2023-12-16
In 2023 a lot happend in AI. Here I list the most for me interesting AI models, systems, etc.
NOTE: This page is dynamic.
Table of contents:
AI from x.com aka. grok.
Anthropic's model. Currently not available european countries. Anthropic’s Claude models can also used with Amazon Bedrock API.
Poe is a neutral platform. You can access different providers (OpenAI, Anthropic, Google, and open source models like Llama 2 or SDXL).
Currently no developer API.
Aside the known google bard. Google provides gemini (ultra, pro and nano). Nano is dedicated for mobile systems. Pro is already available.
replit.com it seems to use nix as backend OS.Example use case: pair programming with LLM.
You need a API key. PaLM API (legacy) or Gemini API. Note, that Gemini API is not available in some countries: https://ai.google.dev/available_regions#available_regions
Why use LLM during programming?
OpenAI API provides various models for app integration.
| Category | Model Type | Models |
|---|---|---|
| Language Models | GPT-4 Turbo | gpt-4-1106-preview |
| gpt-4-1106-vision-preview | ||
| GPT-4 | gpt-4 | |
| gpt-4-32k | ||
| GPT-3.5 Turbo | gpt-3.5-turbo-1106 | |
| gpt-3.5-turbo-instruct | ||
| Assistants API (Tools) | Retrieval | |
| Code interpreter | ||
| Fine-tuning Models | gpt-3.5-turbo | |
| davinci-002 | ||
| babbage-002 | ||
| Embedding Models | ada v2 | |
| Base Models | davinci-002 | |
| babbage-002 | ||
| Other Models | Image Models | DALL·E 3 |
| Audio Models | Whisper (speech-to-text) | |
| TTS (text-to-speech) | ||
| TTS HD |
Sidenote regarding ada v2 and embeddings: Posted by @karpathy on X.
LLM OS. Bear with me I'm still cooking. Specs:
- LLM: OpenAI GPT-4 Turbo 256 core (batch size) processor @ 20Hz (tok/s)
- RAM: 128Ktok
- Filesystem: Ada002
The Assistant's API allows you to create agent-based applications using our text generation models.
The assistans api consists of tools:
Function calling allows developers to more reliably get structured data back from the model.
Source: https://openai.com/blog/function-calling-and-other-api-updates
Learn how to connect large language models to external tools.
https://platform.openai.com/docs/guides/function-calling
From November 2023, OpenAI's ChatGPT introduces the ability to create bespoke GPT models. Anticipated shortly is the launch of a specialized store for these models.
Currently ChatGPT offers a way to create a GPT
Customize a version of ChatGPT for specific purpose.
Enhancing Security in Custom GPT Model Creation
When creating custom GPT models, it's crucial to safeguard against prompt injections that could reveal sensitive instructions. To enhance security, employ mechanisms like 'rules' or a 'seed hash'. During the creation process, you can generate a unique seed hash. This seed hash acts as a key; only those who possess it can access or view the specific instructions of the model.
This approach ensures that your custom GPT model remains secure and its usage is controlled, thereby maintaining the integrity and confidentiality of your application.
Powered by mozilla. It combines llama.cpp with cosmopolitan to make a single file executable. Either in console or browser.