| Happy Wednesday, it's Rachel in London. I've been thinking a lot lately about the names tech companies give their generative artificial intelligence models, and why so many of them reference Greek mythology and animals. But first… Three things you need to know today: • Arm prepares to cancel Qualcomm's chip design license, escalating an ongoing feud
• Anthropic's new AI tool analyzes your screen and acts on your behalf
• Polymarket cracks down on US-based users as whale bets on Trump Over the past few years, tech companies large and small have released a slew of AI models that can generate text, songs, images and more from simple queries or commands. But while the AI industry is largely focused on how to make these models faster and more capable, I've found myself paying particular attention to the monikers companies give their software — be it Llama, Gemini, or GPT-4 — and what those names communicate. I recently conducted an informal review of the names chosen for a slew of AI models. I mostly looked at large language models, such as the software that powers OpenAI's popular ChatGPT chatbot — but at also image- and video-generating systems. I found, in short: There are lots of allusions to astrology and Greek mythology (think Google's Gemini or Amazon's Titan models), a menagerie of models named after animals (Meta's Llama, for instance), and a few less-inspiring names (ahem, o1-mini). I asked Rohan Krishnan, a strategy director at branding agency Red Antler — whose clients range from Google and Zillow to insulated tumbler company Corkcicle — to help me make sense of these names. I wanted to know what, if anything, they mean, and why such naming choices matter, especially when it comes to AI. In general, these product names are meant to evoke emotions related to how a brand wants to position itself out in the world, Krishnan said. Google's Gemini, for instance, is a nod to the constellation, but it also "sounds like something that could be put on the side of a spaceship," giving it a futuristic, ambitious feel. "It's got a little bit of that superhero energy to it, and I think that is why you're seeing a lot of players ideate in that general space," he said. Names inspired by cute animals, meanwhile, seem like an effort to make the software sound, well, adorable — or at least not too serious. And there are plenty of them out there, particularly referencing the camelid family: Meta released its first version of Llama, which stands for "Large Language Model Meta AI," in 2023, encouraging developers to use and build atop it. Since then, at least three Llama-based projects — Vicuna, Alpaca, and Guanaco — have been named after related animals. Krishnan said some of this may simply be about how a word sounds and looks (he pointed out that Llama in particular is easy to say and unusual with its successive l's at the start). Companies drawing on the animal world for inspiration may also be doing so because many animals have attributes that are aspirational to humans, like a cheetah's speed or an elephant's strength. And, he added, animal names can be useful for building a brand with multiple models. Google did this last May, for example, when it debuted four versions of its PaLM 2 large language model, named Gecko, Otter, Bison and Unicorn — all meant to reference their sizes (though, truly, I'm unsure about whether a unicorn would be larger than a bison). There are also plenty of model names that sound more clinical and less exciting, such as OpenAI's GPT series of models. GPT stands for "generative pre-trained transformer." For successive AI models, OpenAI follows the initials with a dash and a series of numbers and/or letters, like GPT-4o. These, too, have depth, according to Krishnan. Such technical-sounding labels can give off a scientific feel, and even though they lack a dictionary definition they can still evoke something powerful in people's minds. "There's something that feels really state of the art," he said. That's not to say that the people behind these models always love the names they give them, however. Take it from Sam Altman. On a podcast last year he joked with Trevor Noah about ChatGPT. "It's a horrible name," Altman said. "But it may be too ubiquitous to ever change." —Rachel Metz Elon Musk's PAC spends just 4% of social media ad dollars on X. Musk has turned his personal X account into a billboard for Donald Trump's election campaign, but the billionaire's political action committee has made the company a much smaller part of its marketing strategy, funneling just 3.6% of its total social media ad spend to his own network. Texas Instruments says it's nearing recovery after sales slump. Meta bans accounts tracking private jets for Zuckerberg, Musk. Huawei Technologies' latest AI chips were produced by TSMC, suggesting that China is still struggling to reliably make its own advanced chips in sufficient quantities. | 
No comments:
Post a Comment