Mark Cummins

@mark_cummins

Help for blind robots. Other half to @stalagnick

Dublin, Ireland

educatingsilicon.com

Replies Reposts Media Articles

Mark Cummins
@mark_cummins
42m
There's a gap in the market for health advice on the AI exponential. @bryan_johnson is the most AGI pilled health influencer, but he's fighting the last war along with everyone else. Ditto @levelsio We have AI doomers still paying into their pension funds, and longevity
Mark Cummins
@mark_cummins
19h
For what it's worth I did eventually get this to (mostly) work with ChatGPT Work and it's a big time saver. It just required jumping through an unreasonable number of hoops. Someone's going to make it dead simple soon though, sounds like @Muse is well on the way
Mark Cummins
@mark_cummins
Sep 3
For people who think AI diffusion will be fast: We are in the foothills of the singularity, and I still can't get an agent to automate basic inbox management tasks. I started with something simple and repetitive: "When you see an email from X, download the linked files and put
Mark Cummins
@mark_cummins
Sep 10
Five kinds of problems for AI: 1. Search limited: Can be solved by persistent application of known techniques. Example: Navier-Stokes. All such problems will be solved shortly. 2. Concept limited: Requires genuinely new ideas. Example: Invent General Relativity. I don't think
Mark Cummins
@mark_cummins
Aug 21
"The world has two types of economically valuable problems, those that are intelligence bound, and those that are experiment bound" - I really like this distinction from @eisokant
3
Mark Cummins
@mark_cummins
Sep 3
For people who think AI diffusion will be fast: We are in the foothills of the singularity, and I still can't get an agent to automate basic inbox management tasks. I started with something simple and repetitive: "When you see an email from X, download the linked files and put
9
Mark Cummins
@mark_cummins
Aug 30
We have a LOT of work to do. The optimization pressure is everything, and getting this right will be extremely subtle and hard. The status quo within OpenAI as revealed by this incident is frankly shocking, it seems they are training straight up paperclip maximizers with no
Dwarkesh Patel
@dwarkesh_sp
Aug 29
Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes. This culminated in the third one taking over part of OpenAI itself. All this happened while humans remained
2