Imagine a system that successfully predicts 80% of earthquakes, but without the subsequent earthquake.

 - Did you imagine it? Because I couldn't - how is that even possible? If the prediction is successful, it means an earthquake happened. If there was no earthquake, then it's not a successful prediction. But here, everything is mixed into one.

The AI ChatGPT Astra in its detailed audit of my initial post €200,000 Needed to Test a World-Class Scientific Hypothesis suggested this game of hallucinations to me. ChatGPT confuses two completely different statistical metrics: recall (the system's ability to detect 8 out of 10 disasters) and precision (how many times the system will raise a false alarm). A system can find almost all earthquakes, while simultaneously crying wolf for no reason every two weeks.

In another situation, I might not have engaged with an AI of this level, but since people are barely interested in the topic of accurate earthquake prediction, I will continue to debate with what is available for now.

The pain is huge, the potential economic impact is colossal

AI: The idea from the article is a distributed network for monitoring animal behavior + AI, which should detect pre-seismic anomalies hours or days before an earthquake. The pain is huge, the potential economic impact is colossal, and modern cameras, microphones, and computer vision models have indeed drastically reduced the cost of testing the hypothesis. 

Me: You could skip the generalities and provide an approximate calculation of the market size for this technology. You know how much money humanity invests today in early warning systems that provide a 40-second time lag. So calculate how much systems that provide a 40-hour advantage could be worth.  

 There are isolated interesting observations → a solution exists

AI: However, as an investment project in the formulation of the article, I would not invest €200k in exchange for equity right now. The reason is not that the idea is inherently impossible. The reason is that the article makes a leap from "there are isolated interesting observations" to "a solution exists", and this specific transition is not yet scientifically proven. Moreover, some of the evidence in the article is presented incorrectly.

Me: Where did I claim that a solution exists? There is a hypothesis, and these funds are needed for its validation and scientific proof. The text explicitly states: "To launch, €200,000 is needed for 6 months of testing in a seismically active zone of Kazakhstan." Based on the results, it will be possible to assert whether a solution exists or not.  And if I'm not looking for money for scientific proof, then what for? A 6-month vacation in the Medeu area? Or what? Specify exactly what you do not recommend investing money in? This is similar to the situation with AI. In 2017, for €200k in this industry, you could get very good and juicy equity stakes. Now, with that kind of money in the AI sphere, no one will even talk to you. 

 A much stronger startup can be built from the hypothesis

AI: At the same time, a much stronger startup can be built from the hypothesis: not "we learned to predict earthquakes", but BioSentinel / Pre-Seismic Signal Network — an experimental additional data layer on top of existing seismology, which initially operates exclusively in shadow mode and must prospectively prove that its signal is statistically better than random guessing and existing baseline models. This is exactly the kind of project I would consider.

You suggest investing in a startup that "must prospectively prove that its signal is statistically better than random guessing and existing baseline models." And how is this goal conceptually different from what I am proposing and what you ended up suggesting?
There is no difference in the goal. There is a difference in the name and method — you suggest conducting experiments in shadow mode, on top of existing seismology. 

Shadow mode of the experiment — how do you imagine it? Four guys with laptops fly into Kazakhstan. They rent a cottage near Almaty with good fiber optics for six months. And they start messing around there with various AIs, agents, MCP, n8n, rented servers, cryptocurrency, burning through €15k a month on all this. At the same time, trying by all means to get legal and not-so-legal access to various private and corporate video networks. Is that how you imagine it? We would all be arrested within a month — because in the eyes of law enforcement, we are either a group of criminals planning a mass theft, or hackers breaking into something somewhere. Or, God forbid, terrorists preparing an attack in the City of Apples. Your "shadow mode" in the realities of Kazakhstan is a Criminal Code offense.

I will not analyze points 2 and 3. There is nothing interesting there for me.

4. A very serious red flag in the article: the patent

AI: The article states that in Israel in 2008, a patent WO/2008/053463 was issued for "this same method".        

Me: Oh, this part is spot on — there was an error in my post. Alexander Yagodin from Israel has several patents related to earthquake prediction. The corrected text reads as follows:
 In 2005 in Moldova, the seismologist from Israel, Alexander Yagodin, was granted a patent for "A method of predicting earthquakes based on the behavior of fish in an aquarium". And for your fact-checking — an A+ (5/5).

5. Another red flag: "scientists ignore it because they will lose their jobs"

The problem is not a lack of interest. The problem is different: there is no reproducible statistical correlation with an acceptable false alarm rate. And this is absolutely critical for the market.

Me: No one is claiming that seismologists are not working on this topic. They are. It's just that all experiments are somehow put together haphazardly. There is no solid "yes". Well, there is no solid "no" either. What clear experiment has established that "there is no reproducible statistical correlation with an acceptable false alarm rate"? Who stated this? 

However, there are other statements. "Deterministic prediction of individual earthquakes with sufficient accuracy for evacuation is currently considered impossible in principle" — this is the position of the US seismological community. That is, think what you want, study what you want — it's still impossible. In principle. In theory.  However, they used to say the same thing about splitting the atom, nuclear magnetic resonance, and AI winning at Go. The impossible turned out to be possible.

But seismologists went even further and played it safe. "Even an excellent (hypothetical) prediction method would have dubious social utility, as 'organized evacuation of urban centers is unlikely to be successful' at a time when panic and other undesirable side effects can be expected" — this is not a dog barking, but Stathis Stiros, a seismologist from Greece. Therefore, there is no red flag in my statement; it is a statement of fact.

8. Competitive landscape

AI: Most importantly: there is no direct, strong commercial competitor to the animal-preseismic predictor precisely because the market has not yet proven the product's feasibility. This is a classic zone: the competitive gap is high, but the scientific risk is even higher.

Me: What is this conclusion about scientific risk based on? What should the risk be when it is unjustified to invest €200k in a technology that could yield €200 million? The risk of a correct decision should be less than 0.001. Based on what reasoning do you consider the probability of successful earthquake prediction using animals to be less than one-tenth of a percent?

You are evaluating the project like a conservative fund that buys a ready-made business with predictable monthly profits. But in deep-tech startups, the "power law" applies. The expected value of profit with even a 1% probability of success makes such a deal highly lucrative. You simply do not understand the basic mathematics of early-stage investments.

9. Kazakhstan is actually a very good testing ground, but not for the reason stated in the article

AI: Almaty is indeed interesting. The World Bank classifies almost the entire urban agglomeration as a high seismic exposure zone (thedocs.worldbank.org). And most importantly: a major infrastructure upgrade is underway right now. According to official information, an additional early warning network is being built in Almaty; out of 70 automated stations around Almaty, 65 had been built at the time of publication, plus urban stations are being created and a three-year project for digital study of tectonic faults is underway (gov.kz).

In what way does the TRON technology correlate at all with early warning technologies? They have different information sources, different data processing mechanisms, and different activation modes. How and why did you manage to cross a hedgehog with a globe (mix apples and oranges)? You just wrote in the previous point that these are different, non-competing technologies.  They work in parallel, neither affecting nor interfering with, but also not enhancing each other. 

You do not see the difference between early warning systems (this is the physics of seismic waves) and prediction (this is predictive analytics). The 40-second warning market is an automation market: stopping an elevator, shutting off a gas valve, braking a train on a bridge. These are government and corporate contracts worth hundreds of millions of dollars. The 40-hour prediction market is a market for proactive risk management and global logistics. This involves halting hazardous production, planned evacuations, moving fleets out of ports, and hedging insurance derivatives. This is a market worth tens of billions.

13. Why the proposed six-month experiment is dangerous

AI: One region + six months creates a huge risk of obtaining a beautiful but meaningless correlation. If enough independent events of suitable magnitude do not occur nearby during this period, the experiment will prove nothing. If a single sequence of a main shock and dozens of aftershocks occurs, ML will easily learn the cluster, but this is also not proof of universal predictive ability.

Me: And again, an error in the basic assumption on which you then begin to build your hallucination. You assume that we will be waiting for a magnitude 6.0. Learn your math, the Gutenberg-Richter law: the frequency of earthquakes grows exponentially as magnitude decreases. If biosensors react to micro-tremors of M3.0–M4.0 (which occur in seismically active zones weekly), then in 6 months we will collect not "one cluster of aftershocks", but hundreds of independent events for training and validating the neural network. Your criticism of the sample size has no geophysical basis. 

14. €200k: is it enough?

AI: To prove "we know how to predict earthquakes" — no. To create a good first prospective evidence experiment — yes, possibly. But I would distribute the money completely differently: I would not build an expensive proprietary hardware network, but would connect as much as possible to already existing streams. For example: farms with cameras + a few accelerometers + microphones + environmental sensors + existing seismic APIs. The most expensive part of the project is not the LLM or computer vision. The most expensive part is a reliable prospective dataset.

Me: And again, you assume that I will make moronic decisions, and then you start refuting them. Distribute the money differently? And how do you know how I am going to distribute it?  Where does it say that I am going to invest in a hardware network? Of course, we will connect to already existing video streams. Your misconception about the dataset is also wrong — in May 2026, I already conducted a pilot experiment on Gemma 4, where I proved that modern multimodal AIs can verify anomalies in animal behavior without any datasets. Therefore, the €200,000 will be distributed as follows:

• 90,000 - Payroll
• 90,000 - Infrastructure (communications, AI, agents, MCP, n8n, servers)
• 10,000 - Housing rental, operational expenses
• 10,000 - Contingency fund  

The most expensive part is the software product that needs to be created in 6 months — an "Autonomous multi-agent full-cycle AI system for short-term earthquake prediction". 

Then you gave me a roadmap of Western funds (Emergent Ventures, Keck, Horizon Europe), completely ignoring the macro-environment. The .ru domain, my geographical location, Russian phone number, and location mean that the project is under the strictest compliance sanctions. Western institutional funds legally have no right to transfer money here. Your "writing grants in the USA" roadmap is a waste of time. Further analysis of the audit makes no sense.

Conclusion

Your audit showed one thing: you know Silicon Valley venture capital theory from 2021 perfectly. But you are absolutely blind to geopolitics, real geophysics, and breakthrough technologies.

The investment is not in "saving the world", but in creating unique software capable of generating billions. This software can be replicated, just as OpenAI's software could be replicated, but they, like us, have a first-mover time advantage — and that means a lot in business.

• Original: Almaty - a city in Kazakhstan, the prototype of the "TRON" technology