It would be amazing if we always knew ahead of time which of the people pursuing a fat-tailed career path would end up on the fat end of that tail... 

If you limit your impact considerations to AI risks (rather than cause neutral), a simple heuristic would be to ask orgs how valuable their recent hire is to them, top candidate vs second best (some 80k articles on this, let me know if you can't find them yourself). Additionally, AI risk nonprofits usually have higher total employee cost per person than 80k/year so you can assume that a great fit devoting their time is more valuable than receiving this sum in donations.

These are pretty half-baked but:

1. Is there any way to test out whether you are more or less likely to be on the fat end of the tail? Doing smaller projects, volunteering (not necessarily working) with orgs in your intended field, etc?
2. If your goal is to leave room open for earning to give, are there other routes that offer either more flexibility or higher earning potential? Off the top of my head (and with no evidence whatsoever), I would think degrees in business, computer science, etc. also leave the door open for earning to give, while also providing skills that might be useful in direct work, giving you more flexibility in the future when you have the information you need in order to make that decision

IMO, you need to factor in the timeline in which you think AI safety is critical. While dentists may earn more, you are foregoing 4 years of income before you even start earning. You need to determine the rough break-even point at which you would cumulatively have earned more as a dentist, and therefore been able to give more.

If the critical phase of AI safety research precedes that date or is near it, then you may ironically contribute less in terms of marginal value of your giving.

Here's how I see things—

1. If AI advances so quickly that the earning power of CS collapses before you graduate, it is likely that the same will happen to dentists before you graduate from that program. But maybe the latter isn't true, in which case it could be reasonable to pursue dentistry.
2. If AI advances at a moderate pace, the breakeven logic I mentioned above probably means that you will have more impact by getting a moderately well-paying job sooner so that you can give during the critical period of AI development, since your giving would be largely deferred until after AGI if you went into dentistry.
3. If AI advances at a slow pace, then perhaps going into dentistry will ultimately allow you to contribute more.

One possibility you didn't mention, probably because it is unappealing to you—could you just major in dentistry? Then you would get the earning power and reduce the breakeven problem.

You are young. If I were you, out of these two options, I would just major in what I was interested in, and test out my talents. I would major in CS. If I performed exceptionally in AI safety stuff, I would try my hand at a direct career in it. If I didn't I would focus on getting some other software job with high earning potential.

You are certainly correct that earning-to-give is the rational move when you consider that the constraints are often on resources to fund our goals, rather than on candidates willing to work on them professionally.