Isn't the whole point of notes to address the uncertainty of valuation? In your example, what happens when there's no 'fair price today'? If you think there's a 30% chance Amazon will be worth 3k in a year, what price would you pay now?
Would it be more appealing if you the round was closed in the next few months, so you investing early would be well rewarded with the discount?
I agree that 20% discount by itself is not appealing enough incentive for the risk of early stage investment. But what about 40%? ...50%? A discount is just the mechanism that most accurately captures the variable nature of valuation. Finding the right number is where the modeling happens.
As was mentioned elsewhere, smaller discounts tend to apply to shorter horizons, such as when bridging immediately into a round. In those cases, they can be a good way for VCs to get on the cap table with prorata rights without having to muck around in series-A drama.