pull down to refresh

Why Does Every AI Product Name Have a Decimal?
181 sats \ 15 comments \ @kr 15 Apr AI
People really suck at math, where did all these AI companies get their branding advice from?
It’s not an accident that brands like Nike and Apple use whole numbers that increase by 1 each season to denote product improvements.
I imagine there is some computer science reason behind the decimal labeling systems teams like OpenAI and Claude use for their models, but it’s a losing strategy if they want to build a great brand.
write
preview
related posts
50 sats \ 2 replies \ @nullcount 15 Apr
Its common practice in software dev to label releases with a major.minor.patch (i.e. v0.1.23) version notation. A bump in major version represents a huge architectural difference, or rewrite, or release of many big features, and/or incompatibility with previous versions. Whereas a bump in minor version is less significant, but still probably has lots of improvements. A bump in patch version is likely just a bugfix or less noticeable change.
The naming convention for LLMs likely stems from this as well. It has little to do with fractions or doing math. Its just a way to see at a glance how a piece of software is improving over time. Bigger number usually means better/battle-tested.
reply
0 sats \ 1 reply \ @kr OP 15 Apr
Good background, it makes sense that they might take this approach since they are writing software... but it still makes for awful branding.
reply
0 sats \ 0 replies \ @nullcount 15 Apr
Base models are technical products. Apps like ChatGPT use the base model to deliver value. I suppose apps benefit more from branding than the base models themselves.
reply
0 sats \ 0 replies \ @Lazy_AMA 16 Apr
It does feel like each AI item is named something like a Model Name v2.5 or an AI X v1.3 The decimal thing as a rule comes from these reasons:
  1. Versioning culture from software/dev world:
It's a long-time convention to form computer program like that, let's say "v1.0", "v2.1", etc. It gives a sense of movement and change over time without inferring a completely modern item.
  1. Inconspicuous promoting flex: A decimal overhaul says the latter is superior than the former without requiring a showy title alter. It's a way to show that development is continuous.
  2. Dodges overhyping: Hopping from 1.0 to 2.0 sounds tremendous. But in case the alter is incremental, like a little show change or an overhaul in preparing information, they go for something like 1.1 or 1.2 to flag it's unused, but not groundbreaking.
These are some of the reasons it appears that way.
reply
0 sats \ 0 replies \ @Dkryptoenth 16 Apr
Great question! The decimal in AI product versioning (like GPT-4.0, Claude 2.1, etc.) usually comes from semantic versioning, which is a common way to track software updates. Here's what it generally means:
Major version (before the decimal): Big changes, overhauls, or milestones (e.g., GPT-4 vs GPT-3).
Minor version (after the decimal): Smaller improvements, tweaks, or new features (e.g., GPT-4.1 might be faster or more accurate than 4.0).
Sometimes a third number is used (like 4.1.2), which refers to bug fixes or patches.
It’s not just AI—this is standard in software dev. But in AI, where models evolve quickly, versioning helps people understand how much a model has changed and whether it’s still the same core architecture or a new generation.
reply
0 sats \ 1 reply \ @ek 15 Apr
I imagine there is some computer science reason behind the decimal labeling systems teams like OpenAI and Claude
I think they just want to reflect how much it improved. Claude 3.5 and Claude 3.7 means it's not as big of a jump as if they used Claude 3 and Claude 4. It's not that deep. 👀
reply
0 sats \ 0 replies \ @kr OP 15 Apr
What about ChatGPT 4.5 and 4.1?
reply
0 sats \ 3 replies \ @zuspotirko 15 Apr
Every? gpt3.5 or Claufe Sonnet 3.7 have decimals in branding. Gpt o1 and o3 do not. Gpt 4o does not. Deepseek-r1 does not. Llama 4 does not. Mistral-large-2411 does not.
reply
5 sats \ 2 replies \ @kr OP 15 Apr
They have dashes and letter/number combinations that are even worse.
reply
0 sats \ 1 reply \ @zuspotirko 15 Apr
Users just use "Claude" or "ChatGPT" or "Deepseek" without knowing which one.
Let me ask you this: Do you know if the Tshirt you're wearing right now is Jersey knit or Oxford cloth or Piqué or Twill? ... Exactly.
reply
0 sats \ 0 replies \ @kr OP 15 Apr
I do, but that's not important.
Notice that t-shirt brands don't advertise their thread patterns like AI platforms do.
You seem to understand that people don't refer to these platforms by their advertised names, so why do you find these product naming conventions useful?
reply
0 sats \ 2 replies \ @SimpleStacker 15 Apr
To make it sound smart?
Nothing made me sadder than seeing a college student not know how to simplify \frac{2}{4}
Sometimes I want to leave this world and float away
reply
0 sats \ 1 reply \ @kr OP 15 Apr
To make it sound smart?
You mean to make your customers feel dumb? Losing strategy.
reply
0 sats \ 0 replies \ @rafael_xmr 15 Apr
meh it sounds like a mixture of the Apple model, with a versioning model to sound smart, that way they can brand it as a new version, but not really that great because it's a "middle version", so there's a lot of hype for "whole number" versions, like OpenAI milking what they can right now to create huge expectations for GPT-5, "if version 4.1 and 4.2 and 4.3 etc.... are this good, just imagine version 5!!!!"
then they keep raising money from investors in the meantime
Apple can't release a new phone every month due to logistics of a physical device, and people don't switch phones so quickly, so they just release a new whole number version every year (or a variant like iphone SE)
reply
0 sats \ 0 replies \ @grayruby 15 Apr
I think they should lean into it and just go with fractions. ChatGPT 2 5/8
reply