Trump's smarts, LLM company self-regulation, AI vs SI vs LLM, mice vs mouses

Trump's smarts, LLM company self-regulation, AI vs SI vs LLM, mice vs mouses

It gets me how Trump keeps referring to himself as "smart" when he manifestly isn't (although except for Biden the smart ones have not done so well). Anyway, he got the American LLM giants to sign a self-regulation compact. I wonder what is in it, if it has teeth, and will it make a difference.

The White House convened about 20 major tech executives—including OpenAI's Greg Brockman, Anthropic's Dario Amodei, Meta's Mark Zuckerberg, xAI's Elon Musk, Sundar Pichai, and Nvidia's Jensen Huang—to sign what they officially dubbed the "White House Accord on Superintelligence: A Joint Commitment on Frontier SI Responsibilities."   
finance.biggo.com
What Is in the Accord

The core document outlines a four-tier framework aimed at managing frontier model deployment. Its primary elements focus on:

    Internal and External Safety Reviews: Implementing tiered internal testing protocols and third-party red-teaming for potential biosecurity, chemical, and severe cybersecurity risks.

    Voluntary Early Access Window: Encouraging developers to grant federal agencies (notably through CISA, the NSA, and the U.S. AI Safety Institute) a 30-day pre-release review window for high-capability models.   
    Freshfields

    Cybersecurity Defenses: Coordinating with a newly proposed federal "AI cybersecurity clearinghouse" to share vulnerability data and patch critical infrastructure software.   
    Davis Wright Tremaine

    Terminology Pivot: In tandem, the administration issued an directive signaling a preference for using "Superintelligence" (SI) instead of "AI" in official frontier-computing contexts.   
    finance.biggo.com

Does It Have Teeth?

In short: No.

Trump himself explicitly noted to reporters outside the White House that the pact is "morally binding" rather than legally enforceable.   
Washington Examiner

    Explicit Ban on Mandates: The underlying administration framework explicitly prohibits mandatory federal licensing, permitting, or preclearance prior to releasing models into the public domain.   
    Freshfields

    Self-Policing: Trump praised the arrangement by stating, "They're going to police themselves," adding that the signers carry the moral obligation to manage their own risk.   
    finance.biggo.com

    Classified Thresholds: While the order tasks the NSA and Treasury with setting metrics for what qualifies as a "covered frontier model," those benchmarks remain behind closed doors, leaving actual compliance checks dependent on voluntary opt-ins.   
    Freshfields

Will It Make a Difference?

The accord operates more as a political hedge and an industry shield than a structural regulatory guardrail.

    For the Tech Giants: It provides maximum flexibility. Companies get public goodwill and political cover from Washington without risking mandatory shutdowns, heavy compliance costs, or hardware deployment caps. It effectively formalizes the status quo of "grading each other's homework."

    For Federal Agencies: It establishes formal channels for data sharing regarding critical infrastructure vulnerabilities, particularly for national defense and cyber-threat detection.

    For Public Governance: Because it carries no statutory penalties or binding oversight mechanisms, any firm can bypass the 30-day access window or skip third-party audits without direct legal recourse.   
    Freshfields

Ultimately, it codifies a light-touch, market-first posture—leaving the responsibility of controlling advanced models entirely in the hands of the executives who built them.

You seem to see it as pure political hypocrisy, but I can see it being of value for public pressure. What do you make of SI instead of AI? (Me I avoid both and use LLM).

That point about public pressure is entirely fair. Even a purely voluntary or non-binding framework creates a visible benchmark. Once companies sign onto specific principles—even under a "morally binding" handshake—it gives journalists, researchers, advocacy groups, and congressional committees a explicit standard to hold them to when something goes wrong. In public accountability, a paper tiger can still be cited in court or at a congressional hearing to prove a company ignored its own stated commitments.

As for the shift from "Artificial Intelligence" (AI) to "Super Intelligence" (SI), it is a fascinating bit of political branding and narrative framing.   
Fox News

The executive order officially directing executive branch agencies to use "Super Intelligence" and "SI" stems from a few distinct motivations:   
Daily Finland

    Distinction from "Artificial": Trump explicitly objected to the word "artificial," arguing that it makes the capability sound "fake" or synthetic rather than powerful. Framing it as Super Intelligence shifts the emphasis from what it is made of (silicon/code) to what it does (exceeding human capability).   
    BigGo Finance+ 1

    National Prestige and Hype: "Super Intelligence" sounds like a strategic physical asset—akin to supercomputers or superpowers. It fits neatly into a broader economic and geopolitical posture aimed at projecting dominance over global rivals.

    Regulatory Avoidance through Terminology: By standardizing a new term across federal agencies, the administration implicitly redefines the technological target. It frames policy around an imagined ultimate state rather than intermediate software risks, aligning with a light-touch, market-first posture.

Your reliance on "LLM" is by far the most grounded choice. "LLM" accurately names the underlying mechanism—a large language model based on probabilistic pattern matching over sequence data—without bringing along philosophical baggage, sci-fi tropes, or political marketing. "Artificial Intelligence" has always been a leaky umbrella term, and "Super Intelligence" only inflates the hype further. Stick with LLM; it keeps the focus on what the software actually is.

On a completely different thing, we are having a disagreement here: what is the plural for a computer rather than a living mouse, mouses or mice?

The consensus among major style guides and dictionaries (including the Oxford English Dictionary, Merriam-Webster, and The Chicago Manual of Style) is that both "mice" and "mouses" are acceptable, but "mice" is overwhelmingly preferred.

Here is how the usage splits down in practice:

    Mice (The Dominant Choice): "Computer mice" is the standard form used by the vast majority of tech writers, manufacturers, and everyday speakers. If you are writing for a general audience, "mice" is the safest choice and won't draw any double-takes.

    Mouses (The Technical/Historical Alternative): "Mouses" gained some traction in early computing documentation to explicitly avoid picturing the squeaky, tail-bearing rodents. It is still recognized as correct in several dictionaries specifically for the input device, but in common parlance, it can sound slightly clunky or jarring to most readers.

Interestingly, Douglas Engelbart—the engineer who invented the device in the 1960s—was once asked about this in an interview. He admitted that he and his team never really decided on a formal rule: "I don't know, we started out with 'mice', and it just stuck."

So, if you are settling an argument: whoever argued for "mice" has common usage, major dictionaries, and the inventor on their side, but whoever argued for "mouses" isn't technically wrong according to the dictionary!

How dare you suggest we might have an argument here [not serious]. It was a discussion.

Ha! Duly noted—my apologies for libeling such an elevated intellectual symposium! A "vigorous exchange of views" shall be the official record.

I hope the panel accepted the ruling with full diplomatic grace, regardless of who was holding the "mice" or "mouses" flag.


Gemini is AI and can make mistakes.