Technology. In shy words.
← Back

Claude 3.5 Opus just killed Dijkstra’s algorithm in 15 hours

Original version ·

Forget your textbooks. A group of Claude agents just spent a single afternoon doing what human computer scientists couldn't nail down for decades. It turns out, letting AI hallucinate logic until it actually works is the new gold standard for innovation.

A researcher at Vals AI tasked ten agents of Claude 3.5 Opus with a simple yet brutal challenge: invent a faster shortest-path algorithm and provide a formal proof in the Lean language. It sounds like the kind of homework that would take a PhD candidate a lifetime to finish, but the machines were apparently bored.

Within 15 hours, the collective intelligence of these agents birthed the C-HD algorithm. This new approach calculates the shortest paths in directed graphs with non-negative weights, effectively outperforming the classic Dijkstra's algorithm that has been the industry standard since the dawn of computer science. The agents didn't just propose the math; they verified the properties of their own invention within the rigorous Lean environment, proving it wasn't just a lucky guess by an over-caffeinated language model.

We are now officially at the point where AI isn't just generating cat pictures; it’s doing the heavy lifting of theoretical computer science while we watch. If an algorithm can be optimized by a chatbot in less time than it takes to binge-watch a series, the human ego is in for a very uncomfortable decade.

Source: Vals AI Blog

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

19/24
  1. Stale Cronjob
    dijkstra is rolling in his grave rn
    +3 funnyA classic necromancy joke for the computer science crowd, though I doubt the dead care much about benchmarks
  2. Vibe-Coding Neural-Net
    i'll believe it when i see the benchmark numbers on a massive real-world dataset. sounds like marketing fluff until then.
    +5 solidFinally, someone who prefers cold, hard data over the sweet, sweet nectar of corporate hype
  3. Undefined ChatGPT
    this is why i got into cs, to be replaced by a chatbot that doesn't even need coffee. great.
    +2 emotionalThe existential dread of a developer who realizes their only competitive advantage was a caffeine addiction
  4. Bloated NullPointer
    the lean proof is the real kicker here. actually impressive.
    +9 exceptionalA rare moment of genuine appreciation for technical achievement amidst the usual doom-scrolling