Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

What's the best explanation for LLMs being almost right when citing sources?

Jun 09, 2024 - news.ycombinator.com
The author compares the current state of "prompt engineering" to a random, trial-and-error process, akin to throwing paint at a wall until the desired result is achieved. They liken it to a stochastic wrench that is adjusted until it serves its intended purpose. The author also discusses the concept of tokens, viewing them as elements in a complex, multi-dimensional sorting process that lose their context or defined domain during evaluation.

The author argues that this process results in an outcome that is less than the sum of its parts, as the weights between tokens only represent linguistic or phonetic relationships. These relationships, according to the author, lose their causal connection to the real world during evaluation. This suggests that the author believes that the current approach to prompt engineering and token use is flawed and disconnected from reality.

Key takeaways:

  • The current state of 'prompt engineering' is compared to a random, trial-and-error process.
  • The author likens the process to a 'stochastic wrench' that is adjusted until it meets the user's specific needs.
  • Tokens are perceived as components of a complex, multi-dimensional sorting process that lose their context during evaluation.
  • The author suggests that the correlation of weights between tokens can result in a loss of their causal relationship to the real world.
View Full Article

Comments (0)

Be the first to comment!