Posts

Comments

Comment by Zian on Open and Welcome Thread – June 2021 · 2021-06-18T00:14:29.051Z · LW · GW

Potentially trivial math question:

Imagine that you have a computer with the following properties:

  • flat
  • black
  • hot
  • speck of red on one side

And your prior at this point for p(slow given it's a computer) = 0.5.

If an article about computers said p(slow given flat, black, and speck of red) = 0.25, then would you use the new # as your prior or combine the two pieces of information to calculate p(slow given flat, black, hot, and speck of red)?

I'm inclined to say that I should use .25 as the new prior and forget about 0.5 but I may also be making a silly logical error here.

Comment by Zian on Power Outage Chances · 2021-05-16T21:40:35.757Z · LW · GW

You could look for something that can provide a more relevant reference set for your situation. For example, you point out that power outages have different probability distributions based on your location in time and space. 

You could also look at others' revealed preferences.

My direct answer your question of estimating an outage for the average person in the US over the next 10 years is that the question is not very useful.

I'll flesh this comment out with a specific example to show what I meant with the first two paragraphs as I have time to do so. By publishing the comment early, I intend to provide some quick value and prevent computer bugs from eating the post.

Goal:

p(>3d outage for a residential location in the next 10 years in downtown San Diego, CA)

Thought Process:

Sources of power outage information:

Potential Primary Sources:

  • Angry articles and social media posts about extended power outages in the last 10 years
  • Other studies of power outages in San Diego County, the City of San Diego, the ZIP code containing downtown San Diego, or the census tract(s) that comprise downtown San Diego
  • Raw counts of power outages in San Diego County, the City of San Diego, the ZIP code containing downtown San Diego, or the census tract(s) that comprise downtown San Diego
  • Reports of telecom outages
  • Reports of datacenter or server outages

Likely Locations of the Information:

  • Local newspapers (searchable via databases)
  • Reddit.com (searchable via Reddit and Google)
  • Patch.com (Google and site-search)
  • Local online-only newspapers (searchable via Google and site-search)
  • San Diego Gas & Electric
  • California Public Utilities Commission
  • California Air Resource Board
  • Just ask the data source directly

Revealed Preferences:

  • Type of power redundancy in use by self-interested entities that want to always have power
    • City of San Diego
    • County of San Diego
    • Local public safety agencies
    • Local datacenters
    • Telecommunications companies
    • Hospitals
    • Transit agencies
    • Jails
  • Sources of information about the revealed preferences:
    • Satellite imagery
    • RFPs, purchase orders, and financial statements
    • Just ask them

From the CPUC:

SDG&E had 68.64 SAIDI minutes in 2019 (data is available going back to 1997) excluding MED. There were 122.96 SAIDI minutes including MED.

Assuming that ">3 day outage" = a MED, then we can do a little algebra.

Given:

SAIDI = Total minutes every customer was without power due to sustained outages / total number of customers

SAIDI without MED = Total minutes every customer was without power due to sustained outages - total MED minutes / total number of customers

68.64 = (Min - Med) / C

122.96 = Min/C

MED minutes = 1358 * # of customers / 25

Since we want to know the p for a given customer, then we further calculate:

MED minute / customer in a year = 1358/25 * # of customers / # of customers

= 54.32 minutes

Therefore:

p = 1.033E-4 that any particular minute will be without power during a major power outage in San Diego, CA

Over the span of 10 years, you'd expect to sit through just over 9 hours of lonely extended darkness.

From a sewage treatment plant RFP:

The "[fuel] tank shall hold 3 days of diesel fuel storage at 70% generators load" for the "two Owner Supplied Trailer Mounted 2MW Caterpillar XQ2000 backup generators".

If we could find the power consumption of the Penasquitos Sewer Pump Station, then we could calculate (3 * amount of fuel consumed at 70% load of a 2MW Caterpillar XQ2000 generator * 2) / (daily power consumption) to find the # of expected consecutive blackout days.

Let's pretend that # is "3" and simplistically assume that it corresponds to 3 days of doom per year. We'll also multiply by 2 as an extra safety factor because the City can probably get fuel easier than normal people.

p = 3 days / 365 days  * 2

= 0.016

Comment by Zian on MIRI location optimization (and related topics) discussion · 2021-05-10T07:34:04.639Z · LW · GW

I'm both up voting you and commenting because I used to live (more than 4 years) and continue to monitor (in a very real time way) the DFW Metroplex. Then, I moved to Southern California, so I've kind of made the MIRI move in reverse.

Beware of motivated reasoning when it comes to things in TX that you think will change for the better soon. For example, if you had listed Dallas as a possibility, I would be warning against counting on improved DART service.

If you find a positive attribute in a particular area of Austin, make sure the other positive and negative attributes about Austin still apply; things can be quite different as you move around the state. Administrative boundaries (e.g. School attendance boundaries) may surprise you.

Strongly recommend looking into sources of expected life expectancy and healthcare costs for all locations being considered.

Comment by Zian on Why are the websites of major companies so bad at core functionality? · 2021-05-09T19:09:20.467Z · LW · GW

Then I have a better answer for the question about how I would Goodhart things.

Let U_Gurkenglas = the set of all possible x that may be randomly checked as you described Let U = the set of all possible metrics in the real world (superset of U_G) For a given action A, optimize for U_G and ignore the set of metrics in U that are not in U_G.

You will be unhappy to the degree that the ignored subset contains things that you wish were in U_G. But until you catch on, you will be completely satisfied by the perfect application of all x in U_G.

To put this b concrete terms, if you don't have a metric for "nitrous oxide emissions" because it's the 1800s, then you won't have any way to disincentivize an employee who races around the countryside driving a diesel truck that ruins the air.

(the mobile editor doesn't have any syntax help ; I'll fix formatting later)

Comment by Zian on Why are the websites of major companies so bad at core functionality? · 2021-05-08T22:57:59.638Z · LW · GW

I'm trying to think through your point using the above stapler and office supplies example.

If you hired an aggressively lazy AI to buy office supplies for you and you told it : Buy me a stapler.

Then it might buy a stapler from the nearest trash can in exchange for one speck of dirt.

Then, you would go to Staples and buy a new ergonomically friendly stapler that can handle 20 sheets (model XYZ from Brand ABC) using your credit card with free five day shipping.

You proposed that we would reward the AI by calculating the distance between the two sets of actions.

I don't see a way for you to avoid having to invent a friendly AI to solve the problem.

Otherwise, you will inevitably leave out a metric (e.g. Oops, I didn't have a metric for don't-speed-on-roads so now the robot has a thousand dollar speeding ticket after optimizing for shipping speed).

Comment by Zian on Prediction-based medicine (PBM) · 2021-04-19T06:36:53.540Z · LW · GW

Many conditions have self-assessment questionnaires used in the research community. They could be given to patients at intervals.

Comment by Zian on What are some real life Inadequate Equilibria? · 2021-04-19T06:25:53.477Z · LW · GW

An EY category #3 example from emergency medical services (EMS) in the USA

>How things are currently, and why that's bad.
When someone tells a public safety access point (PSAP) that there might be a problem, the dispatcher generally sends an ambulance at a minimum. Then, the city or ambulance provider takes the nearest obvious patient(s) to the emergency room. The resulting EMS bill is often extremely high as well as the ER bill. Sometimes, the providers are not in-network and the patient must pay in full even if there was not a true emergency.

>How they could be instead, and why that's better.
The PSAP could pre-commit to only sending certain types of units when a set of criteria are met. And, if none of the criteria are met for a public safety response, then the patient could be given self-care instructions ranging from watchful waiting to calling the patient's doctor. If a crew is sent to the scene, then the crew could send the patient to the most appropriate place (ranging from "stay home for a week" to "Uh oh, you really do need an emergency room."). This would result in greatly reduced spending by the PSAP, public safety agencies, ambulance providers, hospital, patient, and other healthcare-related entities. The patient would also have better outcomes due to decreased financial stress, decreased risk of iatrogenic infections, etc.

>What's blocking the transition from 1 to 2.
A public safety agency/ambulance provider is only paid when they transport a patient to the emergency room because that's what all the insurance companies have agreed to do.

If a hospital dares to tell an ambulance provider that they can send patients to another type of location, the hospital will lose money.

If an ambulance crew does not transport a patient to an emergency room, the crew takes on liability for practicing unconventional medicine.

Dispatchers are generally not permitted to send anything less than an ambulance and a fire engine. Sometimes, ambulances are only provided at the most costly level (ALS). The crew may even be required to go lights and sirens as a matter of policy despite the increased risk of death and injury. But if a city does not do a "full response", then it is on the hook if anyone complains about reduced responses (aka political fallout) and there may be legal liability for the reduced response. Dispatchers are also often not RN or MDs. So they may not be permitted to definitively diagnosis or recommend treatment outside of a very narrow set of pre-defined situations.

For what it's worth, Coronavirus disease 2019 has forced many cities to change their practices because there are not enough units. CMS has also had on-again-off-again efforts to send patients to alternate destinations (e.g. https://www.cms.gov/newsroom/fact-sheets/emergency-triage-treat-and-transport-et3-model) but many such programs eventually end and the local city must find alternative grant funding in the long-term. As usual, the necessary know-how has existed for decades (e.g. https://en.wikipedia.org/wiki/Medical_Priority_Dispatch_System#Response_Determinant).

 

April 25, 2021 Update: Today, there is a New York Times article talking about the closure of ambulance services in rural America. As the article says, the situation was entirely foreseeable (article reasonably claims that the direct proximate cause was lowered patient transport volumes from Coronavirus disease 2019).

Comment by Zian on On Sleep Procrastination: Going To Bed At A Reasonable Hour · 2021-04-18T07:07:15.771Z · LW · GW

To expand on this point, I get the impression that you've worked really hard. It may be time to take this problem to a medical doctor to find out if there is something they can suggest such as a diagnosis of one of the many conditions that result in sleep phase delay a la HPMOR.

Comment by Zian on Rationalism before the Sequences · 2021-04-05T19:33:12.476Z · LW · GW

3

My knowledge before reading the article and comments could be summarized as :

  • These are some really great articles by ESR. I wonder why no one had taken them super seriously yet...
  • somewhat of an outsider perspective as FeepingCreature described
  • I wonder why some people have such strong opinions about this person
Comment by Zian on Urgent & important: How (not) to do your to-do list · 2021-03-14T06:59:10.806Z · LW · GW

Have you found a need to reserve time to visit the lower-priority categories on a routine basis?

For instance, it is very unlikely that I will run out of "must / later" items. I can think of many tasks to do that will end in disaster if they are not done.

But I also have things that should be done eventually like making small tweaks to my workspace or other things that qualify as moderate improvements (e.g. based on the expected return on investment from https://xkcd.com/1205/).

How do you make time for those types of tasks that shouldn't be ignored forever?

Comment by Zian on How can I protect my bank account from large, surprise withdrawals? · 2021-02-23T03:28:03.874Z · LW · GW

This type of bank account is sometimes called a "controlled disbursement account" (e.g. https://www.wellsfargo.com/com/treasury-management/payables/controlled-disbursement-account). It's often used by businesses. Other people have explained how to turn things into net 30 with a credit card.

Combined, you can maintain some control over the activity in your checking account.

Comment by Zian on Overconfidence is Deceit · 2021-02-18T22:36:35.041Z · LW · GW

Thanks for posting this article here. Sometimes it feels like I got into this rationality stuff too late or only after a lot of people scattered away.

(I hope no one minds that this comment doesn't talk about the article's contents.)

Comment by Zian on Your Cheerful Price · 2021-02-15T07:59:31.858Z · LW · GW

There's a cost to figuring out that someone is trustworthy.

You could think of it this way :

Cost of finding a friendly assisting intelligence + normal cost of a good or service >> cheerful price

(pun intended)

Comment by Zian on How do you optimize productivity with respect to your menstrual cycle? · 2021-02-09T21:12:03.699Z · LW · GW

Real DisabilityTM

That sounds like a really useful idea.

For example, perhaps it would help to quantify one's quality of life over a period to identify patterns.

And maybe there are certain metrics that should lead to certain actions (like the idea of a normal range in routine blood tests).

Comment by Zian on How would free prediction markets have altered the pandemic? · 2021-02-09T21:00:08.196Z · LW · GW

large, liquid, free prediction markets being made suddenly available to the public

I am going to be pessimistic and guess that very little would change because this scenario does not include any change to the type of people who take prediction markets seriously and we already have some prediction markets.

Specifically, I predict that if there were an absolutely perfect prediction market for "Vaccine X will be X% effective" and the resulting prediction (in January of 2020) was mostly 98 to 99, the FDA would not behave any differently from what we observed with Pfizer or Moderna.

Comment by Zian on Open & Welcome Thread - January 2021 · 2021-01-31T06:53:37.395Z · LW · GW

Considering how much LW and the Sequences talk about doing Bayesian updates, I figured it's worth talking about a downside I am experiencing. I closely monitor Metaculus and adjust my vaccination expectations accordingly. I have certain events that are on hold until I get vaccinated. Therefore, the optimal strategy would seem to be:

  1. Look up a predicted vaccination date range (e.g. late April)
  2. Schedule the appointments for late April
  3. Go to step 1.

What I didn't expect was the level of effort required by step 2 after the first run through. It gets tiring to call people every few months and say something like "Oops, we have to reschedule again..."

Comment by Zian on Covid 1/21: Turning the Corner · 2021-01-22T01:44:29.290Z · LW · GW

Another vaccination call center website : https://www.covid19vaccinemi.com/

Comment by Zian on (USA) N95 masks are available on Amazon · 2021-01-19T00:04:28.655Z · LW · GW

Effects of Sterilization With Hydrogen Peroxide and Chlorine Dioxide on the Filtration Efficiency of N95, KN95, and Surgical Face Masks (JAMA Netw Open. 2020 Jun; 3(6): e2012099.) says:

The Centers for Disease Control and Prevention listed KN95 masks (the Chinese version of the N95) as suitable alternatives to N95s when N95s are not available. 

Figure 1 indicates that a KN95 mask has similar filtration efficiency as a N95 when new. However, it acts differently (larger variation) after H2O2 sterilization and becomes very inefficient after ClO2 sterilization.

Comment by Zian on What is going on in the world? · 2021-01-17T18:25:35.582Z · LW · GW

I'm don't think the universe is obliged to follow any "high level narratives". I'm afraid I don't understand how thinking of events in these terms is helpful.

Comment by Zian on How can I find trustworthy dietary advice? · 2021-01-17T17:51:58.298Z · LW · GW

UpToDate is a good starting point.

Comment by Zian on What skills or habits have lasting value through time? · 2021-01-12T05:04:15.823Z · LW · GW

Diligently followup on and solve (or minimize the impact of) health conditions and illnesses

Comment by Zian on What to do if you can't form any habits whatsoever? · 2021-01-10T01:34:25.921Z · LW · GW

Are you referring to habit formation with treated ADHD or untreated? There are lots of studies that find dramatic differences in quality of life depending on your answer to that question.

(I won't even get into optimizing the treatment a la MTA.)

Comment by Zian on Covid 1/7: The Fire of a Thousand Suns · 2021-01-08T05:01:26.002Z · LW · GW

In the interest of not falling prey to the illusion of transparency, I must ask if you are familiar with normal hospital offload times in Southern California.

It was pretty common to have crews tied up waiting hours at emergency rooms even in normal circumstances.

Sample numbers from Central California : https://www.google.com/url?sa=t&source=web&rct=j&url=https://emsaac.org/%3Foption%3Dcom_fileman%26view%3Dfile%26routed%3D1%26name%3DJohn%20Surface%20-%20EMSAAC%202019%20Surface%204-23-19.pdf%26folder%3DLEMSA/Conference%26container%3Dfileman-files&ved=2ahUKEwiH85mTyIvuAhWiCjQIHYZnDhY4ChAWMAd6BAgEEAE&usg=AOvVaw1RolpERTpzHHNVMq1_abxc

Comment by Zian on Covid 1/7: The Fire of a Thousand Suns · 2021-01-08T04:18:48.676Z · LW · GW

full triage mode:

The crews are instructed to stop transporting patients who are unlikely to be resuscitated. In other words, before this point, patients were being transported to hospitals for the sake of warm fuzzies, not utilions.

From the perspective of terminal values, I'm not sure if this change actually changes anything.

Comment by Zian on 100 Tips for a Better Life · 2020-12-23T08:52:39.457Z · LW · GW

The $20 may be a stand in for consumer - hostile behaviors like bad customer service or high ATM fees.

Comment by Zian on 100 Tips for a Better Life · 2020-12-23T08:50:26.800Z · LW · GW

I don't know the best way to phrase this but some of this advice is really foundational (eg exercise, sleep, etc). There should be an item here about rapidly and persistently getting professional medical and mental health when you have trouble with the basics.

Comment by Zian on Open & Welcome Thread - December 2020 · 2020-12-11T01:03:35.110Z · LW · GW

I had a distinctly probabilistic experience at a doctor's office today.

Condition X has a "gold standard" way to diagnose it (my doctor described it as being almost 100%) but is very expensive (time, effort, and money). It is also not feasible while everyone is staying at home.

However, at the end of the visit, I had given him enough information to make a "clinical" diagnoses (from a statistically & clinically validated questionnaire, descriptions of alternative explanations that have been ruled out, etc) and start treating it.

In hindsight, I can see the probability mass clumping together over the years until there is a pile on X.

I'm thankful to Less Wrong-style thinking for making me comfortable enough with uncertainty to accept this outcome . The doctor may not have pulled out a calculator but this feels like "shut up and multiply" & "make beliefs pay rent".

Comment by Zian on Final Version Perfected: An Underused Execution Algorithm · 2020-12-07T05:43:06.909Z · LW · GW

How would you say this approach relates to https://www.lesswrong.com/posts/dJQ7BFz9ZPqstP3an/urgent-and-important-how-not-to-do-your-to-do-list?

You wrote that it's important for your preferences to be fairly stable. I've been using a prioritizing grid (e.g. https://www.beverlyryle.com/prioritizing-grid) to go through my entire do list in 1 fell swoop. I wonder if that's more or less efficient than your approach.

Comment by Zian on What are your greatest one-shot life improvements? · 2020-12-07T05:37:31.741Z · LW · GW

Thanks for mentioning the 14 pairs of shoes. I felt a bit silly when doing something similar but I realized that while it's a hassle to carry lots of boxes to and from the local UPS drop off location, I'm not doing anything that is out-of-the-ordinary (if I were buying shoes in person at say, Nordstroms).

To extend your thought a little...if you have any pain walking around and switching shoes doesn't help or it hurts while walking barefoot, consult a doctor.

Comment by Zian on What are your greatest one-shot life improvements? · 2020-12-07T05:31:53.014Z · LW · GW

From one chronic health person to another...

Consider filling up a second dispenser ahead of time. That way, if you're completely exhausted but it's time to take your medication, you can yank the second box off the shelf and refill things when you're more awake. It also gives you a 1 week buffer to refill your prescription.

Pill counting trays are also helpful if you have to take medications that come in a bottle. Buy one that has Amazon reviews from real pharmacists and put the cap under the spout on the right side to catch any runaway pills. This will dramatically reduce the # of pills that fall to the floor.

Comment by Zian on What are your greatest one-shot life improvements? · 2020-12-07T05:23:22.780Z · LW · GW

>sleep

Buy new blanket that isn't falling apart. Semi-permanently reduced the ugh field.

>health

If I take "Doctor diagnosed X" as strong evidence that I have X, then I should find the latest treatment guidelines/summary articles for X once a year. Led to a possibly permanent treatment algorithm (full or nearly-full symptom management) for 2 chronic problems with little to no side effects. Also caught a medication error with the acquired knowledge. Also a found a treatment that greatly helps (but not eliminates/prevents) another chronic condition.

May also have turned up a life-altering diagnoses but that is to-be-confirmed by real medical experts.

>general knowledge acquisition

Using Sci-Hub. Reduces the "Relevant article in PubMed based on the abstract->paste the PMID->oh, this is useful" loop to <1 min.

If necessary, I can try to quantify the benefits of the above items but it's a little like the RENT song about measuring a year.

Comment by Zian on Why are young, healthy people eager to take the Covid-19 vaccine? · 2020-12-03T00:20:27.999Z · LW · GW

Please don't forget that some young healthy people are essential workers who might not really have a choice about this.

Comment by Zian on Open & Welcome Thread – November 2020 · 2020-11-30T08:14:43.016Z · LW · GW

That's not in the Getting Started or the Questions sections, which are the places I looked.

Comment by Zian on Open & Welcome Thread – November 2020 · 2020-11-28T19:43:37.326Z · LW · GW

I found 2 bugs in the Less Wrong website. Where do they go? (this is the first bug; I couldn't find a place to report problems after looking through the FAQ and home page)

Comment by Zian on Plan-Bot: A Simple Planning Tool · 2020-11-28T05:35:30.255Z · LW · GW

I got the following:

>Imagine that it is one week later and your solution failed! Do you really think your solution will work? (Enter "yes" or "no)

Isn't the answer always going to be "no"? You just told me that it failed.

Comment by Zian on [deleted post] 2020-11-28T05:33:18.079Z

Imagine that you are trying to find a doctor in a particular specialty. You are able to think of 12 possible reasons the doctor might refuse to see you. Some are more probable than others and some are easier to minimize/solve than others. You have 5 or 6 doctors to choose from and the 12 failure modes apply to each of them differently. For instance, Dr. A may have a 25% chance of saying "no new patients" whereas Dr. B might be "50%" and Dr. C may be "80%".  What would be the recommended way to reduce the likelihood of failure without spending an inordinate amount of time mitigating things?

Edit:

1 solution would be to identify the probabilities for all 12 for 2 doctors given what is currently known (for instance, if problem X might derail things 80% of the time but you can think of something that will drive it down to 50%, leave it at 80% for now) and then make a decision tree to figure out how many options should be evaluated and which of the 12 things actually need to be mitigated.

For instance, maybe 8 of the 12 things for Dr. A have low probabilities (1%-5%) and the remaining 4 are 25%-50%. Similarly, for Dr. B, perhaps 11 are 1%-5% and the 12th one is 10%. Then you know that even with this worst case scenario, the odds of both A and B turning you down are very low (e.g. 50% * 10% = .5%) and you don't have to even look at the other doctors. Then you can go back to the list of 12  things and figure out which items will increase your odds for the least amount of effort.

At that point, you can pick the desired probability of success and work at each of the mitigations in turn until you reach the desired number.

Comment by Zian on Covid 11/26: Thanksgiving · 2020-11-27T08:22:18.357Z · LW · GW

I hope you don't mind my saying "You're very welcome."

(Healthcare, public safety)

And thank you very much for providing these regular figures and interpretations.

Comment by Zian on Open & Welcome Thread – November 2020 · 2020-11-22T02:17:51.263Z · LW · GW

inflexible

The call taker may be required to follow an algorithm (e.g. https://prioritydispatch.net/resource-library/). This is not to discount all your points; everything you wrote is likely true too.

Finally, it's possible that the high arbitrary cutoff for evidence is a reflection of the agency's priorities and resources.

Comment by Zian on Covid 11/19: Don’t Do Stupid Things · 2020-11-20T13:48:13.336Z · LW · GW

At the end of the article, you wrote "Adding in police and firemen would make it an even 20." Does adding EMS personnel change the estimate at all?

Comment by Zian on Open & Welcome Thread – November 2020 · 2020-11-09T07:29:46.955Z · LW · GW

Wow, that's a bit strongly worded.

I'm going to have to figure out why the journal article gave those figures. Maybe I should send your comment to the authors...

Comment by Zian on Open & Welcome Thread – November 2020 · 2020-11-09T06:29:52.289Z · LW · GW

I'm trying out Bayes Theorem with a simple example and getting really strange results.

p(disease A given that a patient has disease B) = p(b|a)p(a) / p(b)

p(disease B given existing diagnosis of disease A) = 0.21

p(A) = 0.07

p(B) = 0.01

I get 1.47 or 147%. I know that the answer can't be >=100% because there are patients with A and not B.

Where am I going wrong?

Comment by Zian on Where do (did?) stable, cooperative institutions come from? · 2020-11-04T01:16:32.919Z · LW · GW

One possible hypothesis for the question about the lack of new institutions may be related to the ability to assuage oneself with comparatively meaningless activities. For example, I can now write an angry message on a city's social media page and make myself feel like I've done something about the lack of fire protection. I will receive just as much social affirmation (aka likes) that way compared to cold calling all my neighbors and asking them if they'd like to form a committee to raise our taxes and staff an additional fire station.

Comment by Zian on Rationalist Town Hall: Pandemic Edition · 2020-10-22T08:03:05.182Z · LW · GW

Why do you think California is unusually strict?

Comment by Zian on What posts do you want written? · 2020-10-20T05:05:28.570Z · LW · GW

I've read that Less Wrong attracts people with mental health concerns so articles about using mental health related information may be useful.

Comment by Zian on Why Boston? · 2020-10-12T19:46:09.452Z · LW · GW

You can look at the data from the Leapfrog Group and CMS.

Comment by Zian on Msg Len · 2020-10-12T19:38:12.973Z · LW · GW

But then he'd lose the Strunk and White allusion.

Comment by Zian on If GPT-6 is human-level AGI but costs $200 per page of output, what would happen? · 2020-10-10T07:23:22.311Z · LW · GW

I worry that I'm splitting hairs now because it seems that the AI only needs to be clever enough to generate the following in response to a query :

The answer to your question will be provided more quickly if you provide 1 GB of RAM. (rinse and repeat until we get to an AI box)

Comment by Zian on If GPT-6 is human-level AGI but costs $200 per page of output, what would happen? · 2020-10-09T17:36:18.105Z · LW · GW

Based on "Why Tool AIs Want to Be Agent AI's" by Gwern, I would expect an AGI level GPT-6 to self improve and become a world gobbling AI.

The moment it gets a hint that it could answer better by getting (unknown bit of data from the Internet, extra memory, some other resource), the software's own utility function will push the machine in that direction.

Comment by Zian on Fermi Challenge: Trains and Air Cargo · 2020-10-06T02:18:16.040Z · LW · GW

If I have specific knowledge (counts and measurements) about the passenger and freight rail industries, does that count as "object level"?

Comment by Zian on Some thoughts on criticism · 2020-09-24T00:46:28.729Z · LW · GW

That would likely help with "I’m not feeling secure enough to be properly open to criticism." I predict that the more general interaction you have + Alexei's points, the more likely it is that you'll both feel secure.

There may also be external factors that are harder to overcome like power imbalances (imagine a janitor employed by an janitorial company trying to criticize the CEO in the office building who leaves litter everywhere).