JACE, The AI Agent That Controls Your Computer

Buoyed by $2.9 million and led by tech veterans Daniel Gross and Nat Friedman, Zeta Labs, a brainchild of ex-Meta engineers, has launched Jace, an AI agent powered by LLM and AWA-1 models, aimed toward automating complex browser tasks independently.

Unlike traditional chatbots, Jace doesn’t require constant guidance, executing tasks like booking hotels or managing recruitment pipelines autonomously. It employs a hybrid approach: LLM handles interactions while AWA-1 translates plans into browser actions, overcoming web interface challenges. The system verifies task completion to stop loops and secures user credentials like passwords.

Zeta Labs collaborates with design partners to refine Jace before its industrial release, targeting sectors like ecommerce and recruiting. With plans for a $45/month subscription post-free tier exhaustion, Zeta Labs anticipates meeting high demand from small businesses in search of efficient task automation in various operational domains. Read more.

OpenAI’s CEO, Sam Altman, recently disclosed to shareholders that the corporate is mulling over a governance shift, possibly evolving right into a for-profit profit corporation. This potential change would tweak the nonprofit board’s control, bringing OpenAI according to competitors like Anthropic and xAI. The deliberations are as fluid as they’re ongoing, with multiple scenarios still on the table.

Despite these potential upheavals, OpenAI asserts that its primary mission stays unaltered: developing AI for the greater good, with the nonprofit entity holding a central position. Reported by The Information, it underscores the strategic deliberations inside OpenAI because it navigates its future growth and structure. The corporate has reassured stakeholders of its ongoing commitment to creating useful AI. Read more.

OpenAI’s recent board appointment of ex-NSA chief Nakasone has some, including whistleblower Edward Snowden, questioning the corporate’s commitment to open and transparent AI development. Snowden views this move as a possible “intelligence coup” for the U.S. government, raising concerns concerning the alignment of AI development with national security priorities, potentially on the expense of individual liberties.

Nakasone’s role on the “Safety and Security Committee” fuels speculation about collaborations with the U.S. military, further stoking the fireplace. Meanwhile, internal debates at OpenAI highlight a curious chasm between those fixated on theoretical superintelligences and others grappling with more immediate cybersecurity threats.

Snowden’s broader critique extends to the AI community’s hyper-focus on existential threats, cautioning against politicizing AI models to implement minority viewpoints, exemplified by recent controversies involving Google’s AI image generator. Read more.

Researchers from the University of Glasgow argue that the misleading outputs of ChatGPT and similar large language models (LLMs) must be categorized as “bullshit” somewhat than “hallucinations”. Published in Ethics and Information Technology, their study applies philosopher Harry Frankfurt’s definition of bullshit, which emphasizes an indifference to truth somewhat than intentional deception.

In line with the researchers, LLMs like ChatGPT generate statements without concern for accuracy, prioritizing persuasiveness over factual correctness. This distinction between “hard bullshit” (intentional deception) and “soft bullshit” (indifference to truth) clarifies how these AI systems operate, difficult previous notions that described their errors as hallucinations or misperceptions.

The researchers caution against overestimating LLM capabilities by mislabeling their outputs as hallucinatory, advocating as an alternative for accurate terminology to foster a greater understanding of AI’s limitations and promote responsible scientific discourse. Read more.