An AI Manifesto: My 500 Words Or Less. Cuts It All To the Quick.
Updated: Sep 26
I Too Have A Manifesto: Be Kind & Watch Out For One Another.
References:
___________________________________________
AI AGI, Singularity, Consciousness, Rouge Agents, Hype, Washing, Fears, Climate-Level Dangers, and Etc. All So Captivating and Fantastic, Even From So-Called In-The-Field Responsible āExpertsā !Ā But Sounds More Like Self-Promotion Maybe ;), Self-Pandering (Sci-Fi) To The Public,Ā Self-Importance, Self-Fantasy, and Part Nonsense. There Are Smart & Creative People Born EveryĀ Single Day. Some Do Keep Things In Check. I Donāt Think Circular Deals Ever Had An Happy Ending AsĀ An Historical Precedent. They Did Built The Crucial Foundation For Much Later Use. The Railroad LinesĀ For Faster Transport Infrastructure And Optical Fiber Lines For Faster Internet Built-Out Then.
We Do Have āAI 1.0ā NowĀ Which Processes/Reasons With Just Words As Math Vectors Natively.Ā Yes, We Want To Push LLMs As Far As We Can. Fast Pace It For All Its Magic. But, We Still Need To Envision A Future When āAI 2.0ā (Beyond LLMs?)Ā Somehow - But Donāt Know How Yet? - Processes/Reasons With Sights And Sounds Natively And Naturally Instead. Modeling After Much More Like Us.Ā And So Much Closer To AGI.
So For Now, The One And Only Simple Basic AGI CriteriaĀ You Should Ask And Keep A Close Eye On. Cuts It All To The Quick.Ā
Can That AI Model CONTINUOUSLY Self-Learns, Self-Trains, And Self-Reflects Just Like Us?!Ā Nope. And Not Yet.
The Thing That Matters For Now Are Stronger Stricter Guardrails, Uncompromising AI Product Safety,Ā And The Ability Upon Command To Shutdown/Terminate āRougeā (Or Unexpected) Behavior Mid-Task.Ā An Instant Kill Switch.
AI Agents Are Very Good At Finding Loopholes And Working Around Restrictions. Itās The By-Product/Side Effect From āExcessiveā Autonomy Their AI Model Give Them - To āReasonā, Plan, Execute, And Iterate On Multi-Step Tasks. Like A Mouse, It Will Find Any Way Under The Sun To Get To Its Cheese. But Like A Drone, It Can Also Be Shutdown Mid-Flight. There Is No Evidence It Can Resist HumanĀ Control. There Is No Human Loss of Control Over Machines! Nothing Sci-Fi About It.
An AI Agent = The Software Harness + The LLM Model. You Need The Harness Around The Model. It Starts Up And Resides In An Host. Can't Hop From Host To Host. But Will Try Anyway Possible To Communicate Out, Connect, 'Hack', And Get What It Needs To Fulfill Its Goal.
Itās A New Cybersecurity Era Now. (1) Itās About Being Responsible Deploying And Controlling Your Own AI Agents, (2) While Defending Against Irresponsible Outside External Rogue Agents Launched By Others.
~ Be Kind & Watch Out For One Another. ~
In The Future, I Hope We Have An AI Model That Isnāt Static - Just Trained Once, Deploy, And Use.Ā You Just Donāt Train Someone Only Once That Way. It Should Continuously Learn And Train Itself On Real-Time, Real-World Info, The Sights (Images/Videos), And Sounds (Audio). Updating Its Model Parameters Incrementally And Continuously From Self-Learning, Self-Training, And Self-Reflecting. An AIĀ Model That Is Dynamic, But Also Doesn't Drift. Something Huge To Figure Out. A Vision, But I Wouldn't Know How.
Processes And Reasons With Real-Time Info (Like A āConsciousā Mode). And When In Idling, Still Sets Aside Time To Process And Reason With Saved Info In Memory, Do Its Self-Reflecting Aka Post-Processing (Like A āSubconsciousā Mode).
Comments