Tuesday, 2 September 2025
28.3 C
Singapore
29.9 C
Thailand
20.5 C
Indonesia
27.5 C
Philippines

OpenAI unveils initial guidelines for future AI interactions

OpenAI releases the first draft of Model Spec to guide future AI interactions, emphasising helpfulness, legality, and societal respect.

OpenAI, a leading entity in artificial intelligence development, has launched the initial version of its “Model Spec” framework. This new guideline aims to shape how AI tools, including the widely used GPT-4 model, will operate in future interactions. The company has outlined three foundational principles designed to ensure AI models are beneficial and uphold social norms and laws.

The framework established by OpenAI mandates that AI models should provide useful and instruction-compliant responses to both developers and end-users. By assessing potential advantages and drawbacks, these models aim to benefit humanity. Additionally, they are expected to positively reflect OpenAI by adhering to societal norms and regulations.

Comprehensive rules and public engagement

OpenAI’s Model Spec encompasses several specific rules: AI must follow established command chains, comply with relevant laws, avoid disseminating hazardous information, respect the rights of creators, ensure user privacy, and refrain from producing not-safe-for-work (NSFW) content. The framework also introduces a flexibility feature, allowing users and companies to adjust the “spiciness” of AI responses, particularly concerning NSFW content, which OpenAI is considering making available in age-appropriate settings through its API and ChatGPT.

Joanne Jang, a product manager at OpenAI, emphasised the framework’s role in distinguishing intentional actions from bugs in AI behaviour. The guidelines propose that AI models should always assume positive intentions from users, engage in clarifying dialogue, maintain objectivity, promote respectful interactions, and express doubts when necessary.

“We believe these guidelines will facilitate more nuanced discussions about AI models, including legal compliance and ethical considerations,” Jang told The Verge. “This clarity will hopefully simplify policy discussions, making it easier to determine what should be escalated to our policy team.”

An evolving framework awaiting public input

While the Model Specification will not immediately affect currently active OpenAI models like GPT-4 or DALL-E 3, which follow existing usage policies, it marks a significant step towards more refined AI governance. OpenAI plans to update the document regularly based on feedback from various stakeholders, including policymakers, trusted institutions, and domain experts. However, the company has not specified a timeline for the release of a second draft.

As this new framework takes shape, OpenAI is actively seeking public feedback to ensure it aligns with broader societal values and effectively supports the company’s mission of responsible AI development. OpenAI’s approach underscores its commitment to pioneering the field of AI behaviour, which Jang refers to as a “nascent science.”

Hot this week

ChatGPT referral traffic to websites drops by 52% in one month

ChatGPT referral traffic to websites has dropped 52% in a month, as Reddit and Wikipedia rise under OpenAI’s new citation weighting.

Kobo introduces Instapaper integration to replace Pocket on e-readers

Kobo replaces Pocket with Instapaper on its e-readers through a free firmware update, ensuring users maintain a seamless read-it-later experience.

OpenLight raises US$34 million to scale photonics for AI data centres

OpenLight secures US$34M Series A to scale integrated photonics for AI data centres, expanding its PDK and accelerating product innovation.

Yooka-Replaylee remaster launches on consoles and PC on 9 October

Yooka-Replaylee launches on 9 October with digital and physical editions, Switch 2 full cartridge, enhancements, and a 30% upgrade discount.

Apple’s first foldable iPhone tipped to launch with major design changes

Apple’s first foldable iPhone is expected to debut with an 8-inch foldable screen, eSIM-only support, and a high-quality hinge for its 20th anniversary.

FIFAe partners with Lenovo to power FIFAe Finals 2025 in Riyadh

FIFAe partners with Lenovo to power the FIFAe Finals 2025 in Riyadh, featuring top esports teams competing on Legion gaming devices.

Canon Singapore and NLB expand cartridge recycling programme with new green pledge

Canon Singapore and NLB expand cartridge recycling to more libraries with the Inkfinity Green Pledge, supporting Singapore’s Zero Waste goals.

FairPrice Group opens AI-powered Store of Tomorrow at Punggol Coast Mall

FairPrice Group has opened the world’s first AI-powered supermarket at Punggol Coast Mall, transforming shopping and staff experiences.

Banks urged to balance opportunities and risks of agentic AI

Moneythor warns that while agentic AI offers exponential potential for banks, the risks are just as significant without safeguards.

Related Articles

Popular Categories