Bit
A bit is the basic unit of information: the amount of knowledge gained by learning the answer to one yes-or-no question whose two outcomes were equally likely. In a computer it is realized as anything that can reliably sit in one of two distinguishable states, conventionally written 0 and 1, and every file, image, program, and network message is ultimately a very large collection of such choices.
The word and the idea come from information theory, not from electronics. Claude Shannon's 1948 paper "A Mathematical Theory of Communication" used the bit as the unit for measuring information and credited John W. Tukey with coining the term as a contraction of "binary digit." Ralph Hartley had proposed a logarithmic measure of information in 1928, and Shannon's contribution was to tie the unit to probability: one bit is the information in a fair coin toss, so an event of probability one half resolves exactly one bit of uncertainty. This is why a bit is not strictly the digit 0 or 1 but the quantity of information that such a digit can carry at best, and why a predictable signal can carry less than one bit per digit sent.
Counting bits means counting how many distinguishable possibilities can be recorded. One bit distinguishes two cases, two bits distinguish four, and n bits distinguish 2^n, so each added bit doubles the number of possibilities. Ten bits allow 1,024 cases, 32 bits allow 4,294,967,296, and 64 bits allow about 1.8 × 10^19. Going the other way, identifying one item out of N equally likely alternatives needs log2(N) bits, so picking one person from a million takes about 20. Because of this doubling, a modest increase in bit count produces an enormous increase in capacity, which is the logic behind why 128-bit and 256-bit cryptographic keys are considered out of reach for brute-force search.
A bit is a physical thing as well as an abstract one, and its two states can be made from almost any robust difference. A transistor circuit holds a bit as a high or low voltage, a DRAM cell as the presence or absence of charge on a tiny capacitor, a flash cell as trapped electrons, a hard disk as the magnetic orientation of a microscopic region, and an optical disc as a pit or a flat land. Using only two well-separated states makes a bit tolerant of noise: a signal that drifts a little is still clearly nearer one level than the other and can be restored. Physics sets a floor on the cost as well: Rolf Landauer argued in 1961 that erasing one bit must dissipate at least k × T × ln 2 of energy, a limit that real hardware sits far above.
Programmers meet bits directly in flags, masks, and bitwise operations. A single bit is the natural way to store a yes-or-no setting, and a collection of them packed into one number can record dozens of options, such as the permission flags on a file. The operations AND, OR, XOR, NOT, and shifts act on every bit position at once, and they are the primitive steps behind checksums, compression, graphics, and encryption. Bits are also the currency of communication speed: network bandwidth and data rates are quoted in bits per second, and the lowercase b in Mbps means bits, not bytes.
Bit counts now appear in the central trade-offs of machine learning and hardware. Neural-network weights were once routinely stored as 32-bit numbers, but 16-bit, 8-bit, and even 4-bit formats are widely used to shrink models and speed up inference, trading precision for memory and energy. Quantum computing generalizes the idea with the qubit, which can be in a superposition of its two states until measured, but reading it out still yields just one classical bit. A common misconception is that a bit has to be a 0 or a 1 inside the computer in some literal sense: those are labels chosen by convention, and what is physically stored is only a distinction between two states.
Read a bit as the smallest answer a system can give — one yes-or-no distinction worth exactly as much information as an even-odds coin toss — whose power comes entirely from how many of them are combined and how they are interpreted.