🖖 Open-Source AI: Uncensored Models, Building Niche Apps, ChatGPT Alternatives

In this Trends.vc Report, we talk about open-source ChatGPT alternatives, how to easily build and deploy open-source AI models, niche open-source AI models and more.

💎 Why It Matters

Open-source AI helps us build faster by learning from each other.

---

🔍 Problem

Closed-source AI companies are gatekeepers.

They decide when and what you can use AI for.

---

💡 Solution

Open-source AI helps us learn from and build on each others’ work.

This turns an arms race into collaboration.

Now anyone can build ChatGPT.

---

🏁 Players

Open-Source AI Companies

• Explosion • Builds privacy-first AI tools
• Stability AI • Builds audio and image generation models
• Google DeepMind • Researches AI ethics and performance

Open-Source AI Platforms

• Hugging Face • Build and deploy open-source AI models
• Replicate • Build and run open-source models in the cloud
• Google Colab • Platform for machine learning research

Open-Source AI Models

• Stable Diffusion • Text-to-image model
• Riffusion • Text-to-audio model
• BLOOM • Open multilingual language model

Open-Source AI Datasets

• The Pile • Dataset of books, webpages, chat logs and more
• ImageNet • 14,000,000+ annotated images
• OIG • Dialogue data for AI chatbots

Open-Source AI Tools

• PyTorch • Framework for building deep learning models
• TensorFlow • Open-source machine learning framework
• Keras • Deep learning API for AI models

---

🔮 Predictions

• We’ll see more uncensored models. This is “true” open-source AI that follows the Rule 6 of “The Open-Source Definition”.

• WizardLM is an uncensored AI model for text generation.
• Mage is a free, fast and unfiltered Stable Diffusion model.
• NSFW Story Generator is a story generator without morals.
• NSFW Waifu Diffusion is an uncensored anime character generator.

• We’ll see open-source ChatGPT alternatives.

• GPT4All is a free, local, privacy-first AI chatbot.
• HuggingChat is an open-source ChatGPT alternative based on Open Assistant.
• StableVicuna is an AI chatbot that uses Reinforcement Learning from Human Feedback.

• We’ll see more platforms built to host open-source AI models. They will make it easier to build and deploy AI models.

• Replicate lets you use open-source models at scale.
• Hugging Face lets you build, train and deploy open-source AI models.
• Google Colab lets you write, run and share machine learning code in the browser.

---

☁️ Opportunities

• Build a niche AI model. Cater to customers with AI tools designed for specific needs.

• ICON turns images of humans into 3D clothed models.
• Animated Drawings turns simple drawings into 2D animations.
• QR Code AI Art Generator makes background images for QR codes.
• Text2Light turns text into 4K HDR panoramas for 3D scenes, VR and more.

• Offer paid subscriptions to your open-source AI tools. Monetization helps your project sustain the AI race.

• Coqui is a text-to-speech and voice cloning tool.
• Lightning AI helps you train, deploy and build AI.
• Cody helps you read, write and understand code.
• Giskard helps to lower bias and performance errors in AI models.

• Cover new advances in open-source AI. Build an audience by sharing open-source AI news, tools and companies.

• The Breakdown talks about open-source AI agents such as AutoGPT and SuperAGI.
• Last Week in AI writes about open-source models such as PandaGPT and DragGAN.
• Aitrepreneur covers open-source AI models such as Falcon-40B, WizardCoder and more.
• Two Minute Papers covers AI agents and image generators such as AutoGPT and Stable Diffusion.
• The Rundown AI writes about open-source tools and models such as Transformers Agent and Prem.

---

🏔️ Risks

• Race to the Bottom • New open-source models quickly replace the old ones. You can get stuck in a rat race by trying to keep up with the pace. Don’t fall into a coma.
• Copyleft • These licenses let you change code but you must open source any tool made with it. Making it impossible to build proprietary tools with “copylefted” code.
• Copyright • It is hard to verify if contributed code or training data is copyright-free. This can lead to lawsuits for using copyrighted work without permission.

---

🔑 Key Lessons

• Open-source addresses the problem of vendor lock-in and high switching costs. Platforms such as Hugging Face make it easier to find AI models that fit your use cases.
• Open-source AI lets individuals build niche applications that large closed-source companies don't have the time, insight or interest to build. From generating 3D landscapes to turning images into music.
• Clear documentation boosts the quality and adoption of your open-source AI tool. It helps users and contributors understand what your tool does and how it works.

---

🔥 Hot Takes

• AI regulations will force AI companies to open up about their closed-source configurations. OpenAI released GPT-4 without detail on how they built it. As US President Joe Biden weighs in on AI safety, OpenAI has explained how it ensures safety.
• Open-source AI deployment is limited by hardware shortage. Open-source projects are usually run by small teams and solo developers. Who may not have access to the computing power needed for their projects.

---

😠 Haters

“Open-source AI is less performant than AI models made by established companies.”
Open-source AI models can be a little worse but a lot less expensive. Depending on your use case, it can be wise to sacrifice quality without wasting lots of money.

“Companies like Meta root for ‘open research’ until they find an edge. Then become closed-source to maintain it. It’s hypocrisy.”
Tech giants are commercial companies. Google invented and open-sourced Transformers that drive modern AI research. Meta open-sourced the code for LLaMA and its leaked version is used to build dozens of other models. They have spent billions on research and hardware. We can’t judge them for trying to compensate for their efforts.

“Closed-source companies make use of both internal and open-source AI research.”
True. Some open-source projects may not exist without support from closed-sourced companies.

---

🔗 Links

1. We Have an Upcoming Report on Open-Source AI • The tweet behind this report.
2. OSI-Approved Licenses • List of 200+ open-source licenses.
3. Open LLM Leaderboard • List of 100+ open-source AI models with performance tests.

---

📁 Related Reports

• Monetized Open Source • Monetization helps open-source projects to sustain.
• ChatGPT • ChatGPT boosts the productivity of businesses and individuals.
• Voice Cloning • Build an audio content machine that works anytime, anywhere.
• Agencies • Help companies solve problems without hiring and managing large teams.
• Prompt Engineering • Learn how to direct AI.

---

Become a Trends Pro Member to get the full report on Open-Source AI.

Do you want to get the next free Trends.vc Report? Join 50,000+ founders discovering new markets and ideas.