AISafety.com
  • Reading Group
  • Links
  • Map
  • About

AISafety.com

  • Reading Group
  • Links
  • Map
  • About
Author

deluks

deluks

Links

Exploiting Floating Point Non-Linearity by OpenAI

Nonlinear Computation In Linear Networks by Open Ai – Floating point arithmatic is fundamentally non-linear near the limit of machine precision. OpenAI managed to exploit these non-linear effects with an evolutionary algorithm to achieve much better performance than a normal deep normal network on MNIST.

Exploiting Floating Point Non-Linearity by OpenAI was last modified: September 30th, 2017 by deluks
September 30, 2017 0 comment
0 Facebook Twitter Google + Pinterest
Links

Superintelligence Risk Project: Conclusion

Superintelligence Risk Project: Conclusion by Jeff Kaufman – “I’m not convinced that AI risk should be highly prioritized, but I’m also not convinced that it shouldn’t. Highly qualified researchers in a position to have a good sense the field have massively different views on core questions like how capable ML systems are now, how capable they will be soon, and how we can influence their development.” There are links to all the previous posts. The final write up goes into some detail about MIRI’s research program and an alternative safety paradigm connected to openAI.

Superintelligence Risk Project: Conclusion was last modified: September 17th, 2017 by deluks
September 17, 2017 0 comment
0 Facebook Twitter Google + Pinterest
Links

Learning to Model Other Minds

Learning To Model Other Minds by Open Ai – “We’re releasing an algorithm which accounts for the fact that other agents are learning too, and discovers self-interested yet collaborative strategies like tit-for-tat in the iterated prisoner’s dilemma.”

Learning to Model Other Minds was last modified: September 15th, 2017 by deluks
September 15, 2017 0 comment
0 Facebook Twitter Google + Pinterest
LinksSafetySingularity

Tegmark and Social Control

Tegmarks Book Of Foom by Robin Hanson – Tegmark’s recent book basically described Yudkowsky’s intelligence explosion. Tegmark is worried the singularity might be soon and we need to have figured out big philosophical issues by then. Hanson thinks Tegmark overestimates the generality of intelligence. AI weapons and regulations.

Ideological Engineering And Social Control A by Geoffrey Miller (EA forum) – China is trying hard to develop advanced AI. A major goal is to use AI to monitor both physical space and social media. Supressing wrong-think doesn’t require radically advanced AI.

Tegmark and Social Control was last modified: September 3rd, 2017 by deluks
September 2, 2017 0 comment
0 Facebook Twitter Google + Pinterest
Links

Incorrigibility In CIRL by The MIRI Blog

Incorrigibility In Cirl by The MIRI Blog – Paper. Goal: Incentivize a value learning system to follow shut down instructions. Demonstration that some assumptions are not stable with respect to model mis-specification (ex programmer error). Weaker sets of assumptions: difficulties and simple strategies.

Incorrigibility In CIRL by The MIRI Blog was last modified: September 3rd, 2017 by deluks
September 2, 2017 0 comment
0 Facebook Twitter Google + Pinterest
Uncategorized

More on Dota 2

More On Dota 2 by Open Ai – Timeline of the DOTA-bot’s rapid improvement. Bot Exploits. Physical Infrastructure. What needs to be done to play 5×5.

More on Dota 2 was last modified: August 17th, 2017 by deluks
August 17, 2017 0 comment
0 Facebook Twitter Google + Pinterest
Uncategorized

Dota2 Update

Openai Bots Were Defeated At Least 50 Times – People could play against the openAI Dota bot. Several people found strategies to beat the bot. One of the human victors explains their strategy.

Dota2 Update was last modified: August 13th, 2017 by deluks
August 13, 2017 0 comment
0 Facebook Twitter Google + Pinterest
Uncategorized

Dota2 by OpenAI

Dota 2 by Open Ai – Open AI codes a 1v1 Dota-2 bot that defeaated top players. The bots actions per minute were comparable to many humans. The bot learned the game from scratch by self-play, and does not use imitation learning or tree search. The game involves hidden information and the bot’s strategies were complicated.

Dota2 by OpenAI was last modified: August 12th, 2017 by deluks
August 12, 2017 0 comment
0 Facebook Twitter Google + Pinterest

About Us

About Us

Non-profit organization for ensuring safe Artificial Intelligence

We aim to create public awareness on AI risk, bring people working on AI Safety together for exchanging our ideas, and help to influence business, academia, and politics to work together and take direct and constructive action towards a safe transition to an AI dominated world.

Recent Posts

  • Reading Group Session 93

    April 18, 2018
  • Reading group session 89

    March 22, 2018
  • Reading Group Session 88

    March 14, 2018
  • Reading Group session 87

    March 8, 2018
  • Reading group session 86

    March 1, 2018

Archives

  • April 2018
  • March 2018
  • February 2018
  • January 2018
  • December 2017
  • November 2017
  • October 2017
  • September 2017
  • August 2017

Categories

  • Corrigibility
  • Featured
  • Links
  • Map
  • Organizations
  • Rationality
  • Reading Group
  • Safety
  • Singularity
  • Uncategorized

Popular Posts

  • Map of the AI Safety Community

    September 26, 2017
  • Tegmark and Social Control

    September 2, 2017
  • Reading Group Session 63

    September 6, 2017

Newsletter

Subscribe to our newsletter for staying updated on news about AI safety.

@2017 - AISafety.com. All Right Reserved. Designed and Developed by Volkan Erdoğan.


Back To Top