用户实测吐槽：Astra 的 Computer Use 点个按钮要近一分钟
一位交易者称 Astra 远非 AGI，Computer Use 反应慢、每 15 分钟就停，实际表现与 Fable 5.1 接近，质疑基准分数意义。
GPT-6 Astra is no damn AGI. Computer Use can literally take close to a minute just to click a button sometimes. It doesn’t take a human that long, damn it. It’s overhyped. The marketing made it sound like it would work autonomously for weeks and completely obliterate Fable 5.1. Guess what? It’s stopping every 15 minutes. In reality, from what I’m seeing, they’re much closer than the benchmarks suggested. Fable even wins on some tasks. This is why I’m caring less and less about benchmarks. ARC-AGI score? Cool. Can the damn agent actually operate my computer reliably, keep working, recover from failures, and finish the job? That’s the benchmark I care about. Give me reliable loops over another benchmark record any day. It can barely do anything in the real world, but sure, call it AGI anyway. We’re imagining that this is going to cure diseases, revolutionize science, give us insane productivity, and completely transform civilization. My biggest fear is that it turns into another Internet 2.0 instead. The physical-world usefulness never lives up to the hype, governments realize that, and then discover that AI is extremely useful for surveillance, manipulation, behavioral control, and some new MK-Ultra style experiment. Like they do with everything. It reminds me of how people imagined the internet in the ’90s versus what we actually ended up with. Or how we imagined cheap nuclear energy and flying cars. You imagine this incredible future-changing technology. Then somehow everything gets turned into algorithms, addiction, TikTok, mass surveillance, and control. That’s the damn idea. We keep imagining technologies that will liberate people and transform the real world. Then we find a way to turn them into systems for controlling people instead.
这条是引用着另一条写的：「黄仁勋：AGI 到了」