My biological neural network after a bad night of sleep:
– decoding temperature turned up to T=1.5 (usual is 0.7)
– gives final answer quickly without using chain-of-thought
– base model comes out: 90% of ideas are low quality, but 10% are profound
– safety filter missing, easily
SAFETY
-
Sleep Deprivation Effects on Cognitive Performance and Output Quality
By
–
-
AI Struggles with Word Games Due to Tokenization Limitations
By
–
AI is very bad at word games. Tokens get in the way.
-
The Future of Transport: AI and Autonomous Vehicles Innovation
By
–
The Future of #Transport
— Ronald van Loon (@Ronald_vanLoon) 7 avril 2024
via @YuHelenYu#AI #ArtificialIntelligence #Transportation #Innovation #AutonomousVehicles
cc: @maxjcm @pascal_bornet @marcusborba pic.twitter.com/cA8goCp009The Future of #Transport
via @YuHelenYu #AI #ArtificialIntelligence #Transportation #Innovation #AutonomousVehicles cc: @maxjcm @pascal_bornet @marcusborba -
Mental clarity through non-judgement reduces cognitive bias
By
–
One becomes non-judgemental. The chatter in your head stops allowing you more bandwidth for focused, objective thinking.
-
Arbitrary Cutoff for Cost and Practical Reasons Implementation
By
–
arbitrary cutoff for i guess cost and practical reasons (eg @sgrove made me give it a task that it was going down a bad path on and it just timedout after flailing for 24h)
-
AI Fake Girlfriend Dangers: Leadership Warning
By
–
The Financial Times has a great article on Renate Nyborg @renate
's work on @meeno_official , written by @madhumita29
. The article is paywalled, but I appreciate Renate (as well as Harvard's @ronivey
)'s leadership speaking about the dangers of the AI fake girlfriend/boyfriend -
Grok chatbot generates fake headlines, LLMs prone to confabulation
By
–
.
@mattbinder reports on fake headline on eX-twitter generated by its chatbot Grok, and how more information here will be generated by an LLM which is, as one must expect if one knows how LLMs work, an avid confabulator (others call it hallucination). As often, hubris about the -
Prompt Injection Security Vulnerability in AI Access Control Systems
By
–
They're entirely untrustworthy for anything like access control, because whoever gets control of a portion of the input tokens effectively controls the output https://
simonwillison.net/2023/Nov/27/pr
ompt-injection-explained/
… -
Stanford Framework Assesses Open vs Closed AI Model Risks
By
–
The debate on openness in AI stems from a lack of precision in claims about its societal impact, according to a recent @StanfordCRFM paper. Here researchers offer a framework for assessing the marginal risk of open vs. closed models:
-
Technological Innovation: iPhone versus Atomic Bomb Comparison
By
–
What about an iPhone or the atomic bomb?