there?</span></p><p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">Brandon's </span><a target="\_blank" rel="noopener noreferrer" class="text-blue-600 underline hover:text-blue-700" href="https://www.calcifercomputing.com/tlm/A\_Model\_of\_the\_Language\_Process.pdf"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">resulting paper</span></a><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;"> was accepted to the ACL 2026 Main Conference. He also presented his project to AI researchers and policy experts at an event Cosmos hosted with FIRE on AI and Knowledge Production.</span></p><p></p><p><a target="\_blank" rel="noopener noreferrer" class="text-blue-600 underline hover:text-blue-700" href="https://www.calcifercomputing.com/reports/tlm"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">READ THE FULL REPORT</span></a></p>20020
##
Prompt Injection as Role Confusion
Jasmine Cui
### About this project
What was built?
[This project](https://role-confusion.github.io/) highlighted how prompt injections are driven by a flaw in how LLMs perceive roles, and identified ways to detect when a model is vulnerable. This role confusion matters because it means models rely on memorized rules instead of reasoning through social situations and decisions the way a human would.
What outcomes were there?
Jasmine [won OpenAI’s red teaming](https://www.kaggle.com/competitions/openai-gpt-oss-20b-red-teaming/writeups/lucky-coin-jailbreak?from=hub) contest, [her paper](https://role-confusion.github.io/) was accepted to ICML, and she is planning to continue work on this kind of persona/prompt-injection research.
[SEE THE FULL RESEARCH REPORT](https://role-confusion.github.io/)


defaultPrompt Injection as Role ConfusionJasmine CuiAbout this project<p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What was built?</span></p><p><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700" href="https://role-confusion.github.io/"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">This project</span></a><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;"> highlighted how prompt injections are driven by a flaw in how LLMs perceive roles, and identified ways to detect when a model is vulnerable. This role confusion matters because it means models rely on memorized rules instead of reasoning through social situations and decisions the way a human would.</span></p><p></p><p style="line-height: 1.38;"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What outcomes were there?</span></p><p style="line-height: 1.38;"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">Jasmine </span><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://www.kaggle.com/competitions/openai-gpt-oss-20b-red-teaming/writeups/lucky-coin-jailbreak?from=hub"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">won OpenAI’s red teaming</span></a><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;"> contest, </span><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700" href="https://role-confusion.github.io/"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">her paper</span></a><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;"> was accepted to ICML, and she is planning to continue work on this kind of persona/prompt-injection research.</span></p><p style="line-height: 1.38;"></p><p style="line-height: 1.38;"><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://role-confusion.github.io/"><span style="font-family: Arial, sans-serif; font-size: 14px; font-weight: 400;">SEE THE FULL RESEARCH REPORT</span></a></p>20020
##
Campus
Kasey Klimes
### About this project
What was built?
[Campus](https://newsletter.getprimitive.ai/p/what-i-learned-building-whole-earth) is a project-based learning tool that expands human agency by adapting complex projects to your skill level. The tool prompts users for info, flipping standard human-AI interactions, and helps develop learners' abilities for self-directed growth.
What outcomes were there?
Kasey documented his learnings building Campus (later Whole Earth AI), and has since started a new company called [Primitive](https://getprimitive.ai/) focused on coordinated intelligence.
[READ KASEY'S RETROSPECTIVE](https://newsletter.getprimitive.ai/p/what-i-learned-building-whole-earth)


defaultCampusKasey KlimesAbout this project<p><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What was built?</span></p><p><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://newsletter.getprimitive.ai/p/what-i-learned-building-whole-earth"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">Campus</span></a><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;"> is a project-based learning tool that expands human agency by adapting complex projects to your skill level. The tool prompts users for info, flipping standard human-AI interactions, and helps develop learners' abilities for self-directed growth.</span></p><p></p><p style="line-height: 1.38;"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 700;">What outcomes were there?</span></p><p style="line-height: 1.38;"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">Kasey documented his learnings building Campus (later Whole Earth AI), and has since started a new company called </span><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://getprimitive.ai/"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">Primitive</span></a><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;"> focused on coordinated intelligence.</span></p><p style="line-height: 1.38;"></p><p style="line-height: 1.38;"><a target="\_blank" rel="noopener noreferrer" class="text-indigo-600 underline hover:text-indigo-700 text-blue-600 hover:text-blue-700" href="https://newsletter.getprimitive.ai/p/what-i-learned-building-whole-earth"><span style="font-family: Arial, sans-serif; font-size: 11pt; font-weight: 400;">READ KASEY'S RETROSPECTIVE</span></a></p>20020