"Subprime Intelligence". Edward Zitron makes the case that: "We are rapidly approaching the top of generative AI's S-curve, where after a period of rapid growth things begin to slow down dramatically".

"Even in OpenAI's own hand-picked Sora outputs you'll find weird little things that shatter the illusion, where a woman's legs awkwardly shuffle then somehow switch sides as she walks (30 seconds) or blobs of people merge into each other."

"Sora's outputs can mimic real-life objects in a genuinely chilling way, but its outputs -- like DALL-E, like ChatGPT -- are marred by the fact that these models do not actually know anything. They do not know how many arms a monkey has, as these models do not 'know' anything. Sora generates responses based on the data that it has been trained upon, which results in content that is reality-adjacent."

"Generative AI's greatest threat is that it is capable of creating a certain kind of bland, generic content very quickly and cheaply."

I don't know. On the one hand, we've seen rapid bursts of progress in other technologies, only to be followed by periods of diminishing returns, sometimes for a long time, before some breakthrough leads to the next rapid burst of advancement. On the other hand, the number of parameters in these is much smaller than the number of synapses in the brain, which might be an approximate point of comparison, so it seems plausible that continuing to make them bigger will in fact make them smarter and make the kind of complains you see in this article go away.

What do you all think? Are we experiencing a temporary burst of progress soon to be followed by a period of diminishing returns? Or should we expect ongoing progress indefinitely?

Subprime Intelligence

#solidstatelife #ai #genai #llms #computervision #mooreslaw #exponentialgrowth

"Rapid AI progress surprises even experts: Survey just out"

Sabine Hossenfelder reviews a survey of 2,778 AI researchers. They say 50% chance of "high level machine intelligence" ("comparable to human intelligence") by 2047, but that's down 13 years from the same survey a year ago, which said 2060.

For "full automation of labor", 50% probability by 2120 or so, but that's down almost 50 years from last years' prediction. (So last years' prediction must've been 2170 or so).

I can't help but think, does anybody seriously think it will take that long? I get that the "AGI in 7 months" predictions are a bit hard to take seriously, but still? Do these people not understand exponential curves?

Ray Kurzweil, and before him, Al Bartlett, are famous for saying people extrapolate the current rate of change linearly out into the future, so always underestimate exponential curves. Not implying Kurzweil or Bartlett are right about everything but this does look to me like what is happening, and you would think professional AI researchers, of all people, would know better.

Rapid AI progress surprises even experts: Survey just out - Sabine Hossenfelder

#solidstatelife #futurology #ai #exponentialgrowth #technologicalunemployment

march 2023 #exponentialgrowth and #AI. very very urgent.
https://www.youtube.com/watch?v=xoVJKj8lcNQ

Tim Harford: why we fail to prepare for disasters

Financial Times, April 16, 2020

... Psychologists describe this inaction in the face of danger as normalcy bias or negative panic. In the face of catastrophe, from the destruction of Pompeii in AD79 to the September 11 2001 attacks on the World Trade Center, people have often been slow to recognise the danger and confused about how to respond. So they do nothing, until it is too late.

Part of the problem may simply be that we get our cues from others. In a famous experiment conducted in the late 1960s, the psychologists Bibb Latané and John Darley pumped smoke into a room in which their subjects were filling in a questionnaire.

...

The virus started to feel real to Europeans only when Europeans were suffering. Logically, it was always clear that the disease could strike middle-class people who enjoy skiing holidays in Italy; emotionally, we seemed unable to grasp that fact until it was too late.

...

Finally, there’s our seemingly limitless capacity for wishful thinking. In a complex world, we are surrounded by contradictory clues and differing opinions.

...

What if we’re thinking about this the wrong way? What if instead of seeing Sars as the warning for Covid-19, we should see Covid-19 itself as the warning?

Next time, will we be better prepared?

The dynamics of failed disaster preparedness and response, through the lenses of COVID-19, Hurricane Katrina, and other calamities. Failures in vision, leadership, unheeded warnings, cognitive biases, social normalising.

Harfords wonderful but all-too-brief podcast, Cautionary Tales is highly recommended. Its first episode, featuring the work of Charles Perrow, and airing immediately after Perrow's death last November, especially.

http://archive.is/v1gJ1

#covid19 #cognitiveBias #disasterResponse #risk #manifestation #disasterPreparedness #exponentialGrowth #unheededWarnings #hurricaneKatrina #NewOrleans #TimHarford

COVID-19: A month on, April 5, 2020, and my early-March confirmed-case projections are now ... slightly ... pessimistic.

Global confirmed cases lag projection by about 4 days -- the initial forecast remains quite close.

US confirmed cases lag by about 1 week, which is about one doubling period now (~6 days), vs 2.18 days initially. You can look at current case projections and treat them as next week's reality, for the next few days. These values should continue to diverge.

Deaths are mixed relative to projections. Globally, at 1.2 million cases, there are ~64,700 deaths. By case equivalent, I'd projected only 25k deaths, far fewer, and by date, for April 2, 43k deaths vs. 53k actual, stil 10,000 low. For a naive exponential model that's far closer than I'd have expected.

At ~300k I'd suggested 22k dead, actual is 8,452 for the US, better than projected, though official projections are now for an eventual 100-250k fatalities, quite possibly 1-2 million in worse scenarios.

That's some progress, but relatively slight. A far slower deviation from trend than China showed from my first projections, within 2-3 weeks.

A reminder: at the time I'd posted those projections, the US reported only 288 cases, rest-of-world ex-China, 12,668. The figures now stand at 311 thousand confirmed cases U.S., 1.2 million worldwide, with China's 81,669 cases now a rounding error.

As before, the correct US projection is the one in comments on the linked post, not the post itself, due to formula errors.

https://joindiaspora.com/posts/73d4e930421d0138028e002590d8e506

#covid19 #coronavirus #exponentialGrowth

My math-teaching brother's wish is that 2020 is the year everyone learns that "exponential" doesn't mean "very fast" it means "imperceptibly gradual and then suddenly overwhelming".

-- Richard D. Bartlett @richdecibels@social.coop

https://social.coop/@richdecibels/103850907207907177

#ExponentialGrowth #maths #2020

3Blue1Brown: Exponential growth and epidemics

Video 'splainer:

https://invidio.us/watch?v=Kas0tIxDvrg

#covid19 #covid-10 #coronavirus #ncov2019 #exponentialGrowth #video

COVID-19 A Laycat's US Outbreak Model

This is a non-expert's simple extrapolation of the past 11 days' COVID-19 experience within the US, projecting both further likely spread of the COVID-19 outbreak and the possible actual extent of infected individuals based on a presumed testing lag.

As with my earlier China extrapolation: The real message here is how quickly experience deviates below the projection here, suggesting containment efforts are effective. In the case of China, that began about two weeks after my initial post. I am a space alien cat on the Internet, not an expert.

I've probably fucked up all kinds of things. Cluebats welcomed.

How this model works

I'm using a simple exponential growth formula, and basing the expected number of cases (and deaths) from the 5 March 2020 case and death counts, based on what appears to be native community spread rates through the US from 20 February 2020 through 5 March (the period of visible community spread). This is a short window though one showing rapid growth.

It is overwhelmingly evident that the US does NOT have a solid handle on monitoring, and likely won't for at least another week, possible several. This both makes the data presented and model based on them more uncertain, and means that as monitoring improves, apparent case counts will likely increase rapidly. Again, this reflects experience in China.

Virus behaviour, population behaviour, public health measures, weather changes, sunspots, and timelords could all change things markedly.

Exponential growth function

The fomula for exponential growth is:

y(t) = a * e^(k * t)

See: https://www.mathsisfun.com/algebra/exponential-growth.html

Where:

• y(t): quantity at time t
• a: initial quantity
• e: the natural log constant, about 2.7183
• k: the grow rate per period.
• t: the number of periods.

"Period" here is "days".

We can solve for k:

k = ln(y(t)/a)/t

This gives us the growth rate given two measurements t periods apart.

We can solve for t:

t = ln(y(t)/a)/k

In particular, if we solve for y(t) = 2 and a = 1, we get the doubling time.

I've written a simple gawk script which computes for k and doubling rate, and also projects the weekly (7 day) and fortnightly (14 day) growth rates.

Detection lag

A huge problem within the US is that confirmed cases are lagging actual infection dates by a substantial amount. How long that is is ... not entirely clear, though I'm going to assume a 14 day (two week) lag based on:

• Initial infection is followed by a non-symptomatic period of about a week on average.
• Seeking medical assistance has seen a further lag of several days in getting an appointment / performing a test.
• Test results themselves take 4 days based on information I've seen.

The total lag is about 2 weeks.

I'd suggested that this could lead to as much as a 100-fold understatement of actual cases. Based on current data, that seems pessimistic: it's "only" about 47x greater than the published confirmed cases count -- a number that's moved around considerably, by the way, so don't put too much faith in that either. But it gives an indication.

We also get a doubling time of about 2.2 days, which means that however bad the situation is now, it's going to be twice as bad in a little over 48 hours. When you hear statements that the situation is "rapidly evolving" this is what is being referenced. Things are changing very quickly. Locations which may have low risk today may have a high risk in a day or two.

You should be finalising preparations and supplies runs about now, if not already.

Again: non-expert extrapolation based on early data, a simple model, and many uncertainties. I expect we'll likely see following trend, if not overshooting it, for a week or two, mostly as monitoring catches up to reality. I'm very much hoping we'll start to see a low-side numbers starting about two weeks out (18-22 March), as containment efforts begin to be effective. The caveat is that I don't see effective containment measures being enacted, certainly not on the scale that China performed starting ~22 January. In which case the projection here could well fit actual experience for longer.

As before, I'm posting this as a line in the sand of what my projection was. I hope and expect to be proved wrong on this within a couple of weeks. I'm dying to see how well this matches reality.

The professionals are apparently doing this as well

Dr. Messonier of the CDC mentioned 5 March in an NPR interview that there were numerous groups doing epidemic modelling to try to estimate the actual spread of SARS-CoV-2 within the US, though she pointedly refused to give any numbers herself. I have yet to find any published projections, but would be interested in seeing any.

The script

Hardcoded in (edit to modify) are the initial and current case counts. You'll need to supply days between these measures as well. Data are taken from Wikipedia's 2020 Coronavirus Outbreak in the United States article.

The script calcuates the growth rate, with an arbitrary high and low bound (basically assuming one day more or less error in the reported range -- it's kind of weak sauce but gives some idea of sensitivity), the doubling time, the weekly growth rate, and the 14-day growth rate.

It then produces two reports, one every day for 29 days, the other every seven days for 200 days. Both cut off if the infected population exceeds total US population, given as 330.4 million. Shown are projected deaths, cases, cases at a low or high growth rate, and as "w/ 14 day lag" the possible ground truth of total cases from which confirmed cases are drawn. I'll note that this presently exceeds 10,000 cases, and ... doubles ever 2.2 days or so. A rate which will hit 1,000,000 by 18 March.

By April 25, if present rates continue, the entire US is infected. At the WHO's 3.4% fatality rate, 11.2 million die, and given economic modelling, your retirement fund is trash.

(And then the disease may return in the fall....)

For Rest-of-world, you can substitute in values for that outbreak for a simiilar model. (I've got a separate script for this.) As values are hardcoded, it's a tad inflexible.

## Program Output

Minor reformatting aside, this is output as currently stands.

COVID-19 US Outbreak Model

Assumptions:
- init cases (2020-4-26): 14
- cases (2020-3-5): 175
- deaths (2020-3-5): 11
- daily growth rate: 1.316
- doubling time (days): 2.195
- 7 day growth: 6.83x
- 14 day growth/mon. lag: 46.59x

day	date	deaths	cases	@ lo dbl	@ hi dbl	w/ 14d lag
1	Mar 06, 2020	14	230	224	238	10,726
2	Mar 07, 2020	19	302	287	324	14,113
3	Mar 08, 2020	25	398	367	440	18,569
4	Mar 09, 2020	32	524	470	600	24,431
5	Mar 10, 2020	43	689	602	816	32,145
6	Mar 11, 2020	57	907	771	1,111	42,294
7	Mar 12, 2020	75	1,194	988	1,512	55,647
8	Mar 13, 2020	98	1,571	1,266	2,057	73,216
9	Mar 14, 2020	129	2,067	1,621	2,800	96,331
10	Mar 15, 2020	171	2,720	2,076	3,811	126,744
11	Mar 16, 2020	224	3,579	2,659	5,186	166,760
12	Mar 17, 2020	296	4,709	3,405	7,057	219,409
13	Mar 18, 2020	389	6,196	4,360	9,603	288,680
14	Mar 19, 2020	512	8,152	5,584	13,068	379,821
15	Mar 20, 2020	674	10,726	7,151	17,784	499,736
16	Mar 21, 2020	887	14,113	9,159	24,201	657,511
17	Mar 22, 2020	1,167	18,569	11,729	32,933	865,098
18	Mar 23, 2020	1,535	24,431	15,021	44,816	1,138,224
19	Mar 24, 2020	2,020	32,145	19,236	60,987	1,497,580
20	Mar 25, 2020	2,658	42,294	24,635	82,992	1,970,390
21	Mar 26, 2020	3,497	55,647	31,548	112,938	2,592,474
22	Mar 27, 2020	4,602	73,216	40,402	153,688	3,410,959
23	Mar 28, 2020	6,055	96,331	51,740	209,142	4,487,854
24	Mar 29, 2020	7,966	126,744	66,261	284,604	5,904,742
25	Mar 30, 2020	10,482	166,760	84,856	387,295	7,768,965
26	Mar 31, 2020	13,791	219,409	108,670	527,038	10,221,752
27	Apr 01, 2020	18,145	288,680	139,167	717,203	13,448,923
28	Apr 02, 2020	23,874	379,821	178,222	975,983	17,694,965
29	Apr 03, 2020	31,412	499,736	228,238	1,328,136	23,281,550

day	date	deaths	cases	@ lo dbl	@ hi dbl	w/ 14d lag
1	Mar 06, 2020	14	230	224	238	10,726
8	Mar 13, 2020	98	1,571	1,266	2,057	73,216
15	Mar 20, 2020	674	10,726	7,151	17,784	499,736
22	Mar 27, 2020	4,602	73,216	40,402	153,688	3,410,959
29	Apr 03, 2020	31,412	499,736	228,238	1,328,136	23,281,550
36	Apr 10, 2020	214,403	3,410,959	1,289,346	11,477,413	158,908,518
43	Apr 17, 2020	1,463,411	23,281,550	7,283,681	99,184,812	1,084,632,112
50	Apr 24, 2020	9,988,535	158,908,518	41,146,424	857,129,291	7,403,170,243

Source Code

https://pastebin.com/raw/Sn2jrG5f

Please note any observed errors / corrections.

Earlier

• March 3: "COVID-19: A Laycat’s guide to what to consider / watch for" https://joindiaspora.com/posts/3abaa5e03fca0138fcb6002590d8e506
• February 28: "A Laycat’s COVID-19 / Coronavirus Updates – February 28, 2020" https://joindiaspora.com/posts/bc04cb503c840138f4b8002590d8e506
• ~ January 22: "The Wuhan 2019nCoV Coronavirus Epidemic is growing by a factor of ten a week" https://joindiaspora.com/posts/6eab7f50245f0138ba5f002590d8e506

#coronavirus #covid-19 #covid19 #ncov2019 #epidemiology #epidemics #exponentialGrowth #IHopeIAmWrong #awk

Would Anyone Care for a Game of Chess?

In a famous illustration of the power of exponential growth, the inventor of chess is said to have been asked by the King how he would like to be rewarded. His response sounded modest and straightforward: just pay me one grain of wheat on the first square of the chess board, and double that for each additional square. The King found this to be an absurdly humble respect, laughed, and ordered it so.

During fulfillment operations his treasury staff came to realize that the request would clear not only the granaries of the palace, nor of the entire kingdom, but of the entire world, many times over. If a single grain of wheat were 1/4 gram, the entire chessboard would have held over 4,611 billion tonnes. Worldwide total wheat production in 2007 was only 2.3 billion tons. It would take over 2,000 years to fulfill the inventor's request even now, from every wheat field on the planet.

Legend has it the King was not amused and the game's inventor left this world somewhat shortened.

But let's change a few elements...

Continued at the dreddit

#chess #exponentialgrowth #ebola

http://www.reddit.com/r/dredmorbius/comments/2ifbzi/would_anyone_care_for_a_game_of_chess/