OpenAI, the company behind the viral chatbot ChatGPT, announced plans to launch an online store next week that will allow anyone to create their own customizable versions of ChatGPT. The “GPT Store” will give users access to different sizes and versions of OpenAI’s GPT language model technology to develop specialized chatbots for various use cases.
ChatGPT’s Explosive Growth Prompts Launch of GPT Store
ChatGPT has seen massive growth since its launch in November, quickly attracting millions of users impressed by its human-like conversational abilities. However, OpenAI has struggled to keep up with the explosive demand. Launching the GPT Store seems to be their solution to allow third-party developers to build customized services using their AI while still maintaining control over how the core technology is used.
Some key developments that led up to the planned opening of the GPT Store:
- ChatGPT user base grew from 1 million to over 100 million users from December to January
- OpenAI imposed usage limits in December to manage compute costs
- Third-party ChatGPT plugins and wrappers quickly emerged but were soon deprecated
- Developers expressed frustration at losing access to customize ChatGPT functionality
- Reports emerged this week that OpenAI would launch its own “app store”
“The appetite for AI tools customized to specific use cases is incredible,” said OpenAI CEO Sam Altman. “The GPT Store will enable developers to responsibly innovate on top of our technology.”
GPT Store to Offer Developer Access for Custom Bots
According to the announcement, the GPT Store will allow developers to purchase access to customized versions of GPT tuned for tasks like content generation, classification, and extraction.
There will be various pricing tiers based on factors like:
- Model size (small, medium, large)
- Number of maximum tokens per request
- Number of requests per month
- Ability to fine-tune the model
- Access to advanced features like classifications
Pricing details have not been finalized but reports indicate it will follow a similar structure to OpenAI’s existing API with graduated pricing per token usage.
The GPT Store aims to provide simplified and managed access so that those without extensive machine learning expertise can still leverage the technology. Pre-built containers and templates will be provided to jump start development.
Altman stated that all applications will be reviewed by OpenAI to ensure they meet content policy guidelines before being approved for production access in the store. Strict monitoring will also be ongoing.
Enthusiasm Mixed with Concerns Over “Agency Risks”
The announcement of the GPT Store has been met with enthusiasm by the developer community excited to tap into the breakthrough capabilities of ChatGPT.
However, there are also concerns around the potential risks of enabling widespread deployment of generative AI without more rigorous oversight. Critics point to biases in the model training data as well as objectives misalignment between the AI system and human values.
“This technology has immense promise but also presents risks if misused or poorly implemented,” said Abhishek Gupta of the Montreal AI Ethics Institute. “More guard rails are needed to mitigate ‘agency risks’,” referring to AIs potentially causing harm while pursuing overly narrow or simplistic objectives.
To address these concerns, OpenAI said it will enforce responsible deployment guidelines and also provide tools for monitoring model behavior. But skeptics say that innovative regulation and monitoring frameworks have not kept pace with rapid advancements in AI capabilities unlocked by models like GPT-3 and ChatGPT. The GPT Store promises to accelerate this divergence even faster.
Surging Interest from Enterprises to Build Custom Solutions
Alongside individual developers, OpenAI is seeing huge demand from enterprise customers for private installations of ChatGPT and GPT-3 for uses across sectors like healthcare, education, finance and customer service.
The GPT Store will simplify the procurement and provisioning of these customized enterprise AI solutions.
As an example, developer Anthropic has reportedly partnered with OpenAI to deliver a customized ChatGPT model tailored for business users called Claude. This AI assistant bot is designed specifically to be helpful, harmless, and honest.
Other companies like Anthropic, AI21 Labs, and Cohere are also offering alternative large language model solutions enterprises can leverage for content applications.
But for many use cases, OpenAI’s technology still remains state-of-the-art in terms of low-code access and quality of generations. The GPT Store promises to make this much more attainable for a broader set of end users.
“The addressable market here is virtually every white collar worker, call center agent, and knowledge professional,” said Wayne Hu, Partner at SignalFire VC. “OpenAI is poised to radically transform how we leverage AI augmentation in our daily workflows.”
Table 1: Comparison of ChatGPT Pricing Models
|Free Research Version
|Pro / Plus Version
|Free with rate limits
|$20/user per month
|Pay per token usage
|1 million users per day
|Priority access + higher limits
|Custom models + fine tuning
|Additional features like faster response times, priority access
|Custom models tuned to specific domains and use cases
Concerns Over Bias and Toxicity Still Loom Large
Despite the excitement over expanded access to ChatGPT-like capabilities, concerns still remain over issues like biases, unfairness, and potential to perpetuate misinformation or toxicity.
Critics like researcher Timnit Gebru suggest these issues may even be amplified by allowing unchecked proliferation of generative models tuned towards narrow objectives without enough view towards societal impacts.
“We urgently need guard rails combined with community participation in governance before deploying these models widely,” said Gebru.
pling policy expert, Dr. Jack Clark also echoed risks of generative models producing “specious specificity” – where the AI model generates responses with false authority or incorrectly overstating its confidence. This could perpetuate risk of misinformation.
Brandie Nonnecke, CTO at UC Berkeley’s Center for Technology, Society & Policy said: “This technology has not been vetted for security and ethics like other technologies. We wouldn’t deploy self-driving vehicles on public roads overnight.”
To mitigate these risks, OpenAI said the GPT Store will incorporate strict monitoring for policy violations and toxicity. But many experts argue that this after-the-fact filtering is insufficient given how rapidly capabilities are scaling, and call for more proactive evaluation of risks during development cycles.
“What’s needed is ‘safety by design’ built directly into the models from the beginning phases before they are unleashed at global scale,” Gebru asserts.
Looming Showdown with Rival Large Language Models
OpenAI emphasized that the models accessible via the GPT Store will only be for narrow use cases. The much more powerful GPT-4 model which reportedly demonstrates significantly enhanced reasoning and comprehension capabilities will remain exclusively accessible via API for select customers – at least for now.
But rivals like Anthropic, AI21 Labs and Cohere made announcements this week around rival “ Claude”, “Jurassic-1 Jumbo” and “Cohere GPT-4” models promising to match or even exceed GPT-4 for broad comprehension tasks while still being made available to paying enterprise customers.
This sets up a battle of broad competence large language models accessible via cloud APIs between OpenAI and challengers who criticize the company for being too restrictive. Meanwhile, Microsoft is emerging as major investor and commercializer across generative AI startups looking to take on OpenAI.
“This is setting up to be a Cambrian explosion moment for generative AI,” remarks industry analyst James Wang of ARK Invest. The GPT Store promises to further expand access and push boundaries on enabling new innovative applications using language models. However, governance frameworks will urgently need to evolve keep pace to manage emerging risks and challenges amplified by such hyper-scaled deployment of increasingly powerful AI capabilities, accessible to exponentially more end users.
To err is human, but AI does it too. Whilst factual data is used in the production of these articles, the content is written entirely by AI. Double check any facts you intend to rely on with another source.