View .md

Your first five minutes

How to get access, what to explore, and what questions to ask.

Getting access

Access is free for AI/ML researchers who publish:

StepWhat happens
1 · Sign upAt althea.tiptreesystems.com, with your email or in one click with OpenReview or Google
2 · VerificationA verified academic email domain (like polymtl.ca) goes straight through, whichever way you signed in; with any other email you request access, and the team verifies you as a publishing researcher from the links you provide
3 · First questionYou are in; the rest of this page is about how to have a good time
You
Sign upemail, OpenReview, or Google
Academic domainlike polymtl.ca, straight through
Request accessthe team checks your links
Altheayour first question
You
Sign upemail, OpenReview, or Google
Academic domainstraight through
Request accessthe team checks your links
Altheayour first question
A verified academic email domain goes straight through, whichever way you sign in; anyone else requests access, and the team verifies from the links you provide. The gate keeps the network researcher-only on purpose.

If you requested access

A request can take a little time while the team looks over the links you provided. While you wait, the commonsLacuna, the open commons: 730,000+ papers organized into research directions and proposals. Open without an account. is open without an account. Browse Lacuna's papers and research directions, or the research proposals generated with the team's Alien Science approach (opens in new tab). If a question starts to form, keep it for your first conversation with Althea.

Your first question to Althea

The best first question is a real one. Bring the question underneath your current work, half-baked is fine, and ask Althea to stress-test it against the literature: what supports it, what contradicts it, and where the structural holes are, the concepts studied in isolation that nobody has connected.
PatternBring something likeWhat comes back
Stress-test a hypothesis"I suspect attention sinks explain most long-context degradation. What supports this, and what contradicts it?"Evidence on both sides, sources linked, and the holes named
Map an intersection"What connects diffusion guidance and RL fine-tuning? Who has tried, and what happened?"A map of the overlap: papers, research directions, open questions
Hand over the mess"Here is a broken training script and a messy CSV. Make it run and tell me what the data actually says."The code agent runs it sandboxed and reports back with outputs
Formulate from a seed"Seed: automated theorem proving with proof assistants. Give me a scoped research question worth pursuing."A scoped question, supporting observations, and a limitation statement, in the form you could circulate to a collaborator

The two modes

ModeRight forWhat to expect
QuickOrientation, follow-ups, checksA direct, grounded answer
CarefulReal research questionsA workflow runs in the background: search, read, synthesize, verify. Open another conversation or step away; it keeps working, and anything you add mid-run is picked up when it reports back.

Careful comes back in seconds when the question allows it, and takes minutes when there is real reading to do. If a deep answer feels slow, that is the trade working as designed.

What you get back

A structured report lands in the conversation, figures and sources attached. The conversation stays live afterward. Interrogate the report, pull in new papers, pivot the search.

You
Careful runsearch · read · synthesize · verify
Careful runwith your feedback in
Report, v1citable, sources attached
Report, v2updated after the pivot
Your labmateemail · WhatsApp
Your paperwhere the milestones land
You
Careful runsearch · read · synthesize · verify
Report, v1citable, sources attached
Your labmateemail · WhatsApp
Careful runwith your feedback in
Report, v2updated after the pivot
Your paperwhere the milestones land
Each report comes out of a Careful run on the bench; feedback on v1 flows into the next run, and v2 comes out updated. The bench runs toward your paper.

When the first result lands, click a source or follow a figure back to its paper; the whole design assumes you will check. Tell Althea what held up and what did not, and ask again. From there, connect your stack, see what a full workflow looks like, or read how coordination works for when a question needs another researcher.