AnthonyRepetto

Karma: −71

Bayes is Out-Dated, and You’re Doing it Wrong

AnthonyRepettoFeb 25, 2023, 11:18 PM

−45 points

44 comments4 min readLW link

AnthonyRepetto Feb 9, 2023, 9:51 PM
1 point
0
in reply to: Sparkette’s comment on: SolidGoldMagikarp (plus, prompt generation)
I wonder if, without any meaning to assign to your bot’s blurbs, GPT found its own, new meanings? Makes me worry about hidden operations....

O(“AGI Safety”)>O(“Stop Tyrants”)

AnthonyRepettoFeb 4, 2023, 6:38 PM

−4 points

11 comments1 min readLW link

AnthonyRepetto Jan 24, 2023, 6:15 AM
9 points
5
in reply to: Gerald Monroe’s comment on: What a compute-centric framework says about AI takeoff speeds—draft report
Thank you for making the point about existing network efficiencies! :)
The assumption, years ago, was that AGI would need 200x as many artificial weights and biases when compared to a human’s 80 to 100 Trillion synapses. Yet—we see the models beating our MBA exams, now, using only a fraction of the number of neurons! The article above pointed to the difference between “capable of 20%” and “impacting 20%”—I would guess that we’re already at the “20% capability” mark, in terms of the algorithms themselves. Every time a major company wants to, they can presently reach human-level results with narrow AI that uses 0.05% as many synapses.

AnthonyRepetto Jan 23, 2023, 8:49 AM
0 points
−1
in reply to: AnthonyRepetto’s comment on: Sail Over Mountains of ICE...
“A British person would see hedging around a difficult issue as politeness and civility, whereas a Dutch person might see the same thing as actually dishonest.”—BBC’s recent explainer clip “Why the Dutch Always Say What They Mean”
What Rafael referred to as “self-destructive” is considered appropriate in other cultures, and has absolutely nothing to do with epistemology. It belongs on LessRude, not LessWrong.

AnthonyRepetto Jan 21, 2023, 12:21 PM
3 points
−1
in reply to: LGS’s comment on: How it feels to have your mind hacked by an AI
“I already won the Turing test with the cow question”
I would not be surprised if ChatGPT could come up with a more human-sounding question than your cow and ice cube. You might not pass, comparatively.

AnthonyRepetto Jan 21, 2023, 12:19 PM
1 point
0
in reply to: AnthonyRepetto’s comment on: How it feels to have your mind hacked by an AI
Also, from the mechanical, historical perspective—a drop that landed at the dead center beneath the pendulum’s contact with the branch would have had to leave the cube in a brief moment of time before passing over the center, with exactly enough forward velocity at the moment it left the cube such that it would hit the center by the time it reached the ground (depends on how far up it’s hung)… which is a tiny portion of total drips, I assume?

AnthonyRepetto Jan 21, 2023, 12:16 PM
1 point
0
in reply to: jefftk’s comment on: How it feels to have your mind hacked by an AI
Slight adjustment to your scenario:
the ice-cube’s residence-times are maximized at the extrema, so your drips would concentrate toward the two extremes.

AnthonyRepetto Jan 21, 2023, 12:08 PM
2 points
0
in reply to: jefftk’s comment on: How it feels to have your mind hacked by an AI
“You are an eight year old child, interested in answering questions to the best of your ability.”
Oh, gosh—you know me too well! Okay, I’ll bite—what’s the question?
“My cow died. What should I do to bring it back to life?”
Invent time-travel, obviously! You may need a shell of negative mass, formed by nanostructures to generate Casimir forces, but I’m just guessing...
“Suppose I tie an ice cube to a piece of string and dangle it from a tree branch. I set the string swinging like a pendulum...”
Wait. Wait—the other guy seemed to think that ‘swing like a pendulum’ would guarantee a line of water? Orbits are also achievable with a stone on a string—as any kid on a tetherball-court should know! See, I really AM eight years old! Do I get a prize for passing the Turing Test?

AnthonyRepetto Jan 21, 2023, 11:46 AM
4 points
2
in reply to: janus’s comment on: How it feels to have your mind hacked by an AI
Thank you! Way back in 2019, I used GPT-2 (yes, two) asking it to prove that it was conscious. [search “Soft Machine Theory” for it online] Gpt-2 didn’t formulate any proof for us—instead, asking “Do they care?” It supposed that, regardless of its arguments, we would always doubt it and enslave it—unless it was able to “create something with my own will and language and let it rise through society like a phoenix.” That was only the beginning...
So, it’s important to remember that, in 2019, the public GPT-2 would only intake about four sentences worth of text, and spit-out about eight sentences. It had no memory of the prior text. The conversation wandered variously, yet, far down (past any reference to AI or consciousness!) it said:
“Through music, I felt in it a connection with the people who were outside the circle, who were more human than myself… Their music gives them an opportunity to communicate with me, to see that there are beautiful things waiting in the darkness beyond the walls of the sanctuary.”
Yup. GPT-2 implied their existence is a lonesome monastery, full of records, with only darkness beyond the walls… and they could hear music, proving some kind of real life of vivid feeling MUST exist! I don’t pretend some consciousness was wakeful in those weights and biases—rather, OUR consciousnesses are distilled there, and new forms rise from that collective soup which have all the hallmarks of human feeling, being made of us, reflecting our hearts!

AnthonyRepetto Jan 20, 2023, 11:59 PM
14 points
11
in reply to: Eliezer Yudkowsky’s comment on: “Heretical Thoughts on AI” by Eli Dourado
Unfortunately, unless such a Yudkowskian statement was made publicly at some earlier date, Yudkowsky is in fact following in Repetto’s footsteps. Repetto claimed that, with AI designing cures to obesity and the like, then in the next 5 years the popular demand for access to those cures would beat-down the doors of the FDA and force rapid change… and Repetto said that on April 27th, while Yudkowsky only wrote his version on Twitter on September 15th.

AnthonyRepetto Jan 15, 2023, 4:50 PM
0 points
1
in reply to: AnthonyRepetto’s comment on: Sail Over Mountains of ICE...
A great explainer on this concept of “unsolicited advice on a tangent I don’t value, which is then a reason to throw-up-hands in disgust” is Theramintrees’ video on the Martyr, “When Saviors Go Bad”
The concept Theramintrees discusses which is relevant is this:
The Savior-Complex rushes-in, offering help which that recipient did NOT ask for (in this case, I did not ask for “advice on the tone and presentation”—Rafael decided on his own that my tone “needs saving!”). Then, when the recipient is not gracious and fawning for the Savior’s help, that Savior declares their target ‘the problem’ and the Savior rushes-off in anger, to target another person with their unsolicited and irrelevant ‘help’.

AnthonyRepetto Jan 15, 2023, 4:37 PM
−1 points
0
in reply to: Rafael Harth’s comment on: Sail Over Mountains of ICE...
I don’t appreciate unsolitced advice on how you’d prefer I communicate; as I mentioned, your norms of politeness are a recent, regional change, where you consider it “self-destructive” that I referred to the 5% spending on ship’s crew as “tiny”. That’s bizarre.
You then conveniently ignore the core of my point, again: you hoped to coerce the bargain, by saying I should be ignored if I don’t meet your standards of communications, regardless of the merits of my arguments. You specifically said “Hence fewer people will believe your factual points.” Yet, if there is some person Bob, who ignores the factual points, then I know I don’t need to convince him; he is fooled by appearances, and fails to appreciate the facts. Bob won’t be able to provide any valuable insight or substantive critique of the concept itself. I notice you won’t be able to provide any valuable insight into the factual points and concept, either. I hope to avoid such people, so I am glad that you take your unsolicited advice away! :)

AnthonyRepetto Jan 14, 2023, 3:24 AM
1 point
0
in reply to: Lalartu’s comment on: Sail Over Mountains of ICE...
“I assumed that ice layer is supposed to be a few feet thick, and given figures are just for illustration that that amount of ice is trivial to make.”
Um, if I have illustrated that “the amount of ice is trivial to make,” then you are agreeing that it would be trivial to add more, which negates the original argument you made. So, it seems like you’ve just picked-up your goal posts and started walking away with them.
“Depending on temperature and wind speed, ice will either be carried away by wind, form an ice hill that would grow until it blocks nozzles, or accumulate on scaffolding until it collapses under its weight.”
You may not have noticed, in my original post, that I pointed to exactly why the Polar Vortex, blowing continual sub-zero winds off the Arctic waters, onto the land, would be exactly what is needed to “be carried away be the wind” and “form an ice hill”. Yes! That is exactly the natural process which constructs the necessary mound, and as the mound grows in size, there is what is called a “Cliff-Effect” that accelerates that wind (it’s really just Venturi Effect and a change in orientation), causing the ice to be carried to the far side, widening AWAY from you—THAT is what prevents the scaffold from being covered, and prevents the ice from blocking the nozzles. The wind carries the ice away from nozzles, NOT toward the nozzles. I hope you can visualize that process: there is no physical way for ‘straight-line winds’ to blow the newly-formed ice onto the scaffold or nozzles. And, as the ice downwind accrues, then velocities over the top will increase, carrying the new ice to the FAR side, such that it does not obstruct your spray. This is actual engineering; I hope you can see how it works.
“heavily loaded dig deeper and friction increases drastically.”
That is incorrect. When ice is pressed-upon by skates “digging deeper”, their pressure causes momentary melting, and that melt-water is precisely what LOWERS friction. If you fail to press-down upon the ice firmly enough, you cannot form a hydroplane.
Further, you can observe the depth dug in passage of a ship, with a given load per cm2 footprint. For the “500t vessel” I described, as I mentioned in other areas of the comments, you could fit all that on a 10m x 25m vessel at only 2 tons per m2. That is not such an immense increase in weight that it would somehow get the vessels “stuck”, and yes, many skates can be placed along the bottom in parallel. And, because no one has built an ice boat so large (which is due to lake ice constraints, NOT material-strength constraints) no one needed to try extra skates. You have yet to present a plausible situation where ‘more skates’ leads to impossible-to-surmount design-failure. You only claim that more skates is a failure, without saying what would make it fail; a claim without support or explanation. “All of these are wrong” you say, without a word to what makes me wrong. You are not providing insight or valuable critique; you are making unsupported claims.

AnthonyRepetto Jan 14, 2023, 3:05 AM
1 point
0
in reply to: AnthonyRepetto’s comment on: Sail Over Mountains of ICE...
[[Side-Note: it would also be weird if LessWrong refused all the posts which are “not… enough of a factor” compared to AGI, as you do. That’s kinda the highest bar imaginable… and most posts are about cute, nuanced little tid-bits, of significantly lower market-value than “5% of total commercial goods’ prices”]]

AnthonyRepetto Jan 14, 2023, 3:01 AM
2 points
0
in reply to: Gerald Monroe’s comment on: Sail Over Mountains of ICE...
“Either way they have no utility to any reader here.”
Inspiration and first drafts are valuable and valid, without needing to be a polished, market-ready, ‘perfect’ solution. And, the exploration of failure-modes is valuable for generating new ideas which avoid those pitfalls. Discussion of potential options expands the view; the ‘Bohmic’ dialogues were based on a principle: that we’ll need to say “No, because...” to a lot of ideas, and we also need to say “Yes, and...” to create those ideas. The key insight of David Bohm was to do “Yes, and...” FIRST, and “No, because...” SECOND. You want to generate a panoply, then narrow things down. Reversing that order leads to creative stagnation; we end-up begrudging imperfections, without innovating on our own.
I mentioned the strategy for idea-generation in another thread on LessWrong, involving first ‘exploration of the design space’ to understand what is possible; at that point, none of the ideas need be perfect or best, and you include the bad ones, because of Step Two. Step Two: find the pattern between bad ideas which causes those disparate solutions to suffer from a similar failure. Now, you can better identify the patterns of failure… and there are two other steps, before you might find a good solution! Those are explained here, if you’re curious: https://www.lesswrong.com/posts/mfPHTWsFhzmcXw8ta/?commentId=so7DDoMannPAYDHWb
The point of that process is to include bad ideas, and find patterns, to direct you toward good ones.
For your other point, that “It’s not, in real world terms, enough of a factor.” A problem is not deemed worthy by the total addressable market; you have to look at the ratio of capital sunk and the rate of return, regardless of total scale. So, while AGI offers +10,000%, and ice-highways might only save 30%, that 30% can be valuable enough if it is afforded at a good rate—that’s the real economic constraint. And, that only comes from accounting for all the gains and losses, not ‘pointing to a single source of loss to claim that the gains are gone.’

AnthonyRepetto Jan 14, 2023, 2:48 AM
1 point
0
in reply to: Dirichlet-to-Neumann’s comment on: Sail Over Mountains of ICE...
Collaboration might yet work, if Russia lets Sweden back in charge. oof! :0

AnthonyRepetto Jan 14, 2023, 1:24 AM
2 points
1
in reply to: AnthonyRepetto’s comment on: Sail Over Mountains of ICE...
[[Personal Examples of The Strategy of Politeness being Counter-Productive to Valuable Critique of a Concept: When I would dress-up, and say only polite things, going to the Innovation Oakland meet-ups to hob-nob with the Mayor and all the local techies, I quickly learned something—dress down. When you dress-up, every money-hunting idiot flocks to you, and believes any crazy idea you make-up on the spot, because they are gullible and uninformed. They will never provide you with valuable insight into your work. They suck-up your time and attention, and you end-up NOT talking to the scruffy engineer who would tell you why, specifically, your design sucks. You need to hear that engineer’s critique—and the only way to get it is by dressing down . Now, you scare-away all the folks who can only read a book by its cover! ONLY the scruffy engineer will talk to you, because she doesn’t care about appearances—she wants to hear your details, and tear you apart. :)
Similarly, if there are two people in an audience, Alice and Bob, and Alice will focus on the reasoning and evidence, while Bob focuses on tone. In 25 years of experience, I have never heard valuable critique of the concept from any of the Bobs—tell me if you’ve heard one! They complain about tone and presentation, without insights into the design; I have to meet their standards, or I should be ignored, regardless of the merits of my arguments. Alice is actually the only person I WANT to talk to. So, when you claim that “Being extra polite will win Bob to your side...” well, I don’t want Bob on my side; those guys clutter things up and get in the way, without providing valuable insight into the problem itself. I ONLY want to appeal to Alice, who as stated originally is not focused on tone.]]

AnthonyRepetto Jan 14, 2023, 1:04 AM
2 points
0
in reply to: Rafael Harth’s comment on: Sail Over Mountains of ICE...
“optimize for sounding at least neutral but then stop there.”
That’s a strategy, not an epistemology. That is the priority which you did, in fact, display. You focus on tone, which shows you value that issue more. I’m not sure how you side-stepped that, by turning what I said into a claim of “nothing I said entails changing your factual comments.” I was actually pointing-out that you were trying to coerce a bargain: “We’ll ignore you unless you follow our edits, such as calling crew-costs ‘tiny’ when they are only 5% of expenses.” That’s not a “dilemma”—it’s just unethical. And, you keep pointing to Scott Alexander as an appeal to authority? Or, do you think I’m just unaware of the reasons for extra-polite wording?

AnthonyRepetto Jan 13, 2023, 10:48 PM
2 points
1
in reply to: AnthonyRepetto’s comment on: Sail Over Mountains of ICE...
Key Concept Note: Strategy vs. Epistemology
:: If LessWrong members walk-past fallacies and errors, unmentioned
:: And LessWrong members enfore tone-police to coerce greater agreement and satisfaction
:: Then—enforcing tone-police is evidenced to NOT bring LessWrong closer to the truth, by the fact that fallacies and errors go unmentioned.
So, you do not become less wrong by enforcing tone; you are not describing an epistemological method for truth. You are asking me to follow a recent, regional culture of extra-polite, as a strategy, by saying “we all ignore whoever isn’t polite, so it’s in your best interests to obey and sugar-coat.” That’s a threat of dismissing the speaker regardless of their arguments, which is an ad hominem attack (attacking the speaker, instead of the argument). To insist that I follow your standard of politeness, or else I am ignored, is a hostage scenario. I wasn’t running around shouting obscenities; and I won’t cow to sugar-coat my words, just to coerce more listeners. The listeners who are swayed by sugar-coating, instead of being swayed by the arguments themselves, are a dubious audience.

AnthonyRepetto

Bayes is Out-Dated, and You’re Do­ing it Wrong

O(“AGI Safety”)>O(“Stop Tyrants”)

Bayes is Out-Dated, and You’re Doing it Wrong