Verification is easier than solving.

solverification

Christopher C. Smith@christophcsmith

The fundamental computer science insight that makes LLMs useful despite the fact that they make mistakes and hallucinate is this: **verification** is computationally easier than **solving**.

1:30 AM · Feb 17, 2026 · 37.7K Views

---

38 Replies · 48 Reposts · 1.05K Likes

The above tweet inspired this post, and reminded me of a quote by Andrej Karpathy:

“I always struggle a bit when I’m asked about the ‘hallucination problem’ in LLMs. Because, in some sense, hallucination is all LLMs do. They are dream machines. We direct their dreams with prompts... most of the time the result goes someplace useful. It’s only when the dreams go into deemed factually incorrect territory that we label it a ‘hallucination’.

I think this insight has a few broad implications:

• Guess and check can be a surprisingly useful strategy when problem-solving.

• LLMs surely are capable of novel insight. A system that can only output established facts likely cannot generate new ideas outside its imagination.

• Verifiable domains (medicine diagnoses, computer science, maths, material science, etc) will get solved within the next 30 years (probably a lot sooner than that). (I’ve also been meaning to write a post on my thoughts on the future of medicine)

• In life, the road to success is nearly always unintuitive and windy. Once achieved, it’s easy to look back and think the path obvious.

  • (this is probably lowkey why I don’t feel like I’ve achieved anything of note in life yet. everything i’ve accomplished feels easy, anyone could do it if they took the same path).

• This is likely why mentorship/tutoring is so successful, and the education/information market is massive ($7.6 trillion per year). It’s trivial for mentors/teachers to distill skills and learnings onto others and help others avoid the same pitfalls they did.

I’ll probably expand on this post in the future, there are heaps of ideas and interesting directions to write about regarding this. Let me know what you guys thought about this insight!