OpenAI and Anthropic Claim AI Has Disproved a 27-Year-Old Computer Science Hypothesis - HackerNoon
By Google News AI ·
In two weeks, a paper by Peter Jonsson, a computer scientist at Linköping University in Sweden, goes up at what WIRED calls “one of the most prestigious conferences in theoretical computer science.” As of Monday, he told WIRED, much of that work is “dead.” OpenAI and Anthropic each claim that an AI model has knocked out a 27-year-old assumption half of the paper stands on, and both companies say their model found the idea on the same day, October 16. WIRED’s headline put the word in quotation marks: decades of computer science research, “irrelevant.” I’m not a complexity theorist, so I read both papers, the reviewers’ open letter, and what the researchers whose work is affected told reporters. What I found is a result that I think is likely real, that would change what researchers know about hard problems, and that makes nothing you run go faster. The puzzle and the bet Picture a row of light switches, each on or off, and a list of rules. Every rule names three switches. One might say switch 1 is on, or switch 4 is off, or switch 9 is on, and it’s satisfied as long as at least one of those is true. The question is whether any setting of the switches satisfies every rule at once. Computer scientists call this 3-SAT, short for satisfiability. Checking a proposed setting is easy: you walk down the rules and look. Finding one is hard, because five switches have 32 possible settings and every switch you add doubles the count. Ten switches have about a thousand settings; 50 have about a quadrillion. Trying every setting is called brute force, and with brute force each new switch multiplies the work by two. People have beaten that since the 1980s. According to Anthropic’s paper, the best methods known before this month cut the multiplier to about 1.3 (1.328 for the best…
Originally published by Google News AI