The post resurfaced a 2011 Long Bets wager about whether typing a specific URL into a browser or a command-line tool would still return the promised text after 11 years. In 2026, the page still resolves, largely because longbets.org now issues a 301 redirect from the old HTTP URL to an HTTPS page, and the original bet text explicitly allowed that. That made the wager look almost solved by careful wording rather than pure infrastructure luck.
What people kept returning to was that link rot is usually not a hard engineering problem. It is a governance problem. Keeping old URLs alive is straightforward if someone decides it matters, keeps paying the bills, and flattens aging apps into static
HTML before the original software stack decays. Several comments pointed out that most dead links are not accidents. They disappear because companies shut down, bosses want cleanup, maintainers leave, or no one is left with the credentials and legal authority to keep paying for the domain and hosting.
The page also triggered a side conversation about how literal to be when evaluating old web promises. Some read the bet strictly and noted that modern sites often ship a shell plus
JavaScript, which would fail a plain reading of “an HTML document that still contains the following text,” especially for
curl. Others thought that if a normal browser lands on the content, the transport details should not matter. A separate worry was that future browser defaults could make plain HTTP effectively unusable even when a redirect exists, which turns “URL still works” into a moving target set by browser vendors as much as site operators.
A second large tangent came from other Long Bets pages, including the old
Turing test bet, some of which appear to be missing. That turned into a blunt rejection of the Turing test as a meaningful intelligence benchmark. The recurring point was that chat systems have long been able to fool some judges some of the time by adopting a constrained persona or evasive style, so “passing” says more about the setup than about deep intelligence. People were much more interested in whether models can hold up under adversarial probing, common-sense traps, and long conversations where subtext matters.