Why The confusinig labeling of AI models and tools is funny/not funny
Did you notice? There seems to be a bottleneck in available names for products and solutions in the AI world. Quick check of your AI-bubble-knowledge status:
io is
- name of a company founded by Sam Altman and Jony Ive to develop an AI device
- name of the device they are going to sell
- name of an inhabitable moon around Jupiter
- the name of Google's annual conference this week
No cheating! Answers below*
the problem defined
What's in a name? For AI products, no clear guidance for customers. As artificial intelligence tools multiply faster than we can keep track of them, we're witnessing basically the entire industry struggling with naming conventions. This isn't just about marketing confusion – it's creating real barriers for users trying to navigate the AI landscape and make informed decisions.
When Microsoft's Build conference gets confused with builder.ai's bankruptcy or ai-build.com's 3D printing platform, we have a problem. When users can't distinguish between OpenAI's Codex agent, Codex CLI, and the archived OpenAI Codex from 2021, as Simon Willison found out, (also discovering that the latter is still the current topic of Wikipedia's OpenAI Codex page), we're not just dealing with poor branding – we're looking at an industry moving too fast and leaving their communication teams behind...
Mistral's new coding agent Devstral succeeds Codestral (while their public LLM interface is Le Chat, meow).
The Evidence: A Naming Free-for-All
Pi was known to me as the still existing, emotionally intelligent chatbot of Inflection AI, but I learned the other day that there is also an AI model evaluator Pi Labs who uses withpi.ai as URL and has Pi Scorer as flagship model. Needless to say both are not to be confused with Phi, Microsoft's small language model family.
While the label Whisper (by OpenAI) made sense to me for an automatic speech recognition system, I still struggle to remember Whisk for Google's AI-powered tool for visual creativity that lets users remix and generate new images by dragging and dropping photos as prompts (maybe not too important as it is not available for me in Germany yet). And of course another startup was smart enough to launch a speech dictation tool called Wispr Flow. (I also wonder how long partial disemvoweling will be a trend in the digital world - see Flickr, Tumblr, and others)
And with all those agents and assistants the demand for good names increases - they are very likely to be anthropomorphized. So far it's not going too well, if you ask me: Google's new Jules coding agent is a different thing than Joule, the effective SAP AI assistant.
When Function Meets Confusion DEEP
Some AI labeling followed a smart idea, since it's common sense that customers appreciate if things' names refer to their functionality.
So of course the term DEEP (think, search, seek) comes to your mind when you develop an AI model that offers the respective capabilities. Still, after Google Deep Research launch in December 2024, Deepseek released their extremely smart reasoning model R1 by the end of January 2025 (and here 'Deep' is even part of the company name) OpenAI didn't mind labeling their agent Deep Research on 2 February, while Perplexity came up with an identical feature mid of February.
count your children
Around the name of Google's Gemini model, there are a lot of rumors and conversations, but no corporate statement. Users believe in zodiac inspiration - in astrology, Gemini is known for being adaptable, communicative, and multifaceted—traits also associated with intelligence, curiosity, and quick thinking. Others see a 'twin' element in technical architecture, some also hinting to a MoE (Mixture of Experts) approach. Greek mythology offers another angle - Castor and Pollux (the Gemini twins) were known for their complementary skills—one mortal, one divine—working together to achieve greatness.
However, the company so far sticks to quite coherent labeling of their models, from Gemini 1.0 in December 2022 to the current Gemini 2.5, despite the fact I sometimes need my cheat sheet to be sure about the features of 'Flash' 'Experimental' and 'Pro'.
Microsoft went a different direction, slapping "Copilot" on everything from their standalone chatbot to Office integrations to the acquired GitHub assistant. Users never know if they're getting OpenAI's models, Microsoft's own MAI system, or some hybrid - all hidden behind a more or less identical branding.
They are more transparent but less consistent at OpenAI, expecting people to understand that ChatGPT 4o and o4 belong to very different model development chains and come with distinct focus: 4o was released in 2024 as the first model with multimodal interaction (o for omni, as Latin savvy peers know, of course) while o4, launched in April 2025 (o here rather stands for O-pen AI), offers advanced reasoning and analytics. But fortunately, the makers helped us by first introducing GPT-4.5 by end of February 2025, then the GPT-4.1 series in April. Makes sense, doesn't it? They anyway announced to merge everything into a unified GPT-5 on the public interface soon, so no worries...
🧐 You find all available OpenAI models here: -> https://platform.openai.com/docs/models
names with a meaning
Some names actually work, as they have cultural references or carry suitable associations like Falcon as the label for an LLM family developed by the Technology Innovation Institute (TII) in Abu Dhabi. The only issue is that they were released only days from the agents Crow, Finch, and Falcon, forming the Robin AI multi-agent system, developed by FutureHouse.
By the way: Did you know that the label Grok of Elon Musk's xAI origins from Robert A. Heinlein's Sci-Fi novel 'Stranger in a Strange Land', meaning to deeply and intuitively understand or empathize with something or someone, often used in tech and sci-fi communities to describe profound comprehension? Would anyone have doubted Elon's empathy? (Side note: Groq is NOT an LLM but the name of a tech company specializing in ultrafast inference solutions and LPUs - and they were first, a legal dispute about copyright infringement with xAI is still open.)
The Chinese naturally have a sense of speaking labels. The name of Tencent's video model series Hunyuan is composed of "hun" (浑) meaning "complete" or "whole," and "yuan" (圆) meaning "round" or "circle." In Daoist philosophy, "Hunyuan" is associated with the primordial state of the universe, where all things are interconnected and in balance.
Gives some hope for the AI future in the East and West, doesn't it?
So What? Why This Chaos Matters
This naming mess isn't just an annoying side effect of rapid innovation – from my pov, it's actively harming AI adoption and understanding. When private and enterprise customers can't easily distinguish between similar-sounding tools, they waste time, make poor purchasing decisions, and lose confidence in the technology.
More fundamentally, this chaos reflects an industry that's prioritizing speed over sustainability. In the hectic launch times right now there seems to be a lack of testing phases
anyway - we kinda miss those 'Strawberries' and 'Orions', obviously the same is true for thoughtful naming. The confusion also creates barriers to mainstream adoption. If
tech-savvy professionals struggle to keep AI tools straight, how can we expect everyday users to navigate this landscape?
Or they all do the research with AI and the confusing terms and abbreviations are already part of an eval AGI plan to dumb humanity 🤯😄
a call for better ai naming
The AI industry needs to slow down just enough to get naming right.
You know how to do this, right?
- Trademark research before announcements (looking at you, Grok vs Groq)
- User testing of names with actual customers, not just internal teams
- Namespace coordination between teams, especially within the same company
- Meaningful differentiation that helps users understand what each tool actually does
If you are one of those ambitious founders out there, eager to bring a new model or tool to the market, consider the long-term impact of your naming choices. Your customers and users will thank you.
As alwys: findings and suggestions on the issue are welcome!
*Answer: (update 24 June: amidst a trademark debate with Google iyO 'io' was erased from OpenAI websites, including the PR movie) 1. is true, may change after the integration into OpenAI who bought the company for 6.5bn. Which is a lot, given that 2. is a maybe, as there is no product yet, just a some negations (not having a screen, not a wearable) and big promises in a Hollywood like cinematography PR movie around the acquisition deal. 3. is also true (thanks Alberto Romero) and so is 4. although only phonetically, it's spelled I/O.
Write a comment