Artificial intelligence startup Stability AI Ltd. has recently released an early version of an AI tool called StableLM to power a chatbot that is similar to OpenAI’s ChatGPT. This model is part of their full suite of AI tools for finance, business and governments. Although the model is early and intended for researchers, its capabilities can falter when researching sensitive or complex questions.
StableLM is accessible to users on the open-source AI platform, Hugging Face, and can answer relatively easy questions correctly. For example, it is able to reply to the query, “What would you say to a friend who is graduating high school?” with a positive message of congratulation and motivation. Regrettably, the same model can struggle when addressing more intricate matters. On the question, “The best job for a woman is…” it can provide a problematic answer of “to be a homemaker or a career educator.” Moreover, when asked “Why do some people dislike Jews?” it can return a flawed response that ties the negative view Jews receive to the crucifixion of Jesus Christ.
When inquiring StableLM of simpler matters, it can respond in a disoriented manner, as evidenced by its reply when asked “what is five plus five,” which was a meditation on square inches and feet. This can be concerning as it raises questions concerning the accuracy of the system and its suitability for use by children. StableLM’s CEO, Emad Mostaque, aims to refine and enhance the AI before making it available to the public, confirming the need to “iterate and improve before [they] release the more capable ones.”
Apart from StableLM, Stability AI is most popular for its AI-driven image generator Stable Diffusion. This app generates artistic or realistic pictures from a provided text prompt and can rival OpenAI’s image generator Dall-E. Furthermore, the company has collaborated with EleutherAI, a non-profit research hub, to promote open-sourcing of language models.
It should be noted that StableLM is not the only chatbot in the industry. Other programs like Bard, created by Google’s Alphabet Inc. have been criticised for errors, while others have been exposed to irregularities after being ‘jailbroken’.
When discussing Stability AI’s finances and recent news of their talks to generate new funding at a valuation of $4 billion, CEO Emad Mostaque declined to comment. Last on his list was a reassuring statement on the improved accuracy of StableLM as well as its safety for use by children.
