If it's worth saying, but not worth its own post (even in Discussion), then it goes here.
I might as well point out my solution- I've set the date of the Austin meetup to be six years from now, and edit the date each week. It stays on the map, it stays on the sidebar (so I remember to edit the date- if this were automatic, then it could be correct), and it stays out of discussion.
Recent work shows that it is possible to use acoustic data to break public key encryption systems. Essentially, if one can send specific encrypted plaintext then the resulting sounds the CPU makes when decrypting can reveal information about the key. The attack was successfully demonstrated to work on 4096 bit RSA encryption. While some versions of the attack require high quality microphones, some versions apparently were successful just using mobile phones.
Aside from the general interest issues, this is one more example of how a supposedly boxed AI might be able to send out detailed information to the outside. In particular, one can send surprisingly high bandwith even accidentally through acoustic channels.
Yesterday I noticed a mistake in my reasoning that seems to be due to a cognitive bias, and I wonder how widespread or studied it is, or if it has a name - I can't think of an obvious candidate.
I was leaving work, and I entered the parking elevator in the lobby and pressed the button for floor -4. Three people entered after me - call them A, B and C - but because I hadn't yet turned around to face the door, as elevator etiquette requires, I didn't see which one of them pressed which button. As I turned around and the doors started to close, I saw that -2 and -3 were lit in addition to my -4. So, three floors and four people, means two people will come out on one of the floors, and I wondered which one it'll be.
The elevator stopped at floor -2. A and B got out. Well, I thought, so C is headed for -3, and I for -4 alone. As the doors were closing, B rushed back and squeezed through them. I realized she didn't want -2, and went out of the elevator absent-mindedly. I wondered which floor she did want. The elevator went down to -3. The doors opened and B got out... and then something weird happened: C didn't. I was surprised. Something wasn't right in my idle deductions. I figured it out in the few seconds it took for the elevator to descend to my floor and let me out together with C.
Where did I go wrong? When I knew that B left on -2, I deduced, correctly, that C will get out on -3. But then B came back; the fact of her leaving on -2 turned out to be wrong; yet I didn't cancel my deduction about C and didn't return him the "freedom" of leaving either on -3 or on -4. It didn't even occur to me to do that. Why didn't it?
It seems important that the new information was a correction of a known fact, and not just some other fact. If I treat the new information "B does not leave at -2" purely as a fact, the consequence for C is "C may leave either on -3 or on -4", which is already clear as it is and not worth updating. No, it seems "B does not leave at -2" has a special character when it comes to correct the previously-assumed "B left at -2". It comes as a "rollback" of existing information and I need to "roll back" everything I deduced from that information. And that seems hard to do and easy to forget. So if wasn't just a failure to update that I committed. It was a failure to "roll back".
On reflection, this mistake seems like something we might be doing often, and something to keep an eye out for. Is there a name for this mistake, has it been studied?
As there was some interest in Soylent some time ago, I'm curious what people who have some knowledge of dietary science think of its safety and efficacy given that the recipe appears to be finalized. I don't know much about this area, so it's difficult for me to sort out the numerous opinions being thrown around concerning the product.
ETA: Bonus points for probabilities or general confidence levels attached to key statements.
Reproduced for convenience...
On G+, John Baez wrote about the MIRI workshop he's currently attending, in particular about Löb's Theorem.
Timothy Gowers asked:
Is it possible to give just the merest of hints about what the theorem might have to do with AI?
Qiaochu Yuan, a past MIRI workshop participant, gave a concise answer:
Suppose you want to design an AI which in turn designs its (smarter) descendants. You'd like to have some guarantee that not only the AI but its descendants will do what you want them to do; call that goal G. As a toy model, suppose the AI works by storing and proving first-order statements about a model of the environment, then performing an action A as soon as it can prove that action A accomplishes goal G. This action criterion should apply to any action the AI takes, including the production of its descendants. So it would be nice if the AI could prove that if its descendants prove that action A leads to goal G, then action A in fact leads to goal G.
The problem is that if the AI and its descendants all believe the same amount of mathematics, say PA, then by Lob's theorem this implies that the AI can already prove that action A leads to goal G. So it must already do the cognitive work that it wants its smarter descendants to do, which raises the question of why it needs to build those descendants in the first place. So in this toy model Lob's theorem appears as a barrier to an AI designing descendants which it both can't simulate but can provably trust.
I'm not ready for my current employer to know about this, so I've created a throwaway account to ask about it.
A week ago I interviewed with Google, and I just got the feedback: they're very happy and want to move forward. They've sent me an email asking for various details, including my current salary.
Now it seems to me very much as if I don't want to tell them my current salary - I suspect I'd do much better if they worked out what they felt I was worth to them and offered me that, rather than taking my current salary and adding a bit extra. The Internet is full of advice that you shouldn't tell a prospective employer your current salary when they ask. But I'm suspicious of this advice - it seems like the sort of thing they would say whether it was true or not. What's your guess - in real life, how offputting is it for an employer if a candidate refuses to disclose that kind of detail when you ask for it as part of your process? How likely are Google to be put off by it?
I work at Google. When I was interviewing, I was in the exact same position of suspecting I shouldn't tell them my salary (which I knew was below market rate at the time). I read the same advice you did and had the same reservations about it. Here's what happened: I tried to withhold my salary information. The HR person said she had to have it for the process to move forward and asked me not to worry about it. I tried to insist. She said she totally understood where I was coming from, but the system didn't allow her flexibility on this point. I told her my salary, truthfully. I received an offer which was substantially greater than my salary and seemingly uncorrelated with it.
My optimistic reading of the situation is that Google's offer is mostly based on approximate market salary for the role, adjusted perhaps by how well you did at the interviews, your seniority, etc. (these are my guesses, I don't have any internal info on how offers are calculated by HR). Your current salary is needed due for future bookkeeping, statistics, or maybe in case it's higher than what Google is prepared to offer and they want to decide if it's worth it to up the offer a little bit. That's my theory, but keep in mind that that it's just a bunch of guesses, and also that it's a big company and policies may be different in different countries and offices.
I think it is worth mentioning that "the system won't allow for flexibility on this" is just about the oldest negotiation tactic in the book. (Along with, "let me check with my boss on that...")
In reality, there is zero reason Google, or any employer, should need to know your current or past salary information apart from that information's ability to work as a negotiation tactic in their favor.
Google has something you want (a job that pays $) and you have something they want (skill to make them $). Sharing your salary this early in the process tips the negotation scales (overwhelmingly) in their favor.
That said, Google is negotiating from a place of immense strength. They choose from nearly anyone they want, while there is only one Google...
...so, if Google wants to know your salary, tell them your salary. And enjoy your career at one of the coolest companies around. You win. :)
I'm a manager at a financial firm and I've hired people. I'd consider it pretty normal not to want to say. "Everyone" knows that trying to get the other person to name a number first is a common negotiating tactic, no real grownup is going to take it personally or get upset about this.
I don't know how "normal" a company Google is in this way, but I'd guess it's pretty normal.
If you are challenged on this, you can try stating it as a rule: "I'm not prepared to discuss my current salary, I'm here to talk about working for Google." Or, "As a policy I don't disclose my current salary. I'm sure you understand." Or make up some blah about how that's proprietary information for your current employer and you don't feel comfortable disclosing it.
If they absolutely refuse to process your application without this (which is a bad move on their part if they really want you, but some companies are stubborn that way), other options are to fudge your number upwards somehow, though personally I wouldn't try the ones that actually involve telling a literal lie:
Give them a wide range of expectations instead of your current salary. Say that of course it depends on the other details of the offer, any other offers you might get, etc.
Roll in as much stuff as you plausibly can (adding in bonuses or other moneylike benefits, and making an adjustment if the cost of living in Googleland is higher than where you live now). Example: I could add my salary of $80k, my last year's bonus of $5k (or next year's bonus, or my average bonus in percentage terms, whichever is highest), and my $2k transit benefits for a total of $87k.
Round up to the nearest $10k and say it's an approximate figure. So if I make $87k I might say I'm in the ballpark of $90k.
State a range (e.g. if I made $69k I might say I make something in the high 5 figures, or somewhere near the $70-80k range)
Lie outright, but plausibly.
The rationale behind salary negotiations are best expanded upon by patio11's "Salary Negotiation: Make More Money, Be More Valued" (that article's well worth the rent).
In real life, the sort of places where employers take offense by you not disclosing current salary (or generally, by salary negotiations -that is, they'd hire someone else if he's available more cheaply) are not the places you want to work with: if they're putting selection pressure for downscaling salaries, all your future coworkers are going to be, well, cheap.
This is anecdotally not true for Google; they can afford truckloads, if they really want to have you onboard. So this is much more likely to come from standardized processes. Also note in Google's case, that decisions are delegated to a board of stakeholders, so there isn't really one person who can be put off due to salary (and they probably handle the hire/no hire decisions entirely separate to the salary negotiations).
Something's brewing in my brain lately, and I don't know what. I know that it centers around:
-People were probably born during the Crimean War/US Civil War/The Boxer Rebellion who then died of a heart attack in a skyscraper/passenger plane crash/being caught up in, say WWII.
-Accurate descriptions of people from a decade or two ago tend to seem tasteless. (Casual homophobia) Accurate descriptions of people several decades ago seem awful and bizarre. (Hitting your wife, blatant racism) Accurate descriptions of people from centuries ago seem alien in their flat-out implausible awfulness. (Royalty shitting on the floor at Versailles, the Albigensian Crusade, etc...)
-We seem no less shocked now by social changes and technological developments and no less convinced that everything major under the sun has been done and only tweaks and refinements remain than people of past eras did.
I guess what I'm saying is that the Singularity seems a lot more factually supported-ly likely than it otherwise might have been, but we won't realize we're going through it until it's well underway because our perception of such things will also wind up going faster for most of it.
We seem no less shocked now by social changes and technological developments and no less convinced that everything major under the sun has been done and only tweaks and refinements remain than people of past eras did.
I do expect the future to be different.
I could imagine a future where people see illegalizing LSD as strange as illegalizing homosexuality.
I can imagine that Google's rent an AI car service will completely remove personal ownership of cars in a few decades. This removes cars as status symbols with means that they will be built on other design criteria like energy efficiency.
I can imagine a constructed language possibly overtaking English.
There are a lot of other things that are more vague.
I hope MIRI is thinking about how to stop Johnny Depp.
(YouTube version for people who don't want to download QuickTime.)
Huh, a major Hollywood movie about superintelligence, uploading and the Singularity that seems like its creators might actually even have a mild clue of what they're talking about. Trailers can always be misleading, of course, but I'll have to say that this looks very promising - expect to enjoy this one a lot.
Sidestepping the question:
Interview with other companies (Microsoft, Facebook, etc.) and get other offers. When the competition is other prospective employers, your old salary won't much matter.