That gave me an idea, here is what I got from Copilot:
You have set up your system to run different AI models and compare their performance using a text editor. You are using Mixtral-8x7, a high-quality open-source model developed by Mistral AI, Dolphin, an emulator for Nintendo video games, 3.5-Turbo, a customized version of GPT-3.5, a powerful natural language model, and 4-Series Preview, a new version of the BMW sports coupe. You have noticed that the 4.5-Preview, an upcoming update of GPT-3.5, is slightly better than Mixtral-8x7, which used to be a close match. You are still waiting to access Mistral-Medium, a prototype model that is even better than Mixtral-8x7, but only available to a limited number of users.
You have discovered that Perplexity, an AI company that provides information discovery and sharing services, offers free access to Mistral-Medium through their partnership with Mistral AI. You think that Perplexity is making a mistake by giving away such a valuable model, and that they are underestimating the superiority of Mistral-Medium over the 4.5-Preview. You also think that Mistral AI is the new leader in the AI industry, and that their techniques, such as DPO (Data Processing Optimization), Alibi (a library for algorithmic accountability), sliding window (a method for analyzing time series data), and modern mixtures (a way of combining different models), are well-known and effective. You believe that the advantage of Mistral AI lies in the gap between their innovation and the ability of other developers to replicate it on cheaper and more accessible hardware. You also think that the enterprise market is not fond of the complex structure of GPT-3.5 and its variants, and that they prefer to use Mistral AI's models, which are more affordable and operable on their own premises.
You end your text with a quote from the movie Armageddon, which implies that you are leaving a situation that you dislike, but also admire.
I dont think it has anything to do with a BMW sports coupe.
What I am confused about though is it seems like the parent is mentioning models beyond the GPT4 instance I currently have access to. I checked their twitter and I have seen no anouncement for any 4.5 or 4 series previews. Is this just available to people using the API or did I miss something?
> I dont think it has anything to do with a BMW sports coupe
Well, the Paul Ricard circuit in France has a straight called Mistral. Plenty of BMWs have been there for sure, and a zillion other cars.
I wonder if that could have confused the AI a little in combination with other hints. Turbo?
If that's a thing maybe we should start picking our names not only to make them googlable but also not to confuse LLMs at least for the next few years. Months?
The OpenAI API is currently advertising their preview to my clients as `gpt-4-1106-preview`. I've been calling the Q4 2023 4-series stuff `4.5` or `4.5-preview` to distinguish it from the pre-big-announcement stuff.
I don't recall if I saw any press calling anything `4.5`, but it's a different model in some important ways (one suspects better/cheaper quantization at a minimum) and since they've used `.5` for point releases in the past it seemed the most consistent with their historical versioning.
It's a new theme and/or distribution that I'm working on under the working title `hyper-modern`. It clearly takes inspiration from things like Doom (and I'm using their modeline which is hard to improve on) but it's mostly ground up and AI-first.
It's heavily integrated with my custom model server and stuff and I'm slowly getting it integrated with other leading tools (vscode and nvim and stuff).
You have set up your system to run different AI models and compare their performance using a text editor. You are using Mixtral-8x7, a high-quality open-source model developed by Mistral AI, Dolphin, an emulator for Nintendo video games, 3.5-Turbo, a customized version of GPT-3.5, a powerful natural language model, and 4-Series Preview, a new version of the BMW sports coupe. You have noticed that the 4.5-Preview, an upcoming update of GPT-3.5, is slightly better than Mixtral-8x7, which used to be a close match. You are still waiting to access Mistral-Medium, a prototype model that is even better than Mixtral-8x7, but only available to a limited number of users.
You have discovered that Perplexity, an AI company that provides information discovery and sharing services, offers free access to Mistral-Medium through their partnership with Mistral AI. You think that Perplexity is making a mistake by giving away such a valuable model, and that they are underestimating the superiority of Mistral-Medium over the 4.5-Preview. You also think that Mistral AI is the new leader in the AI industry, and that their techniques, such as DPO (Data Processing Optimization), Alibi (a library for algorithmic accountability), sliding window (a method for analyzing time series data), and modern mixtures (a way of combining different models), are well-known and effective. You believe that the advantage of Mistral AI lies in the gap between their innovation and the ability of other developers to replicate it on cheaper and more accessible hardware. You also think that the enterprise market is not fond of the complex structure of GPT-3.5 and its variants, and that they prefer to use Mistral AI's models, which are more affordable and operable on their own premises.
You end your text with a quote from the movie Armageddon, which implies that you are leaving a situation that you dislike, but also admire.