I've played around with ChatGPT a lot and I worked with an early access version of GPT3 (the underlying model on which Chat GPT is based) for a year before that. Again, yelling into the wind I guess, but it autocompletes good answers because it is a statistical combination of associated text it is trained on and can interleave.

I'm not saying it has no relationship to intelligence or how human language works. We largely leverage autocompletion engines ourselves. Try not completing the phrase "Roses are __" with the answer everyone knows. However, that's essentially *all* that drives these bots. They are based on Transformer architectures whose only training objective for the base model is to predict the next token. How it learns data under the hood to accomplish that task is largely opaque. It is the epitome of the Chinese room (https://en.wikipedia.org/wiki/Chinese_room). Tell ChatGPT to write a paragraph proving that 1 == 2. It will spit out some bullshit trying. Or it will dodge the question ("I don't have enough information to answer that...") because the reinforcement learning outer loop has learned from human downvotes not to attempt an answer to questions that look like that.

What I'm saying is that the jump to sophisticated, consistent long term planning and introspection by an AI is a much bigger one than one might infer from reading ChatGPT interactions. It is not clear that there is any sort of "mind" in there, certainly not one that acts consistently and with intention.

I dearly hope (and suspect to be true) that this was the exact logic behind the name.

As an ML scientist, I feel like I'm yelling into the wind here, but it's incredibly important that the general public not extrapolate the abilities of ChatGPT (or these other language models) too far based on how seemingly fluid GPT3 prose is.

It is not smart. It is not auditable. It is text autocomplete on steroids. It is "smarter" in the sense that it will autocomplete something very cogent based on the corpus of training text. A little upvoting and downvoting in ChatGPT has reinforced it towards avoiding topics where it tends to generate the most bullshit, but make no mistake: it hasn't learned that it was wrong nor does it have an explicit abstraction to store that information. Do not ascribe words like "answer" or "plan" to ChatGPT, because again it has no abstractions for either.

It can certainly still be useful, but researchers have very little idea how to control the internals directly. They're mostly building models to modulate the output. As a home assistant, it will probably sound very pleasant, but it will be very hard to get it to do something as simple as consistently incorporate information about you from its database (i.e. things that you've purchased previously, and so on).

In Mexico, or in the US, would this ban be based solely on intent? Let's say I created a company to geo-engineer the planet to be *warmer* by releasing large amounts of methane. How am I different from an industrial company releasing the same amount of methane as a byproduct of its industrial processes? Would they technically fall under a "geoengineering ban"?

Oh, come on... the standardly reported annual CPI metric from December 13th indicated something like 7.1%. Core inflation was 6%. We'll get a new number in a few days.

This is a messy econometrics problem and the above metrics aren't great, so I'm interested to hear out claims that it's lower, but "inflation was last reported to be at ~2%" is too vague to evaluated. Which number? When was it measured? What goods? Where are you hearing 2%?

Even if it was measured under 0.17% over a month's span (e.g could be annualized to 2%), the noise of month-to-month measurement wouldn't give me confidence yet that it reflected reality.

I tend to be pretty cynical about AGI timing claims in ML (I voted 50 years but I was closer to voting on the longer side)

However, one counterpoint: we create general intelligences every day as new babies are conceived, born, and learn. Do we know exactly how the mechanisms for learning work and how to explain that succinctly? Nope! But the ability to trigger the creation of AGI and the full understanding of the mechanisms of how it came to be may not necessarily go hand in hand.

I think you mean about a dollar and ten cents short.

The lore explicitly reset that limitation two Doctors ago. And it's just more honest that they synced the story limitation with the real limitation, which is "as many regenerations as will bring in eyeballs".

Glad you like those things. In person work sounds good for you.

Cower? Really? Some people like getting at least an hour and a half of life back from commuting, and get very little out of the experience itself. Given the inherent decreased cost, I'm doubtful remote work opportunities are going to go away barring a complete breakdown of the Internet. It's cool that you enjoy the traditional commute; those jobs are not going away either.

Given the overall data, it really isn't unreliable unless you think that the CDC is inventing dead people out of whole cloth.

https://www.cdc.gov/nchs/nvss/...

If you scroll down on that page, it shows a plot of total deaths per week over the last few years. Again, total deaths, regardless of cause. The line indicates a level that would be counted as abnormally high - the average is lower. The toll of the bad flu season in 2018 is clear, and so is the impact of COVID-19. The total deaths are largely consistent with the attributed toll.

I can't stress enough that this data isn't vulnerable to errors in the "cause of death" attribution.

...which is why the summary included it? It's had about a 1/3 the death rate.

Google "CDC excess deaths" and I'm sure you'll find the source. I'm citing the US CDC, so that's going to be about the best aggregation one can get.

I don't really care about whatever media reports were doing to chide people and don't particularly agree. I know that lockdowns in my state (I am in the US, I should be clear about) drastically reduced contact between most people I know. A segment of people ignored it, sure, but particularly early on, people took it seriously. Actual data beyond anecdotal experience (e.g. actual miles traveled from cell phone data, etc.) tends to show a drastic drop in people going to large events. Further, large events and conventions were largely and indisputedly canceled. Losing every large crowded mass event is not minor attenuation. If your point is that the media is alarmist, though, sure.

There are a lot of influences. One, the uptick in deaths happened prior to lockdowns really starting. There are a huge number of influences here. Car deaths went down due to miles traveled. I'm sure some deaths went up from a hesitance to seek care. Suicides, for instance, surprisingly went down, against expectation. However, the volume change is very large relative to any of the other death sources you mention.

If we can't talk about these things in the abstract, then we can't quantify.

Yes, I will be interested to see if there's a dip in the coming years. I was honestly curious to see if there was a dip over last fall, indicating exactly what you said. This will be a difficult calculation to do, because per capita is likely still not subtle enough, and it would have to be calculated based on age cohort. That didn't really seem to happen.

However, the longer it takes for that dip in each cohort to happen, the less one can "discount" it. Many elderly people losing the last five years of their lives is still a tragedy, both emotionally and from a family support perspective. I personally suspect people probably lost much more (decade(s)).

And the obesity percentage in the US is high, as is diabetes and other complicating risk factors. Testing per person was much lower at the beginning allowing easier spread. You can't throw out the negatives but keep the positives in formulating a projection. Everything you're saying was probably said (and will be said again) in wargame simulations of pandemics, but I find it very hard to square with the data.

Like many other things, I can't assert it didn't ever happen. It would just require fraud on a scale that just doesn't account for anywhere near the bulk of the volume. Again, if your assertion was "flu deaths got counted as COVID and inflated the death rate" just can't work because it would require a number of flu deaths that is phenomenally beyond the statistics of recent decades. A flu like that would have been noticed by the large community that studies (and is funded to study it!) and publicized as such.

Let's say you're right.

Following annual trends, flu deaths would have contributed in the tens of thousands. COVID has resulted in hundred of thousands.

Let's go one farther - I frequently hear the argument that COVID deaths are overreported on death certificates, for instance. However, the total count of *all* deaths, ignoring the reported causes, show a clear large increase over the past year, even before the lockdowns. The total is roughly 500,000 above what has been a fairly stable baseline.
https://www.cdc.gov/nchs/nvss/...

This was debatable (in good faith) a year ago when we had much less data. It's not really debatable now.

I don't think you're correct, regardless. Flu transmits via similar vectors as COVID, so assuming that flu numbers would maintain when the vector for transmitting it is attenuated doesn't seem like a particularly rational hypothesis.

One thing that's easy to miss about these big franchise series - these series are closer to traditional miniseries in terms of viewer time commitment. Let me compare to the CW's Arrowverse and related shows, which absolutely caused fatigue. It ballooned to 5-6 series, each with 20-23 episodes (!) per season each at 45 minutes. I liked the first couple well enough, but it was ultimately enough of a chore that mild interest didn't overcome.

The Mandalorian is certainly costlier to produce than those shows, but a season of The Mandalorian runs at 8 episodes with comfortable runtimes that vary between 35-55 minutes. The Book of Boba Fett is likely to be similar, as are the rest of these. I can watch three of them for the same time investment as a season of The Flash. Plus, the quality is frankly likely to be better partly because (among many reasons) they don't stretch a solid arc out over too much time.

Plus, I'm dead certain most of these series (other than tentpoles like the Mandalorian) will be plotted with reasonably conclusive endings to their season, and they'll play it by ear whether to continue them.

You are correct, but please expand on your point. Why does that mean as a country we shouldn't try to counter it or at least expose when it's happening? Analogously, just because thieves have trying to steal private property for ages doesn't mean we shouldn't make people aware of new lines of attack so they can secure themselves.

Public knowledge moves slowly, and I don't think the public generally appreciates, even if it's obvious given some thought, how much astroturfing can take place on new forms of communication.