Blog Posts

2025:
- 03-23: ⁠Adding bits beats AI slop
- 03-09: ⁠Gwern.net large file support
- 02-07: ⁠LLMs can be faster than you think
- 01-02: ⁠Plumbing vs Internet, revisited
2024:
- 12-04: ⁠Weinersmith’s Trajectoid Words
- 12-01: ⁠Borge’s Tlön: Fascist Idealism
- 11-29: ⁠Virtual comments: LLM idea
- 11-23: ⁠‘Winning’ AI arms races: then what?
- 11-15: ⁠The State of Chinese AI
- 11-14: ⁠Scaling ‘diminishing returns’
- 09-27: ⁠Abs-E, or, speak only positively
- 09-01: ⁠Trying pemmican
- 08-24: ⁠LLM acceptable-use policy
- 08-23: ⁠Simulating ‘tail collapse’ in R
- 08-22: ⁠Hardware hedging scaling risks
- 08-21: ⁠You should write more online
- 08-16: ⁠Marvel: humanizing mythology
- 08-07: ⁠What is an ‘AI warning shot’?
- 07-30: ⁠Towards better RSS feeds
- 07-29: ⁠A Gambler’s Verity World
- 07-22: ⁠Midjourney personalization flaws
- 06-24: ⁠Regexps used to be AI
- 06-19: ⁠Is OpenAI OK?
- 06-16: ⁠GPT-3 Semantic Derealization
- 05-12: ⁠‘Tools For Thought’ Failure
- 04-21: ⁠Kids Are Sick, A Lot
- 04-06: ⁠On First Looking Into Tolkien’s Tower
- 03-24: ⁠Hybridizing forums/wikis proposal
- 02-10: ⁠Wolfe’s ‘Fat Magician’: Marx Brother?
2023:
- 12-29: ⁠Things You Can’t Countersignal Online: Good Writing
- 11-12: ⁠Why do writers still underestimate LLMs?
- 10-27: ⁠Little’s Law in the Wild
- 09-30: ⁠Against Caring About Subtle Poisons
2022:
- 05-28: ⁠GPT-3 2^nd Anniversary & Looking Forward 2 Years
- 01-28: ⁠Cleanup When: Before or After?
- 01-23: ⁠PDF Forgeries Are Surprisingly Rare
- 01-02: ⁠Peak Pre-Modern Human Speed
2021:
- 11-09: ⁠Why Dreams Don’t Matter (Morally or Psychologically)
2020:
- 02-09: ⁠Crowdsourcing The Best GPT-2-1.5b Poetry
2019:
- 02-19: ⁠ThisWaifuDoesNotExist.net
2018:
- 10-20: ⁠WBE & DRL: a Middle Way of imitation learning on brains
2015:
- 04-22: ⁠Estimated Cost of DMT Machine Elf Prime Factorization Experiment
2013:
- 07-20: ⁠Spending Writing ‘Weirdness Points’ Wisely
2011:
- 12-18: ⁠I-zombies: The Hard Problem of Unconsciousness

View Full Posts

2025:

⁠Adding bits beats AI slop

⁠Gwern.net large file support

⁠LLMs can be faster than you think

⁠Plumbing vs Internet, revisited
2024:

⁠Weinersmith’s Trajectoid Words

⁠Borge’s Tlön: Fascist Idealism

⁠Virtual comments: LLM idea

⁠‘Winning’ AI arms races: then what?

⁠The State of Chinese AI

⁠Scaling ‘diminishing returns’

⁠Abs-E, or, speak only positively

⁠Trying pemmican

⁠LLM acceptable-use policy

⁠Simulating ‘tail collapse’ in R

⁠Hardware hedging scaling risks

⁠You should write more online

⁠Marvel: humanizing mythology

⁠What is an ‘AI warning shot’?

⁠Towards better RSS feeds

⁠A Gambler’s Verity World

⁠Midjourney personalization flaws

⁠Regexps used to be AI

⁠Is OpenAI OK?

⁠GPT-3 Semantic Derealization

⁠‘Tools For Thought’ Failure

⁠Kids Are Sick, A Lot

⁠On First Looking Into Tolkien’s Tower

⁠Hybridizing forums/wikis proposal

⁠Wolfe’s ‘Fat Magician’: Marx Brother?
2023:

⁠Things You Can’t Countersignal Online: Good Writing

⁠Why do writers still underestimate LLMs?

⁠Little’s Law in the Wild

⁠Against Caring About Subtle Poisons
2022:

⁠GPT-3 2^nd Anniversary & Looking Forward 2 Years

⁠Cleanup When: Before or After?

⁠PDF Forgeries Are Surprisingly Rare

⁠Peak Pre-Modern Human Speed
2021:

⁠Why Dreams Don’t Matter (Morally or Psychologically)
2020:

⁠Crowdsourcing The Best GPT-2-1.5b Poetry
2019:

⁠ThisWaifuDoesNotExist.net
2018:

⁠WBE & DRL: a Middle Way of imitation learning on brains
2015:

⁠Estimated Cost of DMT Machine Elf Prime Factorization Experiment
2013:

⁠Spending Writing ‘Weirdness Points’ Wisely
2011:

⁠I-zombies: The Hard Problem of Unconsciousness