May We Take A Moment?
热度趋势
趋势数据积累中
百分比基于当前可用热度信号，而非评论数或独立用户人数。
OpenAI 相关模型动态已经出现，适合跟踪能力变化、生态影响和后续可用性。
人工智能社区对机器智能的看法一直在不断变化，随着技术的进步，其标准也在不断提高。图灵测试曾是衡量机器意识的基准，但现在被认为存在缺陷。GPT-3 最初被称为“准 AGI”，展示了人工智能能力的巨大飞跃，有人认为，2016 年的观察者会认为当前的尖端模型就是 AGI。这种对人工智能进步的快速适应，凸显了对真正人工智能构成要素的持续重新评估。
Prior to ChatGPT, the turing test was typically considered to be the defining moment; the event that marked when we could no longer doubt machine awareness anymore than our own. Does anyone here even remember when models started passing it? What model was it? I feel like crossing this threshold was a blip in time and the immediate consensus was, "that's actually a flawed and easily gamed test". I'm not debating this idea, but it doesn't change the fact that we, as a community, as a society, have been quick to move the goal posts as we've become desensitized to the current state of the art.
I'd like to remind everyone that GPT-3, not ChatGPT/3.5, was referred to by the community as proto-AGI. If you were to have shown someone in 2016 a current frontier model, they would have likely considered it AGI. As someone that's been obsessed with AI since I was a child, I remember the moment I read GPT-3 output a convincing and coherent 4chan copypasta (cringe I know, but that was the moment) and realized we had entered a new era.
I constantly see posts in the vein of "it's not AGI until I see x" or "maybe by 2040, likely later". We're watching incremental improvements on benchmarks and half of us are scoffing every step of the way. I'm not a twitter hype train personality, but I can't help but shake the feeling, moreso the last few weeks, that we're climbing on the event horizon and many of us will be clinging to the graph and rationalizing away its existence.
I'm currently fullfilling my childhood daydreams and far fetched ideas by writing a few paragraphs into a terminal and pressing enter. I doubt there are many, if any, frontier researchers that don't at least consult a frontier model as a tool. Many high end developers I know are now telling me of all the cool projects, features, ideas, etc that they've made a reality rather than complaining about tracing bugs.
I suppose this is something I just needed to get out as someone who lurks this sub every day. I feel like we need to appreciate the moment we're witnessing and the shift that we're in. Sometimes it's hard to see it from one day to the next, but I'd like to have real discussions about it rather than alternate between comments that are "WOOO AGI NOW ACCELERATE" and "AGI will never exist, stochastic parrot" etc.
I personally was always in the camp that biological realism, such as Spiking Neural Networks, would have been required, or at least the best way, to achieve real intelligence. I still believe in the benefit, but I'm starting to change my mind a bit.