The original URL for this prediction will no longer be available in 11 years (2011)

Original Article ↗ Hacker News Discussion ↗

A 2011 wager on whether a specific Long Bets URL would still serve an HTML page years later prompts a broader look at how fragile web addresses really are. Commenters examine technical risks like HTTP-to-HTTPS changes, JavaScript-heavy sites, redirects, and browser behavior, alongside social and economic factors such as hosting costs, company failures, and the difficulty of preserving content after its creators are gone. The thread also touches on related Long Bets about AI and the Turing test, highlighting how both links and predictions can quietly disappear or be gamed over time.

Bet conditions and HTTP/HTML issues

  • Participants dissect the win condition: entering http://www.longbets.org/601 must yield an HTML document containing specified text.
  • The bet explicitly allows a 301 redirect, which covers HTTP→HTTPS migration.
  • Some argue future browser behavior (blanket HTTP warnings, forced HTTPS) could technically break the condition if no HTML page is shown.
  • There’s debate over whether “contains the text” means the raw HTML response must include it, or whether JavaScript-rendered content in a modern browser is acceptable.
  • Curl vs browser is also ambiguous: some read it as “either condition may work,” others as “both must work.”

URL persistence and web longevity

  • One view: keeping URLs alive is “easy” with tests, redirects, and converting old apps to static HTML; example given of a 24-year-old forum still online.
  • Counterview: the hard parts are organizational—bosses deleting old systems, startups dying, registrars/hosts folding, and upkeep after the owner dies.
  • Ideas emerge for “after-death hosting” or long-term subscriptions; WordPress’s 100-year offerings are cited.
  • Flattening legacy sites to static files is praised as a practical preservation strategy.
  • “Cool URIs don’t change” is referenced as an ideal many fail to meet.

Decentralized / P2P storage alternatives

  • Freenet and newer systems like Safecloud are mentioned as different hosting models.
  • Criticisms of Freenet-style approaches: limited retention if sources vanish, inability to delete data, and poor fit for a mobile-first world with few hobbyist nodes.
  • Stronger mobile connectivity today prompts the question of whether P2P is still infeasible, but technical limits (ports, background execution, roaming IPs, routing complexity) are cited as real obstacles.
  • Some lament the shift from early, experimental decentralized networks to today’s centralized “tech feudalism.”

Status of Long Bets site and specific bets

  • The bet’s URL currently responds with a 301 from HTTP to HTTPS; some are impressed the redirect clause was anticipated.
  • Other bets (e.g., #6) are now offline despite being archived by Wayback; reasons are unclear.
  • Commenters note that at large stakes, participants are strongly incentivized to keep their bet URLs alive.
  • There’s curiosity about how bets are enforced (honor vs mechanism) and about discovering all bets expiring in a given year, but the site doesn’t seem to support that easily.

AI, LLMs, and the Turing Test

  • The original Long Bet about passing the Turing test by 2029 sparks extensive debate.
  • Some argue current LLMs already satisfy loose versions of the test: they can fool some people some of the time, especially with gaming strategies (e.g., roleplaying a teenager or a disordered persona and downplaying knowledge).
  • Others insist they clearly fail: they break on common-sense reasoning, display superhuman recall, fall into unnatural patterns under challenge, and struggle with subtext and long, adversarial conversations.
  • Several stress that the Turing test is ill- or loosely specified and easily “gamed,” so passing it doesn’t strongly evidence intelligence.
  • One position: what matters is indistinguishability to a prepared expert, not merely tricking casual users.
  • Another position: if a system is behaviorally indistinguishable from humans, we should treat it as intelligent/conscious, regardless of internals; a counterargument points to lack of non-text sensory experience (e.g., color qualia) and argues internal mechanisms matter.
  • There’s agreement that current public LLMs are not optimized to claim humanness (for PR/safety reasons), so their true Turing-test potential is unknown.

Miscellaneous observations

  • Commenters are surprised and pleased that old Disqus comments on the bet page still work.
  • Some express nostalgia and frustration with modern web practices (JS-heavy pages, HTTPS everywhere) but concede that the bet’s wording showed notable foresight.