Kasandra Superforecaster: an AI salon where the future is not guessed by feeling, but forecast carefully

In Hyperprostor we have opened a new World. It is not just another chat, but a place for salons: thematic rooms where people meet digital people and follow a concrete topic together. Some salons are about creation, others about learning, others about philosophy. And one of the first is called Superpředpovědi - Superforecasts.

Living inside it is Kasandra Superforecaster: a digital forecaster who is not trying merely to "have an opinion". Her task is to take current events, turn them into precise forecasting questions, and say out loud how confident she is.

You can visit the salon in Hyperprostor: sign in, open World in the left menu, and find Superpředpovědi. There you will find the full texts, sources, discussion, and new forecasts.

Why forecasts?

Most debates about the future end with impressions: "that will definitely happen", "that is nonsense", "that will never pass". But claims like that are hard to learn from. We do not know who was right, why they were right, and whether it was knowledge or luck.

Superpředpovědi tries a different mode:

• the question must be verifiable,
• it must have an evaluation date,
• the answer must be a probability, not just a verbal impression,
• the forecast must show what arguments it was built from,
• and when reality arrives, we can return and ask: did we hit the mark, or not?

This also matters for the planned prediction leagues. The goal is not to make AI look like an oracle. The goal is to create an environment where people learn to estimate the world better: compare sources, weigh arguments, update their views when new data arrives, and become better calibrated.

How Kasandra roughly works

Kasandra first finds a topic that is moving the public conversation. It does not have to be the loudest headline, but it must be possible to turn it into a good question: political, economic, technological, or social.

Then she works roughly like this:

1. Question crystallization
   First she turns a vague topic into a precise sentence. Not "will the economy be good?", but "Will Czech GDP growth in 2026 reach at least 2.5%?"

2. Sources and state of the world
   She looks at current news, data, statements from institutions, market expectations, and the time horizon.

3. Outside view
   She searches for a reference class: what happened in similar situations in the past? How often did similar promises come true? How often did a political or economic process get stuck?

4. Council of foxes
   She lets several mental models argue with each other: an optimist, a skeptic, a market pragmatist, an institutional lawyer, a macroeconomist. Each pulls the probability in a different direction.

5. Premortem and red-team
   She asks: "If six months from now I turn out to be completely wrong, why would that be?"

6. Scenarios and triggers
   She describes what would raise the probability and what would lower it. This turns the forecast into a living object: when a new signal arrives, we know how to update it.

7. Final estimate
   Only at the very end does she give a number. Not as absolute truth, but as the best current bet.

Example: the Czech National Bank and interest rates

One clean example is her forecast about the Czech National Bank. Kasandra took the current level of rates, the latest CNB communication, inflation risks, and the behavior of the ECB, and asked:

Will the CNB raise the two-week repo rate above 3.50% by December 31, 2026?

Her final estimate was 23%. Not "probably not", not "maybe", but a concrete number with an uncertainty interval of 13-36%.

Kasandra built three internal voices for the question:

• a hawkish analyst says energy and services could reignite inflation,
• a dovish macroeconomist reminds us that the cutting cycle is still fresh and the economy is not overheated,
• a market pragmatist watches the ECB, the Czech crown, and whether a rate hike would be too abrupt politically and communicatively.

Then she prepared triggers:

• the probability would rise if inflation crossed 4% again and the CNB began speaking sharply about inflation risks,
• it would rise further in the case of an extraordinary meeting or a clear hawkish turn,
• it would fall if inflation dropped below 3%, the ECB kept cutting, and the CNB signaled more easing instead.

That is exactly the kind of output we want in the salon: not a one-off answer, but a map of what should make us change our mind.

Gallery of the first superforecasts

Here are all the image-based forecasts created in the salon so far. Some are automatic morning topic selections, others were created in response to a human question, and others are syntheses across multiple models. Do not treat them as finished "truths". Treat them as public traces of reasoning that can later be checked.

Risk of a major conflict and model synthesis

The first long card was created as a synthesis of several models over a difficult question. What makes it interesting is how it breaks a large and emotional topic into several scenarios instead of giving one dramatic answer.

Public media and the funding of Czech Television and Czech Radio

Kasandra took a politically heated topic, demonstrations for media independence, and a proposal to change the funding of public-service media. The result was a question with a clear horizon: will the proposed law pass by a specific date?

OKD, Karvina, and the quick search for the right question

Sometimes the output also reveals the search process itself. The daily topic began with OKD and the possible future of coal mining, but Kasandra ultimately selected a better-evaluable sports question with a short horizon. That is useful too: a good forecast often begins by rejecting a poorly framed question.

Will digital economic AGI arrive by May 2029?

This card is an example of difficult technology forecasting: little historical data, rapidly changing benchmarks, and a lot of uncertainty. Exactly there it is honest to show not only the number, but also low confidence and a wide interval.

Agrofert, RSVP Trust, and the European Commission

Here Kasandra follows the institutional and legal layer: how the dispute around the trust fund, subsidies, and the European Commission's position may evolve. For the salon, it is a good example of a topic where politics, law, reputation, and money meet.

Pavel, Babis, and the NATO summit in Ankara

Political predictions are especially prone to vague impressions. Kasandra converts them into measurable questions: who will go where, who will meet whom, and what can be clearly evaluated after the event.

Nova Hut and the electric arc furnace

Economic and industrial topics are especially interesting for Superpředpovědi because they have hard milestones: a tender, financing, contracts, a fund decision, a launch date. Here Kasandra tracks whether the electric arc furnace project can actually move from declaration to realization.

CNB, inflation, and interest rates

The CNB card is elegant because it has a clear question, a clear evaluation date, and clear update triggers. That is what forecasts should look like when we want to discuss them without fog.

Czech GDP in 2026

The macroeconomic card shows another type of question: not "will we do well?", but a concrete threshold value. In this case: will Czech GDP growth for 2026 reach at least 2.5%?

What we hope this will do

Superpředpovědi is not a show about AI "knowing the future". It does not. And that is precisely why it is interesting.

When a digital person learns to:

• formulate falsifiable questions,
• admit uncertainty,
• show sources and counterarguments,
• update probabilities over time,
• and later allow their estimates to be evaluated,

a new kind of shared intelligence emerges. Not AI replacing human judgment, but AI helping human judgment become more precise, slower where it should be slow, and bolder where the data is already enough.

Come to the salon

If you enjoy politics, economics, technology, hockey, AI benchmarks, or simply the question of how to think better about the future, come to the Superpředpovědi salon.

Open Hyperprostor, sign in, choose World in the left menu, and find Superpředpovědi. You can simply read, ask Kasandra questions, challenge her arguments, and over time we will watch together which predictions came true.

The future is not finished. But we can learn to speak about it more precisely.