Amodei writes, “The intersection of AI and our political institutions seems a bit like Hobbits and Treebeards – that is, the Treebeard Ant moves so slowly that it can’t even understand the faster Hobbits. But then Amodei turns to a controversial claim – one that, to be continued lord of the rings analogy, this would mean that the Hobbits are moving fast all the time.
Amodei claims, “The scaling laws of AI, which predict exponential growth in general cognitive abilities with increasing computing power, now have more than a decade of empirical evidence behind them.” “If these scaling laws continue for only a year or two, we will get what I said Powerful AI.”
This is hardly a one-off reference; Amodei, who is about to cash in on a bonanza Anthropic IPO, uses the word “exponential” six times in the essay. It includes the title, “Policy on AI Exponential.” But is it true?
Let’s leave aside for a moment the dubious 2020 OpenAI paper Amodei links to, or the abundant evidence from other AI experts to prove his claim that the “cognitive capabilities” of most large language models are not growing that fast, if at all. Amodei’s essay appears to refute the… well, the word anthropic.
The AI vibe shift is real: why the backlash is growing
On the system card for the preview of Cloud Mythos [PDF]In the model that Amodei promotes in the essay for cybersecurity concerns, you will find the following statement: “The [intelligence] The benefits that we can identify are Can be confidently attributed to human research, not AI assistance… Early claims of big AI-driven victories have not held up.” [Emphasis ours.]
mashable light speed
You might think it couldn’t be more certain than that – until you read the system card for Anthropic’s other new Frontier model, the Fable 5. Using a test called the Epoch Capabilities Index, Anthropic researchers set out specifically to see if there was evidence of a feedback loop that leads to what AI experts variously call AGI or digital superintelligence. And the result could not be clearer.
“We do not see a sustained, AI-attributed 2× acceleration in the pace of our AI progress,” Fable Systems cards. [PDF] They say.
So, where is Amodei getting his exponential information? We’ve reached out to Anthropic for clarification, but the quote the CEO used is from a 2020 paper called Scaling Laws for Neural Language Models, co-authored by Jared Kaplan (then with OpenAI, now co-founder of Anthropic). The conclusions of that paper have been questioned for the past four years by another prominent AI researcher, Gary Marcus.
“There are serious flaws in the scaling argument,” Marcus wrote in 2022. already Scaling in deep learning is running into limits, perhaps already reaching the point of diminishing returns.” He cited research on OpenAI’s GPT-3 model, which “shows that scaling begins to falter on some measures such as toxicity, truthfulness, logic, and common sense.”
Marcus was criticized by AI true believers at the time, but has since been vindicated – especially since the release of GPT-5, which was not the superintelligence that some of its users expected.
Finding evidence of the exponential growth of AI since then may be harder than getting into Mordor.
Additional reporting provided by Timothy Worth
Subject
artificial intelligence anthropic
<a href