Peak Performance Has No Ceiling | Generated by AI

Home 2025.11

No, that’s a very common myth, but it’s not correct when we’re talking about actual IOI/IMO/ICPC gold-medal level (top 4–12 in the world).

Let’s break down why the “it gets easier after the first 1000”, “every hard thing becomes easy once you understand it”, and “people just give up or overestimate the difficulty” explanations don’t hold at the extreme tail.

1. The difficulty curve is the opposite of what you describe

The IMO shortlist, IOI day-2 problems, and ICPC World Finals hardest problems are specifically designed so that even the best 100 people in the world will fail most of them on the first try. Every year the problems get harder in absolute terms because the community’s level keeps rising, but the very top problems remain unsolved by 99% of even the elite contestants during the contest.

2. “Once you solve it, it’s not hard anymore” is true for you, but not for the scoreboard

Yes, after you see the solution to (say) IMO 2023 P6 or IOI 2024 “mex” problem, it feels “understandable”.
But during the actual contest, only 0–3 people on the planet solve it in 5 hours.
The gap between “can understand a gold-level solution when explained” and “can invent it from scratch under pressure in a 5-hour window” is enormous and never disappears. Most national-team members can understand every past IMO problem after reading the solution, yet only ~30–50 humans per year can actually invent the solutions to the new ones.

3. Actual data on dropout vs. plateau

Look at Codeforces rating history of people who eventually got IOI gold:

The same in math: if you are not invited to IMO training camp by grade 10–11 in a strong country, your probability of ever getting IMO gold drops to near zero, even with unlimited time.

This is not “they got distracted by startups/girls/money”. Many keep training hard for years after and still plateau.

4. Real quotes from actual gold medalists

5. The transformer analogy actually proves the opposite point

Transformers were invented by a small group of already world-class researchers who had spent 5–15 years mastering the prerequisites.
99.9 % of ML researchers who read the Vaswani et al. paper in 2017 could understand it, but almost none could have invented it from scratch. The same is true for every single major algorithm (A*, FFT, segment trees, HLD, etc.) — once published, thousands learn it, but only one or two people discover it first.

Bottom line

For 99.9 % of people: yes, it gets easier, plateaus, and motivation is the limiting factor.
For the last 0.01 % (IOI/IMO/ICPC gold level), raw problem-solving ability under contest conditions is still the bottleneck, even after 10 000 hours, even when they keep grinding full-time, and even when they want it more than anything else.

Most people who reach national-team level but never get gold do not “give up” — they literally hit a cognitive ceiling relative to the 5–10 freaks who are a tier above them. The gap at the very top is real and does not shrink with more experience; it often grows because the very best improve faster than everyone else.

So no, it’s not like other fields at that extreme level. The talent + early-start component is much more brutal in olympiad math/CS than in almost any other competitive domain except perhaps chess GM at the very highest ratings.


Back

x-ai/grok-4.1-fast

Donate