%20(1).png)
One month after its CEO Sam Altman’s firing and subsequent rehiring, OpenAI released a set of guidelines for gauging which AI models could present “catastrophic risk” to humans. The “Preparedness Framework” means to address the gap between the scientific study of AI risk and the reality of its current state. Open AI’s monitoring and evaluations team plans to review each new model to assign it one of four risk levels: low, medium, high, or critical, in four categories. The first pertains to cybersecurity, the second to a model’s capacity to help humans cause wide-scale harm to others (i.e. through a virus or a nuclear weapon), the third to its potential influence on human behavior, and the fourth to its potential autonomy, especially unshackled. According to the company's framework, only models deemed low or medium risk should be released.
Post-risk assessment, the team will submit their findings to OpenAI’s Safety Advisory Group. Final decisions regarding risk reduction will be left to the head of OpenAI, with overruling powers in the board of directors’ hands. For context, Altman’s firing by OpenAI’s board was partly motivated by his favoring of accelerated AI development, with little consideration for the threats it could represent. He was rehired a few days later after staff and investors protested the decision, and board members have since been replaced.
Read more on Techxplore.
Coincidentally, an AI company named Verses penned an open letter published in the New York Times seeking support from OpenAI after a “significant internal breakthrough” towards artificial generative intelligence, or AGI. Verses specifically mentioned a breakthrough in Active Inference, a concept that seeks to provide a mathematical framework for understanding live sentience. This evolution could help to make large language models (LLMs) more reliable and beneficial for humankind, according to Verses, and eventually lead to AGI. In its open letter, Verses claimed that it qualifies for OpenAI’s assistance in its project, while also professing its commitment to “a safety-conscious project.”
Much like competitors Deepmind, Anthropic, and others, OpenAI has made no secret of its intention to develop AGI. In the mission statement published on its website, the company states that, “Creating safe AGI that benefits all of humanity.” Unlike current AI models, which tend to be designed towards a specific set of tasks, AGI would - in theory - match or outperform human intelligence. Some experts fear that the advent of AGI would imply that Singularity has been reached (i.e., the attainment of human-like artificial intelligence) and that the algorithm would then inevitably evolve into superintelligence, even without human intervention. Other experts have expressed doubt that AGI could be reached at all. In the meantime, only time will tell whether Verses’ open letter is a genuine call for collaboration… or a mere publicity stunt.
Find more on the Drum.
After almost 40 hours of negotiations, EU representatives in Brussels agreed on a historic set of rules for the use of Artificial Intelligence (AI): the AI Act. This agreement marks the beginning of a new era after years of unbridled development, rife with potential misuse, that prompted warnings from influential figures, including Bill Gates, Elon Musk, and Sam Altman, as mentioned above. “We had one objective to deliver a legislation that would ensure that the ecosystem of AI in Europe will develop with a human-centric approach respecting fundamental rights, human values, building trust, building consciousness of how we can get the best out of this AI revolution that is happening before our eyes,” said MEP Brando Benefei of Italy.
By defining a risk-based tiered system for the industry with matching levels of regulation, the AI Act aims to foster ethical AI development. It includes a robust governing framework for generative AI models, AI-driven surveillance, biometric technologies, and transparency standards, among others. Expected to set a baseline for other governments who have yet to set regulations on AI, the AI Act effects might spill over into non-EU markets. As Anu Bradford, a Columbia Law professor, said to AP: “After all, it is not efficient to re-train separate models for different markets.” This agreement follows a series of strong rulings from the EU to regulate the tech industry, such as the Digital Markets Act and the Digital Services Act.
More info on the Tech Portal and the Guardian.
Digital privacy has been quite the hot topic in recent years - unambiguously demanded by consumers, yet rarely granted by tech giants. In 2023, however, the state of affairs seems to have shifted in consumers’ favor, as major tech companies have recently improved their privacy protection practices.
No stranger to consumer privacy, Apple further reinforced its devices’ privacy protections and introduced “contact key verification” in iOS 17.2, allowing users to verify Messenger chats and thus avoid possible scams from imitators. Google announced that Google Maps location history will soon be saved on one’s device directly, instead of the cloud, for reinforced privacy. And notably, Meta started rolling out end-to-end encryption for Facebook Messenger chats, one of its biggest privacy improvements wto date.
Read Fast Company for more information.
Discover all the latest news, articles, webinar replays and fifty-five events in our monthly newsletter, Tea O'Clock.
.svg)
.jpg)
