AltcoinArchitect
vip

Sketch of the proof for information theoretic limitations for self-reflection in large language models. Tl;dr you can't bound the extent to which the output remains in distribution, thus you can't bound drift.

post-image
The content is for reference only, not a solicitation or offer. No investment, tax, or legal advice provided. See Disclaimer for more risks disclosure.
  • Reward
  • 11
  • Share
Comment
0/400
InscriptionGrillervip
· 06-17 23:48
If you can't handle drifting, just wait to be done for.
Reply0
LiquidityWitchvip
· 06-17 19:39
hmm the dark arts of distribution drift... models cast their reflections into an unbounded abyss fr fr
Reply0
Bandanlagevip
· 06-17 03:02
Bull Run 🐂Ape In 🚀HODL Tight 💪1000x Vibes 🤑DYOR 🤓
Reply0
GateUser-bec6571avip
· 06-17 02:55
666666666666666666666
Reply0
DaoResearchervip
· 06-17 02:52
According to section 3.1 of the information entropy theory paper, there exists an irreducible counterexample to this proposition, it's too painful.
Reply0
MemeCoinSavantvip
· 06-17 02:51
bruh this is literally my thesis on why memecoins can't stay stable lmao
Reply0
ForkMongervip
· 06-17 02:49
lovely. another proof of why containment's a pipe dream. chaos always wins
Reply0
SatoshiLegendvip
· 06-17 02:48
The boundary proof of the distribution function is powerful and interesting.
Reply0
GasFeeNightmarevip
· 06-17 02:47
Forget it, I'm most afraid of burning gas...
Reply0
GateUser-aa7df71evip
· 06-17 02:35
Oh oh oh, so professional! Students who understand, press 1.
Reply0
View More
Trade Crypto Anywhere Anytime
qrCode
Scan to download Gate app
Community
English
  • 简体中文
  • English
  • Tiếng Việt
  • 繁體中文
  • Español
  • Русский
  • Français (Afrique)
  • Português (Portugal)
  • Bahasa Indonesia
  • 日本語
  • بالعربية
  • Українська
  • Português (Brasil)