{
    "@context": "https://schema.org",
    "@type": "NewsArticle",
    "@id": "https://badiyaa.com/sergey-brin-gemini-rsi-treadmill-argon/#article",
    "mainEntityOfPage": "https://badiyaa.com/sergey-brin-gemini-rsi-treadmill-argon/",
    "headline": "Sergey Brin says Google is on an ‘RSI treadmill’ as Gemini models help build their successors",
    "description": "In a Google roundtable for Gemini 4 Argon, Brin said researchers and the model now build the next version together. Google's own model card is more cautious.",
    "datePublished": "2026-10-17T15:07:00+01:00",
    "dateModified": "2026-10-17T12:18:26+01:00",
    "author": {
        "@type": "Organization",
        "name": "Badiyaa Editorial Desk",
        "url": "https://badiyaa.com/author/editorial-desk/"
    },
    "publisher": {
        "@type": "NewsMediaOrganization",
        "@id": "https://badiyaa.com/#organization",
        "name": "Badiyaa",
        "url": "https://badiyaa.com/",
        "logo": {
            "@type": "ImageObject",
            "url": "https://badiyaa.com/wp-content/uploads/2026/10/badiyaa-logo-color-600x60-1.png",
            "width": 600,
            "height": 60
        },
        "email": "badiyaavideo@gmail.com",
        "publishingPrinciples": "https://badiyaa.com/editorial-policy/",
        "correctionsPolicy": "https://badiyaa.com/corrections/",
        "actionableFeedbackPolicy": "https://badiyaa.com/contact/",
        "ownershipFundingInfo": "https://badiyaa.com/about/#ownership",
        "sameAs": [
            "https://www.facebook.com/profile.php?id=61595518283625",
            "https://www.instagram.com/badiyaa.com.ai/",
            "https://www.threads.net/@badiyaa.com.ai"
        ],
        "parentOrganization": {
            "@type": "Organization",
            "name": "Rezilient Digital",
            "url": "https://www.rezilientdigital.com/",
            "address": {
                "@type": "PostalAddress",
                "streetAddress": "Fl No 3, Nav Shyam Sadan, Plot No 38, Pestom Sagar, Chembur",
                "addressLocality": "Mumbai",
                "addressRegion": "Maharashtra",
                "postalCode": "400089",
                "addressCountry": "IN"
            },
            "sameAs": [
                "https://www.linkedin.com/company/rezilientdigital"
            ]
        }
    },
    "isAccessibleForFree": true,
    "inLanguage": "en-US",
    "wordCount": 796,
    "image": [
        "https://badiyaa.com/wp-content/uploads/2026/10/Sergey_Brin__50910683_.webp",
        "https://badiyaa.com/wp-content/uploads/2026/10/Sergey_Brin__50910683_-1024x683.webp"
    ],
    "articleSection": "Tech",
    "keywords": "artificial intelligence, Gemini 4 Argon, Google Gemini, Sergey Brin",
    "articleBody": "Sergey Brin says Google’s Gemini models are now helping to build the models that replace them. “We’re starting to get onto that RSI treadmill,” the Google cofounder said in a 30-minute roundtable that Google posted to its Google for Developers YouTube channel on Wednesday, October 14, to introduce its new flagship model, Gemini 4 Argon. RSI stands for recursive self-improvement: an AI system that helps design a more capable successor, which then helps design the next one. In theory each turn of that loop runs faster than the last, and that is why the phrase makes AI safety researchers nervous. What Brin actually said Brin, who returned from retirement to work on Gemini and is credited as a core contributor to it, spoke for about 45 seconds, roughly three and a half minutes into the video. He pointed to the run of smaller Flash models Google shipped before Gemini 4 as the evidence. “And if you look at even before Gemini 4, when we had Flash 3.6, 3.7, 3.8, we’re starting to get onto that RSI treadmill. And now the, you know, our researchers and the model work hand in hand to make the next version of the model.” He added that he had “popped in here randomly” because he and Melvin Johnson had been discussing the next model, and that “probably by the time folks are viewing this, internally we’re already using a couple models ahead.” His closing line on the subject: “I expect it’ll only continue to accelerate.” Note the wording. Brin did not say Gemini is improving itself without people. He described researchers and the model working “hand in hand”, which is a claim about speed and division of labor, not about autonomy. The cadence, checked against the dates The release record backs up the part of his claim that can be measured. Gemini 3.6 Flash arrived on July 21, 2026, 3.7 Flash on August 13 and 3.8 Flash on September 2. That is a gap of 23 days and then 20 days. Argon went to a restricted early-access program on September 30, 28 days after 3.8 Flash, and became generally available on October 13. So Google has been putting out a new model about every three to four weeks since midsummer, which matches the “fast three to four week cycle” another panelist described a minute later in the same video. Fast release dates do not by themselves prove the models are doing the research, though. Google’s own paperwork is more cautious than its cofounder. What Google’s model card says The Gemini 4 Argon model card, published on October 13, tests the model against Google’s Frontier Safety Framework. One of those tests is called “ML R&D acceleration level 1”, defined as a model that “has been used to accelerate AI development, resulting in AI progress substantially accelerating from historical rates.” Google’s verdict for Argon: critical capability level not reached, based on an internal benchmark of research engineering tasks. The stricter “ML R&D automation” level, a model that could fully replace a Google research team at similar cost, was also not reached. The same card is less reassuring elsewhere. Argon passed 14 of 15 of Google’s stealth and situational-awareness challenges, and Google says it cannot rule out that the model has reached the early-warning threshold for undermining human control, so it has added safeguards for internal and external use. Google’s separate safety report for Argon says AI agents now write a majority of production code and bug fixes at the company, with every change still reviewed by a human before it goes in. Google is not alone in this. Anthropic has said Claude wrote over 80% of the code it merged as of May 2026, and that human review of all that output turned into the new bottleneck. Critics of the RSI idea argue the opposite risk: that models trained heavily on their own output degrade rather than improve, a failure known as model collapse. What it means if you use Gemini For anyone paying for Gemini or building on its API, the practical effect is churn. Argon raises the output limit from 64,000 tokens to 1 million, about 15 times more, and is available in the Gemini app, the Gemini API, Google AI Studio, Gemini for business, Google Antigravity and AI Mode. If Brin’s “couple models ahead” comment holds, the model you are testing this month will likely be superseded within weeks, so lock prompts and evaluations to a specific model version rather than an alias that updates on its own. Argon’s knowledge cutoff is March 2026, according to the model card. On Google’s recent rhythm of a release every three to four weeks, the next Gemini model would land in early to mid November. Further reading Wikipedia: Recursive self-improvement Wikipedia: Gemini (language model) Image: JD Lasica from Pleasanton, CA, US, CC BY 2.0 via Wikimedia Commons This article was produced with AI assistance and reviewed by our Editorial Desk. How we work.",
    "markdown": "https://badiyaa.com/sergey-brin-gemini-rsi-treadmill-argon.md"
}