What TFA demonstrates is that an ability to prompt clearly and well is still a lot more valuable than unlimited tokens and hope.
The prompt they used was poor (what does growth mean over the limited period - user base or revenue?), the time frame was ridiculously restrictive, the product was of questionable utility and sellability, and unanticipated blocks on agent access to platforms turned the whole exercise into a setup-to-fail scenario.
The prompt was fine for the specific narrow goal. It's a business, so growth automatically means earn more by default. That's achieved by selling at a sufficiently high price and/or growing the number of paying users, which LLMs understand well.
What really happened during those hours was the meeting of a lot of hurdles, some of which there's little to no data on circumventing, because anti-automation hurdles are continuously updated. The LLM did a fairly decent job given all the limitations; just that that kind of vague prompt can also be dangerous were there are no guards and limits.
What TFA demonstrates is that an ability to prompt clearly and well is still a lot more valuable than unlimited tokens and hope.
The prompt they used was poor (what does growth mean over the limited period - user base or revenue?), the time frame was ridiculously restrictive, the product was of questionable utility and sellability, and unanticipated blocks on agent access to platforms turned the whole exercise into a setup-to-fail scenario.
The prompt was fine for the specific narrow goal. It's a business, so growth automatically means earn more by default. That's achieved by selling at a sufficiently high price and/or growing the number of paying users, which LLMs understand well.
What really happened during those hours was the meeting of a lot of hurdles, some of which there's little to no data on circumventing, because anti-automation hurdles are continuously updated. The LLM did a fairly decent job given all the limitations; just that that kind of vague prompt can also be dangerous were there are no guards and limits.
Wait until the AI learns about enshittification