⬥ Level of Understanding
This is more of a starting point and my very own ‘accountability buddy’ to make sure I don’t waste time testing tools I have no use for.
My goal with this list is to document my use of generative AI tools and to keep my actions in check with my generative AI intentions. The structure for this list is inspired by the Technology Radar from Thoughtworks and will be updated frequently.
It might come as a surprise that this list is fairly sparse. That is exactly the point. I don’t believe in a scattered, mass tool adoption practice that really only serves to further inflate the market. Rather, I try as best as I can to evaluate each tool carefully and only test and use things that truly serve me, focusing on the underlying technologies instead of adaptations.
Tools that I use regularly and recommend.
- What it does: generative text-to-text model.
- Why I use it: The uncontested champion when it comes to text generation. The quality of GPT-4 has yet to be beaten by other models. The API is also incredibly versatile and easy to use. For now, this is my go to when it comes to brainstorming, sparring, thought refinement, programming and other creative work.
- What it does: generative text-to-image model.
- Why I use it: The quality of this model is extremely high. Not much prompt engineering is needed to get good results. Unfortunately, there is no free tier, and the user experience leaves allot to be desired. This model is also highly opinionated, there is a certain gloss and finish to these images that might not always be desirable.
- What it does: an iOS and macOS app that allows you to run text-to-image Stable Diffusion models locally on your device.
- Why I use it: This app is actively being developed, it is completely free, and it brings the magic of image generation to your device. No vendor lock in, no internet connection required. You can use whatever model you wish, including SDXL. You can train LoRAs. It is incredibly versatile, but requires a fair bit of tweaking to get good results. I always try to opt for community backed, open source solutions and this is the best example of that.
- What it does: a multiplatform app that allows native, local AI upscaling.
- Why I use it: This is another great example of a free, privacy-friendly software. With this app, you can easily AI upscale images from the comfort of your computer, no need for an internet connection. The makers are working on a cloud version, I can’t speak to that. I only use the local version.
Tools that I’m currently trying out to see if they fit my needs.
- What it does: generative text-to-image model.
- Why I’m trying it: The most impressive thing about this model is its ability to understand nuanced descriptions and relationships. Its integration into ChatGPT also brings us one step closer to a multimodal model. And yet, Midjourney still seems to produce better results from an artistic perspective.
- What it does: generative text-to-image model with refined features.
- Why I’m trying it: This tool allows for commercial use and the dataset used for training adheres to ethical standards. I can’t currently vouch for the quality of the images, but the effort and integration into other Adobe apps is remarkable.
Tools that I’m considering but haven’t started using yet.
- What it does: generative text-to-text model with a focus on ethics.
- Why I’m considering it: With Claude 2 Anthropic has done something fascinating. With its massive context window of 100k tokens and its ethical guidelines for the training data, this is the only real alternative to ChatGPT. Unfortunately, this tool is not yet available worldwide.
- What it does: generative text-to-text model with full integration into the Google ecosystem.
- Why I’m considering it: I have not gotten great results with this one, the reason this tool is interesting is because of the extensions feature, essentially allowing you to connect generative AI to the entire Google ecosystem (search, Gmail, calendar, Google Drive, etc.) And with the looming release of Google’s new model, Gemini 2, this is a tool to look out for.
- What it does: generative text-to-text tool that allows for multiple models.
- Why I’m considering it: This is especially interesting for knowledge workers. Perplexity focuses on delivering sources for all of its answers, something that should really be the standard for all AI tools. The UI is beautiful and the free tier generous. It also allows for the usage of different models, like Claude or GPT-4.
Tools that I’ve decided not to use at the current moment.
- What it does: generative text-to-text model based on GPT-4.
- Why I’m not using it: From the restricted amount of messages you can send, to the lower quality of the answers, I see no upside to using this tool. If I absolutely need internet access, I use Google Bard or a web browsing plugin for ChatGPT.
- What it does: generative text-to-text capabilities for Notion.
- Why I’m not using it: I am a heavy Notion user, but this implementation is just not good enough, besides the obvious integration advantage, there is nothing on offer that can’t be done better with other tools.
- What it does: generative text-to-video model.
- Why I’m not using it: The proposition and execution is impressive, but it’s still not at a point where I would truly recommend using it for your artistic projects. The ease of use is to be commended, but the pricing and quality are just not good enough, yet. A tool like this would allow me to push the envelope further and bring generative AI into my film work.