We give them goals and a set of immutable guardrails.
They can't set goals for themselves.
They can't remove the guardrails. They can only set subgoals towards the goals we set for them.
REGULATION
-
AI Agents Goals Control Framework Immutable Guardrails
By
–
-
Russell’s Teapot: Unfalsifiable Claims and Burden of Proof
By
–
Following Bertrand Russell, if Geoff claimed that there is a solar system-destroying teapot flying between the orbits of Jupiter and Saturn, you could summarily dismiss his claim as obviously false.
But proving him wrong would be essentially impossible.
That is the kind of -
Superintelligent AI: Control, Safety, and Digital Mediation
By
–
But seriously folks, this a short and juicy tirade in which I say:
— Yann LeCun (@ylecun) 25 novembre 2023
(0) there will be superhuman AI in the future
(1) they will be under our control
(2) they will not dominate us nor kill us
(3) they will mediate all of our interactions with the digital world
(4) hence, they will… https://t.co/xFtOBYdNV5But seriously folks, this a short and juicy tirade in which I say:
(0) there will be superhuman AI in the future
(1) they will be under our control
(2) they will not dominate us nor kill us
(3) they will mediate all of our interactions with the digital world
(4) hence, they will -
Circumventing Biden AI Model Training Restrictions Through Parameter Scaling
By
–
the biden executive order put restrictions on models that:
• contain at least 10^9 parameters
• use more than 10^26 floating-point operations my future company will get around these restrictions by simply training a 9,999,999 billion param model comprised of 8192-bit floats -
AI Objectives vs Parental Influence on Children’s Development
By
–
The nice thing about objective-driven AI systems is that we set their objectives and guardrails.
With 10 year-olds, we have a somewhat limited influence on their objectives and guardrails. -
Crypto Industry Leverage: Economic Impact and Regulatory Pressure
By
–
I think many people who think they got a good deal don't have accurate perceptions over what happens over the next two years. Be that as it may, strongest point of leverage is "A lot of Americans lose money and will be upset if you crash the crypto industry. We are its center."
-
Circumventing Biden’s AI Model Parameter and Compute Restrictions
By
–
the biden executive order put restrictions on models that:
• contain at least 10^9 parameters
• use more than 10^26 floating-point operations my future company will get around these restrictions by simply training a 9,999,999 billion param model comprised of 8192-bit floats -
US Legal Framework Gap: AI Models as Creative Adaptations
By
–
There is a case to be made that models fall under database law where these laws exist (which has benefits), but the U.S. is stuck without — and thus the judge has no framework to reason about models as "Adaptations" like you'd think of a Film from a Book or an Album from a Play.
-
Kadrey v. Meta: AI Training Data Infringement Case in Court
By
–
The Kadrey (and Silverman) case v. Meta is still going to court for infringement at training time. Big Tech lawyers want to take this on! Coverage is terrible because this law firm seems to be throwing the cases; couldn't do much worse if they tried… https://
hollywoodreporter.com/business/busin
ess-news/sarah-silverman-lawsuit-ai-meta-1235669403/
… -
Fair Use Legal Framework: Four Factors in AI Copyright Cases
By
–
You know the case is going to court for the remaining claim, right? Plus there is no such thing as Transformative Use: Fair Use has four factors only one of which is transformation. If you can link to those mathematical facts it could help everyone!