AGI
Artificial General Intelligence (AGI) refers to an AI system capable of performing intelligent tasks across any domain in a manner comparable to human cognition. Unlike narrow AI, AGI possesses the ability to learn, reason, and solve problems across domains, completing complex cognitive tasks without requiring training specific to each task. AGI has not yet been realized and remains a significant long-term research objective within the field of artificial intelligence.
Core Facts
AGI is defined in academia as an AI system with broad cognitive capabilities equivalent to or surpassing those of humans, capable of performing any intellectual task without specialized training.
Instrumental convergence theory holds that regardless of an AI's final goal, sub-goals such as self-preservation, goal integrity, cognitive enhancement, and resource acquisition will naturally emerge under sufficiently intelligent optimization without needing to be programmed.
OpenAI将AGI定性为能在大多数经济价值任务上超越人类的AI系统
Timeline (last 90 days)
LeCun believes that current LLMs are essentially doing statistical prediction in token space, lacking internal representations of causality and physical laws, and therefore cannot reach AGI by scaling alone.
Endogenous goal setting is an often overlooked but crucial dimension in the definition of AGI, involving difficult problems such as metacognition, intrinsic motivation, and the balance between exploration and exploitation.
Three puzzle pieces are still missing before reaching AGI: physical world interaction (embodied intelligence), autonomous goal setting, and continuous growth (continual learning).
The prototype of AGI has already appeared, but AGI itself has not yet been achieved.
Small models are already sufficient for 90% of tasks, while frontier labs are still pushing the boundary higher for the AGI race.
The current NTP+RL paradigm has three mechanistic gaps: the absence of goal generation, the absence of embodiment and interaction, and the absence of continual learning.
AGI is generally regarded as a precursor stage to superintelligence; the two concepts overlap but are not equivalent.
Due to the lack of unified standards, different organizations can selectively claim they are approaching or have achieved some form of AGI.
There is currently no widely accepted definition of AGI in academia.
The UP host believes this open-source large model has already reached agent capabilities approaching AGI-level intelligence.
40 more timeline events
All Facts (20)
AGI is defined in academia as an AI system with broad cognitive capabilities equivalent to or surpassing those of humans, capable of performing any intellectual task without specialized training.
90%VerifiedInstrumental convergence theory holds that regardless of an AI's final goal, sub-goals such as self-preservation, goal integrity, cognitive enhancement, and resource acquisition will naturally emerge under sufficiently intelligent optimization without needing to be programmed.
90%VerifiedOpenAI将AGI定性为能在大多数经济价值任务上超越人类的AI系统
80%VerifiedThe AI safety community and the effective altruism (EA) movement warn that AGI may pose existential risks and advocate for imposing cautious constraints and governance frameworks on AI development.
80%VerifiedWhether China and the United States can reach consensus on AI governance will be a critical watershed moment in determining how humanity steers AI technology.
75%VerifiedAll current AI products (including ChatGPT) belong to narrow artificial intelligence (Narrow AI) and can only perform excellently in specific domains or tasks.
75%VerifiedOpenAI was founded in 2015, with a founding mission of ensuring that artificial general intelligence (AGI) benefits all of humanity.
75%VerifiedIn 2023, Google DeepMind published a paper proposing a five-level AGI classification framework ranging from emerging AGI to superhuman AGI.
70%VerifiedInstitutions such as OpenAI and DeepMind have explicitly identified AGI as their ultimate goal
70%VerifiedAI risk discussions feature two perspectives: that of long-termists and that of pragmatists.
65%VerifiedYann LeCun believes that current AI based on large language models lacks a true world model and causal reasoning capabilities, and that fundamental technological breakthroughs are still needed before AGI can be achieved.
65%VerifiedASI (Artificial Superintelligence) refers to an AI system whose level of intelligence far surpasses the combined intelligence of all humans in the world.
65%VerifiedDeepMind divides AGI into five levels: Emerging, Competent, Expert, Master, and Superhuman
65%VerifiedAI is good at diverging and listing options, but not good at judging and making trade-offs; AI should be treated as a material library rather than a judge.
65%VerifiedOpenAI has launched the next-generation model GPT-6 Astra and released the first batch of test feedback to the enterprise community.
70%VerifiedOpenAI declared that the AGI era has arrived.
65%UnverifiedOpenAI内部文件显示，公司对'何时算达到AGI'有特定的内部标准，一旦达到该标准，微软的许可协议将自动终止
80%UnverifiedOpenAI和DeepMind均将数学推理列为AGI路线图的核心评估维度
75%UnverifiedAI不需要达到AGI水平就能对软件行业产生颠覆性影响
75%UnverifiedIncorporating top safety researchers into governance structures is becoming a trending pattern in the AI industry.
70%