Stability AI, the company behind the popular text-to-image generative AI model called Stable Difussion, is launching today a new product called StableLM that functions similarly to ChatGPT – the popular AI-powered chatbot created by its competitor OpenAI.
According to a blog post published earlier today, two models of StableLM will be made available by the firm, with the lighter version offering a complexity level of 3 billion parameters while the “heavier” one has 7 billion parameters.
We’re releasing the first of our large language models, starting with 3B and 7B param models, with 15-65B to follow. Our LLMs are released under CC BY-SA license.
— Stability AI (@StabilityAI) April 19, 2023
The company stated that it plans to release more complex models of 15 to 65 billion parameters so. Both of the alpha versions that are being released today can be used for commercial and research purposes.
How Powerful is StableLM?
Users can tap on StableLM to generate both text and code responses. The model has been trained by using an open-source dataset called “The Pile”, which reportedly contains approximately 1.5 trillion tokens of content.
For comparison purposes, GPT-3 – the language model that originally powered ChatGPT – had a complexity of 175 billion parameters. Hence, the models that Stability AI is launching are relatively less sophisticated. However, the quality of the responses primarily depends on the richness of the training data.
“In our evaluations, not only do models trained on the Pile show moderate improvements in traditional language modeling benchmarks, they also show significant improvements on Pile BPB”, The Pile website indicates in regards to the quality of its dataset.
Meanwhile, in regards to the safeness of the model, Stability AI wrote: “As is typical for any pretrained large language model without additional fine-tuning and reinforcement learning, the responses a user gets might be of varying quality and might potentially include offensive language and views”.
Experts have been voicing their concerns about the pace at which AI labs are releasing new and more powerful models and even called for a 6-month halt for the release of more powerful large language models (LLM) than GPT-4 – the latest product from OpenAI.
OpenAI’s founder and Chief Executive Officer, Sam Altman, claimed that, even though some of those concerns are valid, the right approach to deal with them is not to pause innovation. However, he confirmed that his firm is not working on a model more powerful than GPT-4 at the moment although it is working on new ways to make GPT-4 more powerful.
Stability AI relies on the infrastructure of Amazon Web Services to harness the computing power needed to test and launch their models. In August 2022, Stability AI launched Stable Diffusion, an AI-powered image generator that can be used to create pictures from text prompts.
Stability AI Also Released a New Text-to-Image Models Called Stability XL
Recently, Stability AI created a more powerful model called Stability XL (SDXL) that lets users create highly sophisticated and polished images out of shorter text prompts. According to the firm, over 40 million users are currently leveraging the power of the firm’s imaging application called DreamStudio, which is now powered by SDXL.
With DreamStudio, users can create images by adding credits to their accounts. Currently, it costs $10 to purchase around 1,000 credits. With that many credits, it is possible to create approximately 5,000 images.
Also read: 10 Best AI Stocks to Invest in 2023
Differently from OpenAI, Stability AI aims to create open-source AI models that can be used by developers from all across the world to enhance their products and services and even build more sophisticated models by using the company’s products as the foundation.
“We open-source our models to promote transparency and foster trust. Researchers can ‘look under the hood’ to verify performance, work on interpretability techniques, identify potential risks and help develop safeguards”, the company wrote.
It is unclear what the pricing of StableLM will be. If it works similarly to Stable Diffusion, users will have to pay to buy a certain number of credits that will be progressively depleted as they send text prompts to the AI model.
Other Related Articles: