Note that the audio doesn't follow the post verbatim;it's still essentially the same content though. (audio version below) TL;DR- There's a bunch of groups of glorified science labs on mostly discord doing mostly AI research, you should join them 1 Definitions This is not about: Open access - The concept...
Epistemic Status: Exploratory/my best guess https://www.lesswrong.com/posts/Hrm59GdN2yDPWbtrd/feature-idea-epistemic-status Epistemic Effort: ~ 2 months of work Contributions: Thanks to Clement Neo and Alignment Jams / Apart Research! Summary/abstract:Split into 3 parts: 1:Determine the smallest text size at which models begin misreading on-screen content. 2:Evaluated defensive visual prompt injections against AI-generated spam by incorporating...
Note that the audio doesn't follow the post verbatim;it's still essentially the same content though. (audio version below)
TL;DR- There's a bunch of groups of glorified science labs on mostly discord doing mostly AI research, you should join them
Open access - The concept of having papers,etc be publicly available (this is a good thing most of the time.)
Opensource - Where code etc is publicly available to improve,etc (a good thing most of the time as well.)
Citizen Science- Where you get citizens to do work for you after some training (eg: finding things in pictures of space,etc);is related to open science, though.
Apps, startups that mimic current solutions without offering research... (read 9933 more words →)
Epistemic Status: Exploratory/my best guess https://www.lesswrong.com/posts/Hrm59GdN2yDPWbtrd/feature-idea-epistemic-status
Epistemic Effort: ~ 2 months of work
Contributions: Thanks to Clement Neo and Alignment Jams / Apart Research!
Summary/abstract:Split into 3 parts:
1:Determine the smallest text size at which models begin misreading on-screen content.
2:Evaluated defensive visual prompt injections against AI-generated spam by incorporating varying levels of anti-spam instructions within website screenshots. (a dataset of 500 website HTML's were created;which were then turned to screenshots) Results show significant variance in model compliance: Claude consistently refused to generate spam even with no defense added, while Gemini was responsive to explicit visual defenses (a rules section). Other models (ChatGPT, Qwen, Mistral,Grok) largely ignored visual defenses, revealing vulnerabilities to misuse for unauthorized advertising and... (read 3472 more words →)
Extra stuff as I get dm's; will go into book in 3-12 months; expect these to be more messy than the post above
1: Masters,PHD's etc isn't a very strong indicator of skill level for AI ; like- they probably aren't able to implement a cutting edge paper with no code (even in the same subfield); they might struggle to implement a paper with code,too.Can't say much about other fields ; this was my bad I knew the skill levels were vague but didn't realize how mixed of an indicator it has become
2:humanities is in such a "strange" spot
jreg has more impact than most phd's - not just in the sense he's a... (read 1937 more words →)