The future of AI by Eric Schmidt

The Future of AI by Eric Schmidt

A Snail:

• IF: companies persist in the AI race,
• THEN: humanity might end up with AI agent capable of autonomous problem-solving and decision-making,,
• BECAUSE: the AI race accelerates technological advancements at a breakneck pace, expanding the context window and enabling AI systems sustained and iterative reasoning, which leads to a reduced oversight of their work from humans.

• I recently watched an interview with Eric Schmidt. My overall impression? The usual political rhetoric—China and Russia are to blame, and there’s no free speech. Yeah, yeah, tell me something new about free speech in your so-called “free” country, where even the president can be silenced and deplatformed.
  • Corporations—the very ones you used to run—wield the real power. We’re living in a cyberpunk reality. But enough with the ranting; there were some worthwhile insights I can use to refine my own thinking and cross-pollinate ideas already stored in my mental vault.
• The pivotal phenomenon unfolding before a stunned audience is an AI arms race, in which megacorporations and various states, including Russia, have actively engaged. However, the true revolution is transpiring behind closed doors, driven by three metafactors—where meta signifies that each element amplifies the others in a self-reinforcing feedback loop.
  • Expansion of the context window, which is growing indefinitely. This means I could theoretically use an infinite prompt to query AI on anything. Not by feeding it an endless stream of characters, but by leveraging its responses, refining them, and feeding them back—enabling a chain-of-thought reasoning process. I suspect that DeepSeek’s DeepThink and GPT’s o1 function in a similar manner.
    • Now, let’s take this concept a step further. Imagine I need a meticulous, step-by-step guide to accomplish a complex task—say, synthesizing dynamite. Of course, AIs are programmed with safeguards against such queries, but I wouldn’t ask outright. Instead, I’d take an indirect approach: inquiring about the necessary ingredients, their chemical interactions, the tools involved. With a 1,000-step iterative process, I would eventually uncover the full methodology. This strategy allows for the resolution of almost any intricate problem, even those I have no prior knowledge of. Which brings us to the second groundbreaking element: AI Agents.
  • Imagine we train an LLM exclusively on comprehensive data from a single domain—chemistry, physics, or any other scientific discipline—excluding all extraneous information. Then, we instruct it to analyze, derive insights, or generate hypotheses.
    • Through a multi-step reasoning process, it could propose theories that range from utterly absurd to groundbreaking. If it produces nonsense, we simply restart the process until it yields something of value. This leads us to the final crucial factor: actionable execution.
  • Text-to-action is the frontier AI models are still mastering, but let’s be honest—they’re already writing competent Python code.
• The real question is: when will we fully step into this reality? I believe we already have. Which brings us to the next, more pressing question: what should we do about it?
  • There are two possible approaches: one is to suppress AI development (pull the plug), as suggested by the tech-luddites — Harari, Musk, Schmidt, Brin. The other is to embrace the transformation, harnessing it to either ride the wave or lead the charge. As we say in Russia:
    • q if you can’t compete, take the lead.
  • From my perspective, the most significant drawback of AI autonomy is its lack of discernment.
• There’s a nonzero probability that we will soon witness AI autonomously verifying the work of another AI—without human oversight.
  • In fact, I strongly suspect this is already happening as I type these words.
  • How does an AI determine what warrants attention and what does not? How can it evaluate the significance of any given information? I can’t fully grasp the answer, but a logical starting point would be to engage the scientific community in ensuring we don’t cross an irreversible threshold.
  • Universities possess immense intellectual capital but are severely underfunded. A strategic move would be to grant educational institutions unrestricted access to AI models from major corporations—empowering students to conduct meaningful research with AI or enhance the depth and impact of their existing work.

---

BIO

theBrain mapping

ID: 202502011102 Source:: Friend:: Child:: Next:: When AI starts communication with another AI they develop language

Keywords:

Reference: