The guy who invented the forgetting curve
Hermann Ebbinghaus was a German psychologist who figured out that most of what you learn, you lose within days if you don't do anything about it. He wasn't studying student memory or clinical amnesia. He was studying raw learning from scratch. That distinction matters more than people usually realize. He developed a completely new methodology for studying something that had always been treated as philosophical hand-waving. Before Ebbinghaus, people talked about memory in vague terms. He decided to test it like a chemist would test a reaction. His most famous contribution is the forgetting curve, which he discovered through experiments where he memorized lists of nonsense syllables — things like "DAX," "BUQ," "YAT" — and then tested himself at different intervals afterward. He found that memory drops off sharply at first and then levels out. The shape of that decline is consistent enough that the curve itself is still referenced in textbooks over a century later. He also introduced the concept of savings, measuring how much faster you relearn something compared to learning it the first time. That was his second major contribution. The spacing effect emerged from these same experiments: distributed practice produces better retention than massed practice, which sounds logical in hindsight but wasn't well established before he demonstrated it empirically.
One thing beginners consistently miss is that Ebbinghaus's nonsense syllables weren't a quirky choice. They were essential. Real words come with existing associations and meanings, which confounds the measurement. By using triconsonant sequences with a vowel sandwiched in between, he minimized pre-existing connections and got cleaner data. The tradeoff is that real-world learning almost never looks like memorizing "FIZ" repeated 40 times. That's the core limitation of his work that most introductory psychology courses gloss over. I ran into this exact gap when I tried to apply the spacing algorithm to a vocabulary project for a language I was teaching myself. The standard Anki SRS implementation uses the SM-2 algorithm, which is directly descended from Ebbinghaus's savings method. Here's the specific problem: Ebbinghaus's forgetting curve assumes a uniform difficulty profile across all items. In practice, some vocabulary items stick immediately and others never really do, no matter how many times you review them. The algorithm treated every item the same initially, which meant I was wasting time on words I already knew cold while barely touching the ones that kept slipping away. The workaround I ended up using was to manually adjust the ease factor on stubborn items rather than letting the algorithm decide. Specifically, I'd bump the interval down by 30 percent on any word I failed twice in a row instead of waiting for the system to catch up. It's a small adjustment but it cut my daily review time from roughly 45 minutes down to about 20 without sacrificing retention. The system was technically working as designed; the design just didn't account for the fact that not all material learns at the same rate.
There's another counter-intuitive point worth mentioning. Ebbinghaus found that it took him roughly 12 to 15 repetitions to memorize a single nonsense syllable pair to perfection. That number feels intuitively right to anyone who's actually tried it, but it directly contradicts the common advice about "seven tries being enough" that circulates in self-improvement spaces. The number of repetitions needed depends heavily on whether you're aiming for immediate recall or long-term retention, which Ebbinghaus measured separately. He distinguished between learning to perfect and learning that lasts, and the gap between those two states is where most people get tripped up. Another overlooked finding is the serial position effect. He noticed that items at the beginning and end of a list were remembered better than those in the middle. This has two components: the primacy effect, where early items get transferred into long-term storage because you have more time to rehearse them, and the recency effect, where later items are still fresh in short-term buffer. When you're designing study sessions, this means your most important material should be placed at the start or end of a session, not buried in the middle. It's a small structural change that makes a measurable difference over time. Let me be clear about where Ebbinghaus's approach falls apart. The forgetting curve is an average across thousands of trials. Individual variation is substantial. Some people retain nonsense syllables significantly better or worse than the curve predicts. His sample size was effectively one person — himself. That's not a criticism so much as a boundary condition. The methods hold up because subsequent researchers replicated the general patterns with actual human subjects, but any application of his findings should account for the fact that the original data came from a single extremely disciplined individual doing this work in isolation over many years.
Get the Full Details
He also had no way to distinguish between different memory systems. Modern neuroscience separates procedural memory, episodic memory, semantic memory, and working memory. Ebbinghaus treated memory as a single unitary capacity. His curve describes the decay of declarative verbal material specifically, not motor skills or emotional conditioning or visual-spatial recall. When people try to apply his forgetting curve to everything from learning a language to mastering a piano piece, they're stretching a model beyond its intended scope. The practical takeaway is straightforward. If you're trying to learn anything that involves factual recall, space your reviews. Don't cram. Use the savings method as a way to measure whether you've actually retained something — if you can relearn it significantly faster than the first time, the memory is there even if you can't currently access it. And if you're building a spaced repetition system yourself, don't treat every item as equally difficult from the start. Adjust manually based on your own performance patterns rather than trusting the default algorithm to sort it out on its own.
