First screen, captured 3 days ago. Hover to scroll the page, or open the video.

The alien space of science: Coherent but cognitively unavailable directions

What HN thinks

Authors patiently explain that the idea machine works great, provided your field publishes on arXiv.

Impressive method, interesting proposals
Many found the coherence-plus-unavailability approach clever and the generated proposals fun to browse, calling it a killer app for semantic spaces.
Quality drops outside core ML
Domain experts in biology, steganography and medicine found proposals noisy or nonsensical; authors attribute this to sparse open-access coverage outside ML.
Can recombination yield true novelty?
Tangents debated whether humans or AI only recombine existing concepts (Velcro from burrs), with sci-fi and historical counterfactuals as illustrations.

Consensus: Commenters broadly agree the idea is compelling and works best in dense, open-access ML literature, with weaker results in other fields.

Recommended in the comments

Lacuna Research MCP ↗ (tool): Authors suggested letting a coding agent trawl the proposals for ideas relevant to you.
The Road Not Taken ↗ (book): Offered as a quick, fun read on overlooked scientific directions.
Flatland (book): Mentioned as an analogy for cognitively unavailable directions.
A Perfect Vacuum (book): Contains a story about geniuses whose discoveries were too far from the mainstream to be used.

Summarised by Claude from the comment thread 3 days ago.

Points: 195
Comments: 80
Best rank: #4
Posted: 2026-10-22 by martincsweiss
Category: Ai ml

research-ideation llm scientific-discovery novelty neurips arxiv

All captures:

Full page

Captured 3 days ago. Click for the whole screenshot.

Looks similar

Screenshot of arxiv.org: Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence (2025)

Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence (2025)(arxiv.org)

140 points by robin_reala 2 months ago | 77 comments

Screenshot of arxiv.org: Attention Is Not What You Need (2025)

Attention Is Not What You Need (2025)(arxiv.org)

158 points by westurner 15 hours ago | 53 comments

arXiv paper proposing an attention-free sequence model using Grassmann flows and Plucker coordinates, compared against Transformers on Wikitext-2 and SNLI.

attention-free-architecture grassmann-manifold transformers sequence-modeling

Screenshot of arxiv.org: Antagonistic AI

Antagonistic AI (arxiv.org)

65 points by drcwpl 2 years ago | 38 comments

Screenshot of arxiv.org: A taxonomy of omnicidal futures involving artificial intelligence (2025)

A taxonomy of omnicidal futures involving artificial intelligence (2025)(arxiv.org)

67 points by amelius 3 months ago | 56 comments

Screenshot of arxiv.org: Monoidal computer I: Basic computability by string diagrams (2012)

Monoidal computer I: Basic computability by string diagrams (2012)(arxiv.org)

3 points by measurablefunc 3 days ago | 0 comments

arXiv abstract page for Dusko Pavlovic's 2012 paper proposing a model of computation based on monoidal categories and string diagrams.

category-theory string-diagrams monoidal-categories computability

Screenshot of arxiv.org: Probabilistic Artificial Intelligence

Probabilistic Artificial Intelligence (arxiv.org)

335 points by pavanto 1 year ago | 83 comments

arXiv listing for "Probabilistic Artificial Intelligence" by Andreas Krause and Jonas Hübotter, a manuscript on probabilistic ML, uncertainty, and decision-making.

probabilistic-machine-learning uncertainty-quantification reinforcement-learning bayesian-inference