0.5%的准确率数据揭示了当前AI模型与通用人工智能(AGI)之间巨大的能力差距。这个极低的分数表明，尽管AI发展迅速，但在真正理解复杂推理方面仍处于非常初级的阶段。作者用讽刺的语气质疑行业过度炒作AGI进展的现象。

【洞察】120 倍的人机差距意味着：当前 AI 推理能力的提升是「在已知模式上的优化」，而非「真正的归纳推理泛化」。这对所有声称「AI 已接近人类」的产品宣传都是正面挑战——AGI 时间线的预期需要重新校准，而非渐进式调整。

⚠️【令人震惊的数字】最强前沿模型得分低于 0.5%——而非专业人类轻松超过 60%，差距超过 120 倍。这是继 ARC-AGI-2 之后最彻底的「AI 能力幻觉清醒剂」。推理能力的提升并未自动迁移到「新颖抽象推理」，当所有人在讨论 AGI 即将到来时，这份数据是最直接的反驳。

这是一个值得记录的重要信息，表明作者对AGI的达成持高度信心，并预测将在2035年左右实现。

大多数人认为AI进步主要体现在特定领域的知识获取和模式识别上，而非跨上下文的推理和长期行动能力。但作者强调GPT-5.5在需要持续推理和行动的领域取得显著进步，这一观点挑战了AI能力发展的主流叙事，暗示通用智能可能比预期更早实现。

作者提出AGI需要两个关键要素：持续学习能力和从模式匹配到理解因果关系的能力。这一观点挑战了当前AI发展路径，暗示我们可能过于关注规模和数据，而忽视了真正的理解能力。

yes, this is why I think the AI hype is tech's coping strategy in the face of climate change. A figleaf for inaction.

yes, again a coping strategy. AGI soon is a great excuse to do whatever you want now bc AGI will clean everything up next year. AI is a cope cage much like a tinfoil hat.

Yes, a dream that forgets to factor in that a world without workers would (currently) also mean a world without consumers.

Also, possible missing word.

for - progress trap - ASI - progress trap - AGI

for - quote - AGI - Weapon of Mass Destruction

quote - As we get closer to superintelligence, - it will be seen more and more as an enabler and driver of - weapon of mass destruction (WMD) capabilities, - if not as a WMD in and of itself. - Direct calls for a “Manhattan Project for AGI” are already starting.

for - progress trap - AGI - one year away from automating software work

for - progress trap - AGI

for - progress trap - AGI - Silicon Valley move fast and break things strategy - incompatible with security of AGI

Interesting position paper about how to have useful discussion about AGI and Foundation models.

for - article - Techradar - Top AI researcher says AI will end humanity and we should stop developing it now — but don't worry, Elon Musk disagrees - 2024, April 7 - AI safety researcher Roman Yampolskiy disagrees with industry leaders and claims 99.999999% chance that AGI will destroy and embed humanity // - comment - another article whose heading is backwards - it was Musk who spoke it first, then AI safety expert Roman Yampolskiy commented on Musk's claim afterwards!

for - stats - comparison of cognitive powers - AGI AI agents vs human researcher

stats - comparison of cognitive powers - AGI AI agents vs human researcher - 100 million AGI AI researchers - each AGI AI researcher is 100x more efficient that its equivalent human AI researcher - total productivity increase = 100 million x 100 = 10 billion human AI researchers! Wow!

for - progress trap - debate - nobody is discussing the dangers of such a project!

progress trap - debate - nobody is discussing the dangers of such a project! - Civlization's journey has to create more and more powerful tools for human beings to use - but this tool is different because it can act autonomously - It can solve problems that will dwarf our individual or even group ability to solve - Philosophically, the problem / solution paradigm becomes a central question because, - As presented in Deep Humanity praxis, - humans have never stopped producing progress traps as shadow sides of technology because - the reductionist problem solving approach always reaches conclusions based on finite amount of knowledge of the relationships of any one particular area of focus - in contrast to the infinite, fractal relationships found at every scale of nature - Supercomputing can never bridge the gap between finite and infinite - A superintelligent artifact with that autonomy of pattern recognition may recognize a pattern in which humans are not efficient and in fact, greater efficiency gains can be had by eliminating us

for - key insight - AGI as automated AI researchers to create superintelligence

key insight - AGI as automated AI researchers to create superintelligence - We will reach a period of explosive, exponential AI research growth once AGI has been produced - The key is to deploy AGI as AI researchers that can do AI research 24/7 - 5,000 of such AGI research agents could result in superintelligence in a very short time period (years) - because every time any one of them makes a breakthrough, it is immediately sent to all 4,999 other AGI researchers

for - key insight - AGI in next 5 to 10 years or bust

key insight - AGI in next 5 to 10 years or bust - As we start approaching billion, hundred billion and trillion dollar clusters, hardware improvements will slow down due to - cost - ecological impact - Moore's Law limits - If AGI doesn't emerge by then, then we will need to have major breakthrough in - architecture or - algorithms

Yann LeCun released his vision for the future of Artificial Intelligence research in 2022, and it sounds a lot like Reinforcement Learning.

• difference between
  • ChatGPT (AI)
  • AGI

Behavior here is experience, information that is stored in the memory and retrieved for reflection and learning to happen. Does that mean Believable Agents or Generative Agents can essentially become aware of their own existence and potentially begin to question and compare the virtual/internal environment with the external environment ?

It is unclear what form the "alignment property" would take, and most importantly how such a property would be evaluated especially if there's an arbitrary divide between "dangerous" and "pre-dangerous" levels of capabilities and alignment of the "dangerous" levels cannot actually be measured.

Interesting sounding high level paper about the limits and constraints on general intelligence and how this might relate to the struggles AI/ML research has had historically.

Clicked through to the essay. It appears to be mainly an argument for a super-powerful benevolent general artificial intelligence, of the sort proposed by AGI-maximalist Nick Bostrom.

The money quote:

The only way to avoid having all human values gradually ground down by optimization-competition is to install a Gardener over the entire universe who optimizes for human values.

🔗 This is a great New Yorker profile of Bostrom, where I learned about his views.

🔗Here is a good newsy profile from the Economist's magazine on the Google unit DeepMind and its attempt to create artificial general intelligence.