By admin 
Silicon Valley has long had its own dialect. Venture capitalists are always talking about the value of being “high agency” and making “orthogonal bets.” But lately, tech vernacular has taken a peculiar turn—people have started describing themselves as if they were chatbots. If you misspeak, maybe you’re “hallucinating.” Don’t know the answer to a question? That’s because it’s not in your “training data.” Feeling forgetful? Perhaps you have a case of “context rot,” a phrase that refers to the degradation of a bot’s responses over a long conversation. “I’ve been telling my wife I have context rot for months,” Conor Bronsdon, the host of an AI-focused podcast, told me.
Read more President Trump Has a Pet-Project Docket at the National Parks
Where I live, in San Francisco, such comparisons are inescapable. “I’ve described myself as high temperature,” a friend recently told me; in AI-speak, this means he is prone to randomness. AI experts now talk about updating their “weights” when they learn something new and training on “synthetic data” when going over internal thoughts. “Humans have a huge base model trained over billions of years of evolution,” one software engineer wrote on a popular tech forum. “It’s impressive how quickly we learn, but it’s arguably comparable to fine tuning.”
Such speech easily comes off as unsettling, if not aggressively bleak: Why describe beautiful, tender life in detached, algorithmic terms? At the same time, language is a fossil record of previous technological revolutions, and if this current transformation is anything like previous ones, some of this new slang may well stick around. AI experts have repeatedly cautioned that anthropomorphism, or the tendency to attribute human qualities to nonhuman objects such as chatbots, can be misleading. But now the reverse is occurring in everyday speech—a sort of modelmorphism, wherein people describe themselves as if they were large language models.
The comparisons really took off a few years ago, after a group of AI researchers argued that language models are “stochastic parrots” that link together language based on statistical patterns without possessing any understanding of meaning. Dissenters retorted that humans are also “stochastic parrots.” As one software engineer riffed, “Humans are basically a sophisticated Markov chain. They are very good at pattern matching, but have no understanding of anything.”
Language has historically evolved alongside technology. Expressions such as running out of steam and cog in the machine, for instance, were adopted into common speech after the Industrial Revolution as shorthand for the human condition. AI metaphors are simply figures of speech, Anna Ivanova, a cognitive scientist at Georgia Tech, told me: When someone says, “My gears are turning,” they don’t literally mean that there are gears or gearlike things in their brain that are actually moving; in much the same way, when people say that they have “context rot,” they don’t mean that they literally have a transformer-based AI model running in their brain.
The difference between earlier technologies and AI systems is that artificial neural networks are loosely inspired by the brain itself. Philosophers such as Raphaël Millière and Cameron Buckner have argued that LLMs could serve as potential models of certain aspects of human cognition. In the 2010s, scientists found parallels between how image-recognition neural networks and the human brain process visual information. There may also be “parallels” between how LLMs and humans represent language, Ivanova said. The MIT neuroscientist Ev Fedorenko and others have identified a type of “language network” inside the human brain, which Fedorenko believes is “very similar in many ways to early LLMs.” These comparisons are contested, but they are receiving research attention: “The recent success of LLMs is rooted in the same computational principles that govern biological brains,” scientists wrote in a position paper last week.
Read more Photos: Releasing Animals Back Into Their Habitats
“Because we do not understand the brain very well we are constantly tempted to use the latest technology as a model for trying to understand it,” the philosopher John Searle wrote in the 1980s. “I am told that some of the ancient Greeks thought the brain functions like a catapult.” In the 19th and 20th centuries, people turned to telegraphs and telephone switchboards as models for the mind: “The brain is no more than a kind of central telephonic exchange,” the philosopher Henri Bergson once wrote. By the ’60s, some scholars were arguing that the mind functions like a computer. “Cognitive scientists often say that the mind is the software of the brain,” the philosopher Ned Block explained in 1990. Today, LLMs make for an alluring comparison.
For now, these analogies are being made mostly by those working in the AI industry. But over time, some of the metaphors might start to seep into everyday language. If you look at the growth of railroads, “a whole range of language change began to happen” as people started to use the new transportation system, Naomi Baron, a linguist and professor emerita at American University, told me. Some of the AI slang isn’t introducing new phrases into English so much as transforming the meaning of existing speech.Take the word hallucinating, which is used to describe AI models confabulating. People are “taking a term created for what the human mind is capable of doing, projecting it onto AI, and then projecting it back onto humans with an AI flavoring,” Baron explained.
None of this is to say every AI-inflected term should get a free pass. I’ve started chastising my friends when I catch them unironically referring to themselves as chatbots. Using LLM metaphors to describe our minds could lead to a reductive and dehumanizing understanding of human cognition. Earlier this year, when OpenAI CEO Sam Altman was pressed on AI’s natural-resource demands, he said that “it also takes a lot of energy to train a human.”
Still, the people in tech I spoke with told me that using LLM-speak to describe their own mind helps them reflect on their cognitive habits. Language subsumes new idioms over time and masks their origins. People today say “My plans have been derailed” without much thought to the phrase’s railroad origins. Perhaps my great-grandchildren will grow up saying things such as “I have context rot” and “That’s not in my training data” without pausing to remember that those phrases first applied to machines.
Read more Donald Trump Effectively Just Told Iran to Get a Nuclear Weapon