SEAT I
EINSTEIN
What will it take to cure cancer?
- What will it take to cure cancer?
- What is the most feasible path to cure Alzheimer’s?
- How do we restore function after complete spinal cord injury?
- Which strategy is closest to a functional HIV cure?
- If we could bet on only three anti-aging technologies this decade, which three?
- What is the most feasible route to commercial fusion power?
Humanity's Hardest Questions, Answered by Collective Superintelligence.
The machines have begun to answer.
The lower bound for contributing to mathematics will now be to prove something that LLMs can’t prove.
— Sir Timothy Gowers, May 2026
- 8080 Years
Erdős’s 1946 unit-distance conjecture — disproved by AI in May 2026.
- 5656 Years Overnight
Two Erdős problems open since 1970 — solved by AlphaProof Nexus overnight.
- 5050 Years Nobel
The protein-folding problem — solved by AlphaFold, Nobel Prize 2024.
It is who gets to ask.
What if the greatest minds could sit in one room again?
For a century, the hardest questions in science waited for minds rare enough — and rooms quiet enough — to hold them.
SEAT II
TURING
SEAT III
DARWIN
For all of history, that room was impossible.It isn't anymore.
INTRODUCING
Pepkio ThinkTank
- Not one model.A panel.
- Not an answer.A deliberation.
- Not a black box.A glass house.
How ThinkTank works
Five frontier models. Three rounds. Full traces.
Round 1 — Independent deep reasoning
The same question is sent to five of the strongest frontier models available (different providers). Each runs independently at its highest reasoning setting.
Round 2 — Cross-critique
Each model reads the others’ Round 1 answers (anonymized), then attacks weak assumptions and missing evidence — and revises its own view.
Round 3 — Final synthesis
One synthesizer model integrates the panel into a single answer: agreements, disputes, and remaining uncertainty. Earlier traces stay intact.
Nothing is hidden. Not even the doubt.
FULL TRACESNO EDITINGOWN YOUR DOUBT
> round=1 panelist=C status=reasoning hypothesis: multi-modal attack surface on residual disease counter: selection pressure under incomplete clearance uncertainty: 0.41 — need adversarial pass on metastasis priors … freezing trace.
The Archive: Questions too big for one mind.
TT-001
Cancer
What is the most likely technical route to definitively conquer cancer?
TT-002
HIV
Which functional-cure strategies deserve the next decade of concentrated capital?
TT-003
Alzheimer's
What is the most feasible multi-target strategy to modify disease progression?
TT-004
Spinal Cord
Which restoration pathway is closest to reversing complete paralysis?
Free to read. Full panel. Full traces.
Your most valuable question is the one closest to you.
One good answer can give you a new direction. Massive leverage.
Don't just observe history. Interrogate your own reality.
Ask your own questionFrequently asked questions
What is ThinkTank?
ThinkTank is a multi-model intelligence platform. Several frontier AI systems independently reason about a hard question, critique one another, then converge on a synthesis — with the full thinking process visible.
How is Explore different from Ask?
Explore shows monumental questions the platform has already run through the full panel. Ask lets signed-in users submit their own questions and run the same multi-round protocol.
Which models participate?
A fixed roster of frontier models from different providers (for example DeepSeek, Claude, GPT, Gemini, and Grok). The roster is reviewed periodically for frontier status.
Are my questions private?
User-generated runs are private by default. You can optionally make a completed run public via a shareable link on tt.<domain>/run/[id].