Model Releases
Just because someone puts the word AGI in their benchmark doesn't make it even remotely a reflection of what the general public understands …
Just because someone puts the word AGI in their benchmark doesn't make it even remotely a reflection of what the general public understands by this term. As a layman, I define AGI as a virtual charact
Just because someone puts the word AGI in their benchmark doesn't make it even remotely a reflection of what the general public understands by this term. As a layman, I define AGI as a virtual character that I can reliably ask to do some work a reasonable human with the adequate level of expertise can do, and I can expect that this work will be done the way I described it and reasonable clarfication questions will be asked if needed. Ask yourself a question: would you delegate this AI to negotiate and sign a multi-thousand dollar contract on your behalf, representing your interests with an adequate level of qualification? If the answer is no, it's not AGI. We are going to need a different AGI benchmark. Where is the goalpost moving next?
Source: Gary Marcus (X) | 2026-09-03