Posts

Comments

Comment by nimit on How should TurnTrout handle his DeepMind equity situation? · 2023-10-17T02:33:03.736Z · LW · GW

Assuming your investment portfolio consists of some broad index of stocks, you might modify it to contain every sector except tech, since your google equity compensation makes you over-allocated in tech anyway. So you would be "short" QQQ versus the counterfactual world where you don't have equity compensation. If necessary you could get even more short by buying some SQQQ.

In theory, you could be perfectly indifferent to GOOG by owning SQQQ and all the other stocks in QQQ except GOOG in the correct proportion. Though this probably runs against the spirit if not the letter of any employee trading policy.

Owning some GOOG while being short QQQ probably works pretty well in making you indifferent to GOOG's price in most cases, even over fairly idiosyncratic events like quarterly earnings (though you may have to be more short QQQ in that case). It would fall apart if you were deciding whether to press a button that halved/doubled the value of Google (unless you were unreasonably short QQQ). For that case, precommitting to donate seems like the most reasonable scheme? It feels like any CoI from that should be dominated by whether pressing the button is good for humanity.

Another option: find one or more persons at other companies who also receive stock compensation. Commit to shorting each other's stock so that as a group you have 0 exposure, and then donating everything to charity. This leaves even charity contributions perfectly hedged.

Comment by nimit on Mech Interp Puzzle 1: Suspiciously Similar Embeddings in GPT-Neo · 2023-07-17T01:44:57.914Z · LW · GW

Thinking outloud a bit, I haven't followed mechanistic interpretability besides skimming some transformer circuits posts and half understanding them  - 

Jung'f gur qvssrerapr orgjrra arb naq tcg-2 fznyy?
- arb unf ybpny nggragvba naq hfrf n qvssrerag cbfvgvbany rapbqvat? 
- ibpno fvmr ybbxf gur fnzr

V xabj va uvtu qvzrafvbany fcnprf enaqbz irpgbef fubhyq or begubtbany. Fb jung'f qvssrerag urer?
- gur rzorqqvatf zhfg or nyy pyhfgrerq arneol fbzrubj
- vzntvar nyy ohg gur ynfg 10 qvzrafvbaf ner svkrq ng 1, gung jbhyq qb gur gevpx, ohg vg frrzf hayvxryl gur zbqry jbhyq yrnea guvf
- pbhyq nyfb or nyy gur rzorqqvatf unir n pbafgnag ba gurz, r.t. vafgrnq bs h naq i jr trg (p+h) naq (p+i)

Bx abj V jnag gb fubj gur zbqry pbhyq pbzchgr gur fnzr guvat jvgu na nqqrq pbafgnag va gur rzorqqvat. 
nggragvba ba (p+h) sebz (p+i) vf fbzrguvat yvxr (p+h)^G Jdx (p+i)
= p^G Jdx p + h^G Jdx p + p^G Jdx i + h^G Jdx i

Fb nyzbfg gur fnzr guvat, rkprcg bar fznyy rkgen grez qrcraqvat ba h. 
- vg'f cebonoyl svar gb vtaber guvf? 
- gur ZYCf pbhyq whfg unir n yrnearq ovnf juvpu fhogenpgf bhg p, fb gubfr ner gbgnyyl svar

tcg2 fznyy -- V thrff vg whfg unf n fznyyre p, znlor whfg ol enaqbz punapr bs ubj vg jnf vavgvnyvmrq? Ohg vs guvf vf n obahf dhrfgvba, gur nafjre fubhyq or zber vagrerfgvat. Fbzrguvat gb qb jvgu gur cbfvgvbany rapbqvat be gur nggragvba zrpunavfz? fueht.