scale got us part of the way here. neurosymbolic AI, added in the form of harness, tools, code interpreters etc were a major contributor that people often don’t acknowledge.
@garymarcus
-
Case for Multiple Independent AI Evaluators
By
–
@DKokotajlo makes an obvious point that gets made too rarely: you can have more than one independent AI evaluator. And you should, to balance technical expertise (e.g., human failures vs cybersecurity vs model capabilities), ideology, perspective, etc. (Every grant panel I was
-
Ex-DeepMind researcher testifies on Google AI safety commitments
By
–
Powerful testimony from ex Google DeepMind researcher @Turn_Trout at NYC Hearing on AI, about how he tried to get Google to commit to AI safety, and how Google, sadly including Demis Hassabis, ultimately backed down from earlier commitments to AI safety. So much for “don’t be
-
AI misalignment detection worsening amid industry’s fast-move attitude
By
–
“Our ability to notice misalignment problems is already quite poor and set to get much worse in the near future” “combined with the move fast and break things attitude, the industry is at great risk” – @DKokotajlo
-
NYC hearing: most code now written by AI, poorly vetted
By
–
At NYC hearing, @DKokotajlo echoes @hilbertspaess
’s point that most code is now written by AI and not vetted carefully enough. This is worrisome. -
Debate on unchecked AI-written code and responsibility
By
–
“Majority of code is now written by AI, and people do not check it very carefully anymore. … Handing over that much to the AI is irresponsible” – @hilbertspaess (possibly slightly paraphrased from memory) Fully agree with him on that part.
-
AI control debate at NYC City Council meeting
By
–
Jacob Coxon @hilbertspaess and @DKokotajlo next at the NYC City Council meeting, hosted by @SpeakerMenin
. Coxon cpens with “We do not know how to control any AI system yet”, which, importantly, isn’t actually true: lots of AI systems (AlphaGo, GPS Routing, etc) have no control -
Live testimony at NYC AI hearing on AI policy
By
–
live momentarily on @msnow re today’s NYC AI hearing. my testimony: https://
garymarcus.substack.com/p/coming-soon-
new-york-citys-hearing?r=8tdk6&utm_campaign=post&utm_medium=web
… -
Debate on labeling early AI systems as superintelligence
By
–
So let me get this straight, now we have to call every piece of AI SI, even • ELIZA, 1965 which was just keyword matching and certainly not super or intelligent • Deep Blue, which could play chess but nothing else.
• The original Google search algorithm (which could only do -
Critique of OpenAI’s AGI claims and AI controllability
By
–
A month ago: OpenAI says GPT-6 could be AGI (spoiler alert it wasn’t) Today: OpenAI admits current AI’s uncontrollability is bad, asks you to suck it up.