All of Hauke Hillebrandt's Comments + Replies

Is OpenAI gaming user numbers?

Gdoc here https://docs.google.com/document/d/1os0WNmJ-O1eEGeKr543nkemnXbTmYkE2sC-t51c9OE4/edit?tab=t.0

Some have questioned OpenAI's recent weekly user numbers:[1]

Feb '23: 100M[2]

Sep '24: 200M[3] of which 11.5M paid, Enterprise: 1M[4]

Feb '25: 400M[5] of which 15M paid, 15.5M[6] / Enterprise: 2M

One can see:

• Surprisingly, increasingly faster user growth
• While OpenAI converted 11.5M out of the first 200M users, they only got 3.5M users out of the most recent 200M to pay for ChatGPT

Where did that growth come from? It's not from ... (read more)

This lag effect might amplify a lot more when big budget movies about SBF/FTX come out.

Yes, good catch, this is based on research from the World Value Survey - I've added a citation.

I checked. It's 0.67.

This seems to come from European countries.

Yeah I actually do cite that piece in the appendix 'GDP as a proxy for welfare' where I list more literature like this. So yeah, it's not a perfect measure but it's the one we have and 'all models are wrong but some are useful' and GDP is quite a powerful predictor of all kinds of outcomes: 

In a 2016 paper, Jones and Klenow used measures of consumption, leisure, inequality, and mortality, to create a consumption-equivalent welfare measure that allows comparisons across time for a given country, as well as across countries.^[6] 

This measure of huma... (read more)

You can compute where energy is cheap, then send the results (e.g. weights, inference) on where ever needed.

But Amazon just bought rented half a nuclear power plant (1GW) near Pennsylvania, so maybe it doesn't make sense now.

cf

“The Bootleggers and Baptists effect describes cases where an industry (e.g. bootleggers) agrees with prosocial actors like regulators (e.g. baptists) to regulate more (here ban alcohol during the prohibition) to maximize profits and deter entry. This seems to be happening in AI where the industry lobbies for stricter regulation. Yet, in the EU, OpenAI lobbied to water down EU AI regulation to not classify GPT as 'high risk' to exempt it from stringent legal requirements.^[1] In the US, the FTC recently said that Big Tech intimidates competition... (read more)

Hanson Strawmans the AI-Ruin Argument

I don't agree with Hanson generally, but I think there's something there that rationalist AI risk public outreach has overemphasized first principles thinking, theory, and logical possibilities (e.g. evolution, gradient decent, human-chimp analogy, ) over concrete more tangible empirical findings (e.g. deception emerging in small models, specification gaming, LLMs helping to create WMDs, etc.).

AI labs should escalate the frequency of tests for how capable their model is as they increase compute during training

Comments on the doc welcome.

Inspired by ideas from Lucius Bushnaq, David Manheim, Gavin Leech, but any errors are mine.

— 

AI experts almost unanimously agree that AGI labs should pause the development process if sufficiently dangerous capabilities are detected. Compute, algorithms, and data, form the AI triad—the main inputs to produce better AI. AI models work by using compute to run algorithms that learn from data. AI progresses due t... (read more)

ARC's GPT-4 evaluation is cited in the FT article, in case that was ambiguous.

Agreed, the initial announcement read like AI safety washing and more political action is needed, hence the call to action to improve this.

But read the taskforce leader’s op-ed: 

1. He signed the pause AI petition.
2. He cites ARC’s GPT-4 evaluation and Lesswrong in his AI report which has a large section on safety.
3. “[Anthropic] has invested substantially in alignment, with 42 per cent of its team working on that area in 2021. But ultimately it is locked in the same race. For that reason, I would support significant regulation by governments and a pr

Ian Hogarth is leading the task force who's on record saying that AGI could lead to “obsolescence or destruction of the human race” if there’s no regulation on the technology’s progress. 

Matt Clifford is also advising the task force - on record having said the same thing and knows a lot about AI safety. He had Jess Whittlestone & Jack Clark on his podcast. 

If mainstream AI safety is useful and doesn't increase capabilities, then the taskforce and the $125M seem valuable.

If it improves capabilities, then it's a drop in the bucket in terms of o... (read more)

a large part of those 'leaks' are fake

Can you give concrete examples?

[Years of life lost due to C19]

A recent meta-analysis looks at C-19-related mortality by age groups in Europe and finds the following age distribution:

< 40: 0.1%

40-69: 12.8%

≥ 70: 84.8%

In this spreadsheet model I combine this data with Metaculus predictions to get at the years of life lost (YLLs) due to C19.

I find C19 might cause 6m - 87m YYLs (highly dependending on # of deaths). For comparison, substance abuse causes 13m, diarrhea causes 85m YLLs.

Countries often spend 1-3x GDP per capita to avert a DALY, and so the world might want to spend $2-... (read more)

Very good analysis.

I also thought your recent blog was excellent and think you should make it a top level post:

https://entersingularity.wordpress.com/2020/03/23/covid-19-vs-influenza/

Cheers - have taken this point out.

Cruise Ship passenger are a non random sample with perhaps higher co-morbidities.

The cruise ships analysed are non-random sample: "at least 25 other cruise ships have confirmed COVID-19 cases"

Being on a cruise ship might increase your risk because of dose response https://twitter.com/robinhanson/status/1242655704663691264

Onboard IFR. as 1.2% (0.38-2.7%) https://www.medrxiv.org/content/10.1101/2020.03.05.20031773v2

Ioannidis: “A whole country is not a ship.”

Thanks Pablo for your comment and helping to clarify this point. I'm sorry if I was being unclear.

I understand what you're saying. However:

• I realize that the Oxford study did not collect any new empirical data that in itself should cause us to update our views.
• The authors make the assumption that the IFR is low and the virus is widespread and find that it fits the present data just as well as high IFR and low spread. But it does not mean that the model is merely theoretical: the authors do fit the data on the current epidemic.
• This is not differ

Cheers- corrected.

It looks more like you listed all the evidence you could find for the theory and didn't do anything else.

That was precisely my ambition here - as highlighted in the title ("The case for c19 being widespread"). I did not claim that this was an even-handed take. I wanted to consider the evidence for a theory that only very few smart people believe. I think such an exercise can often be useful.

I don't think this is actually how selection effects work.

The professor acknowledges that there are problems with self-selection, but given that the... (read more)

I do not think that can be used as decisive evidence to falsify wide-spread.

This is a non-random village in Italy, so of course, some villages in Italy will show very high mortality just by chance.

That region of Italy has high smoking rates, very bad air pollution, and the highest age structure outside of Japan.

By the end of its odyssey, a total of 712 of them tested positive, about a fifth.

Perhaps other on the ship had already cleared the virus and were asymptomatic. PCR only works for a week. Also there might have been false negatives. I disagree that the age and comorbidity structure can only lead to skewed results by a factor of two or three, because this assumes that there are few asymptomatic infections (I'm arguing here that the age tables are wrong).

In my post, I've argued why the data out of China might be wrong.

Iceland's data might be wrong because it is based on PCR not serology, which means that many people might have already cleared the infection, and it is also not random.

That's true and that's what they were criticized for.

They argued that the current data we observe can be also be explained by low IFR and widespread infection. They called for widespread serological testing to see which hypothesis is correct.

If in the next few weeks we see high percentage of people with antibodies then it's true.

In the meantime, I thought it might be interesting to see what other evidence there is for infection being widespread, which would suggest that IFR is low.

No. My ambition here was a bit simpler. I have presented a rough qualitative argument here that infection is already widespread and only a toy model. There are some issues with this and I haven't done formal modelling. For instance, this would be what would be called the "crude IFR" I think , but the time lag adjusted IFR (~30 days from infection to death) might increase the death toll.

Currently, also every death in Italy where coronavirus is detected is recorded as a C19 death.

FWIW, if UK death toll will surpass 10,000, then this wouldn't fit very well with this hypothesis here.

The point remains: given that some people have such a different theory, it's unclear how many supporting pieces of evidence your should expect to see, and it's important to compare the evidence against the theory to the evidence for it.

Yes, that's what I'm trying to do here. I feel this is a neglected take and on the margin more people should think about whether this theory is true, given the stakes.

Presumably some of these people are hypochondriacs or have the flu? Also, I bet people with symptoms are more likely to use the app.

With al

I'm not impressed by the comment about this paper here on LW or the twitter link in it.

This paper was written by an international team of highly cited disease modellers who know about the Diamond Princess and have put their reputation on the line to make the case that this the hypothesis of high infections rate and low infection fatality might be true.

I think it is a realistic range that this many people are already infected and are asymptomatic. Above I've tried to summarize and review the relevant evidence that fits with this hypothesis.

But I'm not ruling out the more common theory (that we have maybe only 10x the 500k confirmed cases). I just find it less likely.

There were a few dengue in Australia and Florida where it is unusual

Dengue "popping up in unusual places", makes me think that it's more likely that massive Dengue outbreaks in Latin America might have a high proportion of C19.

One person had persistent negative swab, but tested positive through fecal samples...

“Chinese journalists have uncovered other cases of people testing negative six times before a seventh test confirmed they had the disease.”

This is just to lend credence to the paper that shows there had been 2 million inf... (read more)

This seems pretty hard to evaluate because with a large number of published pre-prints on the outbreak, it's not very surprising that there would be many suggesting higher-than-expected spread.

No, this is different. I'm not just cherry picking the tail-end of a normal distribution of IFRs etc. The Gupta study in particular and some of the other studies suggest a fundamentally different theory of the pandemic.

Presumably some of these people are hypochondriacs or have the flu? Also, I bet people with symptoms are more likely to use the app.

Yes, but... (read more)

If the Gupta study is true, then a rough approximation (ignoring lag) would be that it's:

IFR = Number of UK deaths (~750) / 36-68% of the UK population (66 million).

So 0.002% to 0.003%.

In Italy, with almost 10k deaths it would be 0.02%-0.04%

from supplementary materials:

"DISCLAIMER: The following estimates were computed using 2010 US Census data with 2016 population projections and the percentages of clinical cases and mortality events reported in Mainland China by the Chinese Center for Disease Control as of February 11th, 2020. CCDC Weekly / Vol. 2 / No. 8, page 115, Table 1. The following estimates represent a worst-case scenario, which is unlikely to materialize. • Maximum number of symptomatic cases = 34,653,921 • Maximum number of mild cases = 28,035,022 • Maxim... (read more)

And yet another preprint estimating the R0 to be 26.5:

Quotes from paper:

"The size of the COVID-19 reproduction number documented in the literature is relatively small. Our estimates indicate that R0= 26.5, in the case that the asymptomatic sub-population is accounted for. In this scenario, the peek of symptomatic infections is reached in 36 days with approximately 9.5% of the entire population showing symptoms, as shown in Figure 3."

I think they estimate about 1 million severe cases in the US alone if left unchecked at the peak.

"It is unlike... (read more)

And another preprint saying there were +700k cases in China on 13th of March:

"Since severe cases, which more likely lead to fatal outcomes, are detected at a higher percentage than mild cases, the reported death rates are likely inflated in most countries. Such under-estimation can be attributed to under-sampling of infection cases and results in systematic death rate estimation biases. The method proposed here utilizes a benchmark country (South Korea) and its reported death rates in combination with population demographics to correct the reported CO... (read more)

As mentioned in a comment above, one of the (pretty highly credentialed) authors of this preprint has written two papers on the Diamond Princess, and so, excuse the appeal to authority, but any argument against this paper based on Diamond Princess doesn't seem likely to invalidate conclusions of this preprint .

Also this squares seemingly squares more with John Ioannidis take on Corona:

"no countries have reliable data on the prevalence of the virus in a representative random sample of the general population."

And that airborn-ish transmission ... (read more)

Not sure: the Diamond Princess is mentioned in this preprint and in fact one of the authors of this preprint wrote two papers on the Diamond Princess:

https://scholar.google.com/citations?hl=en&user=OW5PDVgAAAAJ&view_op=list_works&sortby=pubdate

So I think they thought about this,

The first paper that I cite has a very illustrative video and is a seminal paper in this field.

Table 8 in the review paper that you refer to shows a trend of estimation techniques getting better over time. In the latest study from 5 years ago the mean error was down to 6.47.

My broader point is:

• the error rate might be brought down even further by better methods, video quality, and priors
• this might so that it a valid proxy for fever
• This might be very cost-effective on a population level, given the zero marginal cost of software

However, I do agree that this is not trivial.

That's false. The accuracy isn't high. I learned from the last conversation I had with EA who had a startup that did this, that the accuracy isn't high enough to be useful medically.

Interesting data point - there are several papers on this that say it's a reliable way to measure heart rate (less than 10bpm; see "Heart rate estimation using facial video"). Perhaps this could be brought down much further by throwing more engineering brains, computation and priors at it.

Where do those ≥38°C come from? From what I read