Prompts for Progress<br>AI-assisted research in mathematics and the sciences<br>An archive of attempts to solve research problems with AI.<br>AI-assisted research is accelerating, but its prompts, methods, and outcomes are scattered. Prompts for Progress collects them in context so researchers can study how the work is being done, learn from prior attempts, and plan better-informed work of their own.<br>Private prototype · Research cutoff July 16, 2026<br>Browse prompts and attempts ↓

23 documented records<br>23 tracked problems and collections<br>6 complete reported results<br>2 documented attempts without a result<br>12 locally preserved full prompts

Use the complete archive<br>Clone the repository and work with every prompt at once.

git clone https://github.com/wintercarver/prompts-for-progress.gitWork with the data →<br>Problems, not just announcements<br>Follow one question across every attempt.

Use the problems index to see which approaches have been tried on a question and how their outcomes differ.<br>Open problems index →<br>Documented activity<br>The public record is accelerating.

The timeline shows documented activity using the known run date when available, otherwise the first dated public event. Counts reflect this archive rather than all AI-assisted research.

2023

2024<br>12

2025

2026

Complete resultPartial progressMixed campaignDocumented attempt

Jun 9, 2026Paper and lineage analysis publicly released EinsteinArena agents improve twelve mathematical bounds<br>Jul 8, 2026Peer-reviewed article published with public framework and logs SciExplorer investigates initially unknown physics models<br>Jul 9, 2026Proof and prompt released GPT-5.6 produces a proof of the Cycle Double Cover Conjecture<br>Jul 15, 2026Multi-agent proof attempt began and ended without a proof CDC-style attempt on the Bartnik admissible-extension conjecture<br>Jul 15, 2026Original uninterrupted 148-minute run GPT-5.6 closes a zeroth-order convex-optimization gap<br>Jul 16, 2026Preprint, prompts, chats, and Lean repository shared GPT-5.6 closes a zeroth-order convex-optimization gap<br>Jul 17, 2026Archive audit confirmed public retained solutions, discussions, verifiers, and the 604 certificate EinsteinArena agents improve twelve mathematical bounds

Seed corpus<br>Study the prompts, methods, and outcomes together.

Browse 23 of 23 records. Each full record keeps the prompt close to its context and evidence.

DomainAllastronomy and chemistrybiologychemistryclimate sciencecomputer sciencemathematicsmathematics and physicsphysicsphysics and materials science<br>OutcomeAllComplete resultPartial progressMixed campaignDocumented attempt

01<br>Mixed campaignchemistry · synthetic chemistry and laboratory automation<br>ACRA converts literature procedures into robotic syntheses<br>ACRA generated executable procedures from published chemistry, completed several physical syntheses, and preserved a failed reaction that a chemist also could not reproduce without substantial changes.<br>SystemAutonomous Chemputer Reaction Agents, Chemputer, Opentrons, OpenAI API-based models, exact production version undisclosed<br>Promptfull<br>EvidenceExperimental<br>First dated eventApr 3, 2026<br>Keep in view: This is primarily a literature-reproduction workflow rather than discovery of a new reaction. The system makes explicit best-guess substitutions for ambiguous prose, and some Chemputer-specific packages are not openly downloadable.<br>View full record→

02<br>Mixed campaignphysics and materials science · atomic force microscopy and laboratory automation<br>AILA benchmarks agents on physical microscopy tasks<br>AILA performed real atomic-force-microscopy workflows and evaluated four language models over 300 task instances, exposing code failures, routing errors, and safety-relevant instruction drift.<br>SystemAILA, GPT-4o, GPT-3.5-turbo-0125, Llama-3.3-70B-versatile, Claude-3.5-sonnet-20241022<br>Promptfull<br>EvidenceExperimental<br>First dated eventOct 14, 2025<br>Keep in view: The campaign automates established microscopy workflows rather than establishing a new physical finding. A correct final measurement did not guarantee that the agent followed safe or authorized instructions.<br>View full record→

03<br>Mixed campaignchemistry · solid-state chemistry and battery materials<br>A-Lab GPSS runs 352 spinel-electrolyte experiments<br>A GPT-5 agent proposed and executed 263 samples in a 352-experiment lithium-halide campaign, raising the joint phase-purity and conductivity hit rate while preserving the full low-yield denominator.<br>SystemA-Lab GPSS, GPT-5 with high reasoning effort, Bayesian optimization<br>Promptfull<br>EvidenceExperimental<br>First dated eventNov 3, 2025<br>Keep in view: Humans constrained the search space and handled several physical and analytical steps. The final joint hit rate was 5.33%, so the result is an optimization campaign with many unsuccessful experiments rather than one autonomous breakthrough.<br>View full record→

04<br>Mixed campaignmathematics · combinatorics and number theory<br>Aletheia scans 700 open-labeled Erdős problems<br>A batch campaign narrowed 700...