Start with 1,000 coins. Flip them all and remove every coin that lands heads: about 500 are left. Flip the survivors and about 250 remain, then 125, then 63. Each round removes half of whatever is still there, so the losses shrink (500, 250, 125) even though each coin's chance never changes. No coin can tell you when it will go, yet the whole crowd follows the curve closely. Radioactive nuclei behave the same way, and one round is one half-life.
What a nucleus actually does is turn into a different, more stable nucleus. In beta decay a neutron becomes a proton and an electron shoots out: carbon-14 (6 protons, 8 neutrons) becomes nitrogen-14 (7 protons, 7 neutrons). In alpha decay the nucleus throws out a helium-4 nucleus, so uranium-238 becomes thorium-234. Gamma decay releases a high-energy photon and leaves the element unchanged. Heating, crushing or dissolving the sample does not measurably change the half-life, because the electrons and bonds that chemistry acts on barely touch the nucleus.
A candle that burns down 1 cm every hour also shrinks over time, but it loses the same amount each hour, so it is linear and really does run out: a 10 cm candle is gone after 10 hours. Decay loses the same fraction each half-life. To tell them apart, compare equal time steps: if you subtract the same amount each step, it is linear; if you multiply by the same factor, it is exponential decay. After 2 half-lives a quarter remains, not zero; after 10 half-lives about 0.1% is still there.
Think of it like this. A sample of radioactive nuclei is like a crowd in which everyone flips a coin each minute and leaves on heads: nobody can say who goes next, yet half of whoever remains is gone every minute. Real nuclei do not wait for a signal, though; each one has a steady chance of decaying at every instant.
Core idea. Every unstable nucleus of a given isotope has the same fixed chance of decaying in a given time, unaffected by heat, pressure or chemistry. So a large sample always loses the same fraction per half-life, and N=N0(1/2)t/T1/2 predicts how much remains.