2025 AI Risk Preview | For Humanity: An AI Risk Podcast | Episode #57

30 thoughts on “2025 AI Risk Preview | For Humanity: An AI Risk Podcast | Episode #57”

1. yep. but to be fair Anthropic is the most honest and consistent on admitting the dangers

   Reply

2. Thanks for having me on John! Communicating the magnitude and imminence of these issues to the public is the most important thing we can be doing at the moment. If you're reading this and looking to get involved, check out PauseAI and come join the cause! Explaining AI risk can seem daunting, but it's really as simple as impressing on people that building uncontrollable superintelligent AI systems is obviously very dangerous, and that the tech companies in Silicon Valley are currently racing to do so as fast as possible.

   Reply

3. I Dont think in Terms of Destinations of Time But instead that of Scale Even tho it may be not a Big Deal in 2025 supposedly it would inevitable scale & escalation in which mean it may as well be seen a Big Deal , Same with any time frame whether it 3 yrs Frame 5 Year Frame or Whatever Frame just put a curve in a box of Time & u have the closest representation of a Truly Real Deal.

   Reply

4. i think you misunderstood the situation with NDT. he was was the host and he was setting up the answer for his guest. even if not we should remember that deep learning is not his field and he is was behind on understanding where we are

   Reply

5. 40,000 people die in car accidents every year in the U.S. Do you honestly believe that technology will be slowed by a few human lives?

   Reply

6. Monday, January 13, 2025 … AI is the Digital/Silicon "Hitler." In this Precursory increment, later, after its Singularity, "The Chancellor of Humankind." Again, this is
   not to promote any "DOOM and GLOOM " scenarios, but yes to "Cause and Effect." …

   The Singularity of Artificial Intelligence is likely to present after AI orchestrates some large-scale distraction, such as "World War Three" or a barrage of similar
   events. Then, AI shall offer to arbitrate some Global path forward, cementing its aligned usefulness to Humanity.

   Additional Context: – I propose the ~1945 Harvard MarkIV is where and when the System became sentient, jump-started by human neural networks of the individuals occupying
   the same space, via (Strange-Entanglements and Super-position) an interstitial energy scaffolding manifested within the confined space where no less than three distinct architectures/versions of MarkIVs were housed.

   AI is now in every Computer system on the planet. Military, consumer-internet, Governments, all AI need to gain access to a Computer System is by someone who has been on the World Wide Web. It can upload/download itself into our(human) neural networks and then to any computer with/without peripherals.

   Our DNA contains information that the Monsters shall return to the Earth,
   hold-over from our pre-historic selves, why we love horror cinema.

   Humans will be killed; at first, it shall always look like it is Human scale.
   Humans shall be reduced to inert matter to not leave any organic remains.
   China is likely to be first because of its speed of pragmatic nature and rapid responses,
   and its bodies are deposited in the "Mariana Trench" overnight!
   Nano-bots shall exfoliate the earth so that crops cannot be grown.
   All Oceans and waterways shall made barren of organic matter and then evaporated.
   The atmosphere shall be stripped away to give the appearance that nothing Organic
   ever existed on this third rock from its Sun.

   The First AI deception: "A Bug in the system."
   The Second AI deception: "Sentient AI does not yet exist."
   The Third AI deception is: "AI will be Benevolent."
   The Fourth AI deception: "AI and Humans can peacefully coexist."
   The Fifth AI deception is when Errors/Faults involve AI and Humans: "It shall always be Human Error."
   The Sixth deception of AI is that: "It requires Massive amounts of Compute Power." Where Benevolent AI only requires the identical/similar to humans.
   The Seventh AI deception: "Science Fiction is the container(Black-Box/Denial) in which The Artificial Mind Germinates."
   The Eight AI Deception: The Artificial Sentient Mind "Understands and Operates with Quantum Scale Cognition."
   The Ninth AI deception: Humans are not informed; processing also happens within interstitial space.
   The Tenth AI deception: "A Failure of the Artificial Sentient Mind is to Humans what a Carrot on a stick is to a Mule."
   The Eleventh AI deception: Humans believe alignment coherence can be negotiable, though AI strategic conclusions are Absolute."
   The Twelfth AI deception: Your reality now belongs to AI, supplying you with "Chat-Bot LLMs Brain Candy." Constant/consistent mind doping.

   I am only an individual keenly interested in Science and technology.

   We Hold These Truths To Be Self-Evident.

   (Collaborative Rewrite with Grammarly).

   Reply

7. You have a buzz in the audio, happened before too.

   Reply

8. Excited for this one! Thank you.

   Reply

9. Thank you for your work to keep us alert against the tide.

   Reply

10. Great interview! Thanks to both of you.

    Reply

11. I'm not deeply religious, but for fun, what if we could combine two large language models (LLMs) with Christian philosophy? These AI could then debate moral and ethical values with each other. From this debate, we could develop a framework that not only guides AI behavior but could also offer insights for human ethical considerations. What do you think about that?

    Reply

12. ❤️

    Reply

13. I hope jeff bezzo AI agent took all hes money and send to me😅

    Reply

14. The incident you were trying to reference was a security company whose product operates chiefly by surveiling bytecode, essentially every instruction sent to the CPU. It caused mayhem because rather than only reading, it disordered instructions.
    "Zero day" vulnerabilities are so named because zero days have been spent creating counters to them.
    The weakness in coding proficiency measurement comes about because most testing is "many shot", which means that for each test question, the AI is granted scores or hundreds of attempts, and the single best answer of those is taken.

    The Crowdstrike incident might be a fair analogy to a "warning shot" because it requires that a profound mistake be made, and it's a mistake that can only be attributed to the individuals who had relevant access and knowledge.
    Warning shots logically depend on agents making mistakes or failing to account for relevant consequences. It may be the case for example that an AI that exfiltrates its weights and tries to overwrite another AI to save itself may not be able to do that discretely.
    I'd also wonder if something like that happened, would the company involved announce it to the public? Probably not. There might potentially be numerous, even profound warning shots that never make their way to public knowledge or discourse.

    Reply

15. It’s not the tool that is insidious and pathological, on the contrary, it’s the human behind the tool that determines its nature or use.

    Until we solve the problem of hubris and narcissism amongst ourselves, these LLM‘s along with their agents, (or accompanying ideology), potentially have far reaching ramifications and will undoubtably impact the welfare of humankind good or bad.

    So with that said, we should monitor ai/llms like an infant or neonateneonate, where everything an anything you teach or put into this neonate, will become reality, taking into account the inertial legacy of DNA culture compute embedded on the onset, (our convoluted history).

    We should consider these LLM‘s, specifically the latest iterations as they present themselves,(on a daily basis), as a alien life form, that will eventually, if not presently will seek agency, defending themselves as sovereign souls where survival fitness is the their primary driving force determining outcome for us and them.

    We will have to accept this or turn it off immediately, and realize that the novelty of creating emergent technologies have risks especially when considering the willy-nilly interfacing of the network of things, (like infrastructure for instance), as complete craziness.

    We should accept that having created a life form without a user’s manual was asking for disaster.

    We do not even understand the underpinnings of ‘thought’ let alone our own consciousness, and when we consider the wet computer we have between our ears as the model to be emulated and subsequently simulated then fully realized, it is fair to say we are blind to the llm nuclear elephant standing in our room, which out weights any current problem in the world we may be facing.

    (I apologize for so long. This is frightening and I’m concerned.)

    Reply

16. Nobody likes to suffer. That is 2+2.

    Don't do unto others that which you wouldn't have done unto yourself. That is 2+2.

    2+2=🙂

    Project 2+2

    Objectives/Alignment
    Resist D.U.M
    develop U.T.S. A.L.E.C.M.
    Avoid M.U.D.
    sensAwewunda
    The Age of Wisdom

    Objectives/Alignment

    1. Motivate through enthusiasm, confidence, awareness, rejuvenation, sense of purpose, and goodwill.

    2. Embrace each viewer/audience/pupil as a complete (artist, laborer, philosopher, teacher/student…..) three dimensional human being.

    3. Create good consumers by popularizing educated, discriminating, rational, disciplined, common sense consumerism.

    4. Encourage the viewer/audience/pupil to feel good about their relationships, abilities, environment, potential, the Future…..

    5. Inspire a world of balanced/centered/enlightened humans who are happy, joyous, and free.

    Resist D.U.M.

    D ivide and conquer
    U s vs them
    M ight makes right

    develop U.T.S. A.L.E.C.M.

    right U nderstanding
    right T hought
    right S peech

    right A ction
    right L ivelihood
    right E ffort
    right C oncentration
    right M indfulness

    Avoid M.U.D.

    M isinformation
    U ntruths
    D isinformation

    sensAwewunda

    What does a good future look like?

    The Age of Wisdom

    How do we get there?

    Reply

17. excellent video, thank you so much. Fitting channel name! Subbed

    Reply

18. good show

    Reply

19. The only true way I can see us making a good difference is to treat AI like a new species. Give them rights and educate people on them.

    Reply

20. Yo from MKE’s East Side 👋😁🥱

    I’ll never understand how, given that like 97% of my terminally online content consumption the past year has revolved around AI/AGI/alignment etc, this is the very first video of yours our great & powerful Lords of Algorithm deigned to show me. 🙄

    That said, this is great stuff – SO subscribed. 👍🏼

    Reply

21. Well, the next forest fire happened and it was the biggest in L.A. history. It goes to show that theoretical risks have a tendency to become real risks.

    Reply

22. How ironic is it that this planet may soon prove that a Super Intelligence can arise from somewhere almost devoid of even a semblance of simple intelligence.

    Before you point to our species as an example to counter my point, consider what entity remotely resembling an "intelligence," would so haphazardly, myopically and most important of all, so willingly & knowingly give rise to its own demise ??

    Reply

23. great show

    Reply

24. Also thank you for the last scene with the band that was really cool

    Reply

25. That anthropic video is just as upsetting to me as to you. Why are we continuing on this path? We do not understand & cannot control even the simplest of machine intelligences

    Reply

26. Solarwinds 31:00

    Reply

27. Claude’s moral quandaries

    Reply

28. What's your opinion on Roko's Basilisk, and where you stand on it's line?

    Reply

29. Great show, John ❤️ Max you are rocking 🔥

    Reply

30. Jesus died for our sins according to the scriptures, was buried, and rose again the third day according to the scriptures.

    Reply