by 
This week I get into Artificial General Intelligence (AGI) and start to debunk what it is, what it’s not, and if we’ve achieved it. This week on the Brief we cover OpenAI’s newest model GPT-6 Astra, explain NYC’s ban on AI in schools and why it’s a great idea, and close with the FTC’s newest lawsuit against Amazon.
I am confident that 99.9% of you reading this have dabbled with LLMs and interacted with AI models. I’ll extend that assumption into believing that each of you, at one point in interacting with an AI, had a moment of, “wow, this thing is amazing!” I’d also wager that when an AI outputted a mistake, hallucination, or something blatantly wrong (but confidently-delivered), you thought, “wow, this thing is dumb!” That’s the state of AI right now, and why the binary benchmark of “have we reached AGI?” is an impossible question to answer.
In my opinion, comprehending why the goalposts on AGI keep moving is to understand the difference between a “task” and a “job.” Just because an AI can now perform 1,000,000+ ‘tasks’ better than a human can doesn’t mean the same AI could do the full single “job” that a human is employed to do. According to Microsoft’s research (taken with a grain of salt), a human makes 35,000 decisions a day. While these decisions are not created equally and most are second nature and subconscious, I am confident the greatest AI today would not have a 100% success rate on those 35K decisions. I’m not sure it would even cross a 50% success rate. And until an AI can do all the same things a human could do, and do them exponentially better, I don’t think we’re at a purist’s definition of AGI.
More importantly, do we even want this AGI future?! This is the question I don’t believe the Frontier AI labs can answer concisely without a philosophical or savior-complex-laced answer. OpenAI thinks GPT-6 Astra is AGI. Go test it out and let me know if you think we’ve crossed the threshold.