mannykannot

Prior to the Hiroshima bombing, all that the US had told the USSR about the atom bomb was that they had developed a "new weapon of unusual destructive force" (Truman to Stalin at the Potsdam conference.) As it happens, Stalin knew all about it through espionage, but as the US government was unaware of this, we cannot deduce, from what Stalin had been told by the US government, that the US government's position was that using atom bombs against Japan would have no influence on Stalin's thinking.

Furthermore, we have evidence, from the diaries and correspondence of Truman, Stimson and others, that they were concerned about Stalin's territorial intentions, and expressed the hope that the US's demonstration of the atom bomb would curb them. Whether that led to at least a tacit secondary goal, of checking the USSR by leaving no doubt about the US's willingness to use atomic weapons in circumstances it deemed necessary, is an open question debated by historians to this day.

I'm generally OK with the list of push-backs against common misconceptions in the summary, but I have my doubts about the second one:

Claim: Hallucinations are inevitable. Finding: They are not, because language models can abstain when uncertain.

...which raises the question of how reliable the uncertainty estimate could get (we are not looking for perfection here: humans, to varying degrees, have the same problem.)

For a specific context, consider those cases where LLMs are programming and invent a non-existent function: are they usually less certain about that function than they are about the real functions they use? And even if so, abandoning the task with the equivalent of "I don't know [how to complete this task]" is not very useful, compared to what a competent human programmer would do: check whether such a function exists, and if not, decide whether to implement it themselves, or backtrack to the point where they can solve the problem without it.

More generally, I would guess that balancing the competing incentives to emit a definite statement or decline to do so could be difficult, especially if the balance is sensitive to the context.

There is a way to state Parson's point which avoids this issue: hallucinations are just as much a consequence of the LLM working as designed as are correct statements.

Well, it continued to bother Newton. Here, he seems to be foreshadowing the concept of fields:

"It is inconceivable that inanimate Matter should, without the Mediation of something else, which is not material, operate upon, and affect other matter without mutual Contact... That Gravity should be innate, inherent and essential to Matter, so that one body may act upon another at a distance thro’ a Vacuum, without the Mediation of any thing else, by and through which their Action and Force may be conveyed from one to another, is to me so great an Absurdity that I believe no Man who has in philosophical Matters a competent Faculty of thinking can ever fall into it. Gravity must be caused by an Agent acting constantly according to certain laws; but whether this Agent be material or immaterial, I have left to the Consideration of my readers."

https://plato.stanford.edu/archives/win2008/entries/qm-actio...

Your first point is correct, but it (like the second one) has no bearing on the issue of the medical soundness of the department's positions and policy.

I'm surprised the guardrails are aluminum rather than galvanized steel.

I see the ambiguous Vela incident of 22 September 1979 has been excluded.

https://en.wikipedia.org/wiki/Vela_incident

Up to a point I agree, but when someone deploys this vague language in what are presented as strong arguments for big claims, it is they who bear the burden of disambiguating, clarifying and justifying the terms they use.

That's true, but still, a single individual or small group living isolated in the real world will, over a lifetime, learn only a tiny fraction of what we can learn from the written knowledge accumulated over millennia.

Having said that, I tend to agree that having AI interact with the world may be key: for one thing, I'm not sure whether there is any sense in which LLMs understand that most of the information content of language is about an external world.

As the claim you made about copyright being used to take down a video was completely false, how did it contribute to anything?

> ...the courts could interpret however they want in conflict of your entire comment.

Courts sometimes adopt tendentious, cotroversial or disputed definitions in their rulings, and law-makers sometimes (and more frequently) do so when writing the law, but that in no way invalidates what I have written, which is about how to use a dictionary.

This is a myth: in Germany, as in many other countries, copyright covers only specific expression; you cannot copyright either the name of a historical person or a topic of discourse. The videos were briefly taken down as an automatic response to a complaint, but it seems the complaint was not upheld and the videos were restored.

At the time, Germany had a law censoring insulting comments about foreign heads of state, but that only applied to living ones (and maybe only those in office at the time?) That law was repealed in 2018.

The videos remained blocked in Turkey, but on account of a specific law banning criticism of Ataturk, not copyright.

I have heard it also puts you in a better position when an airline is bumping passengers or carry-on luggage, though the soft rules are always changing, so I don't know if this is still the case (or ever was.)

This is not how dictionaries work. When multiple definitions are given, it does not follow - and in many cases, it is not even possible - that they are all applicable in any context.

Ignoring context leads to things like "English as She is Spoke" https://en.wikipedia.org/wiki/English_as_She_Is_Spoke

Decent dictionaries give some guidance as to the contexts in which each each definition is applicable, but for thoroughness you probably cannot beat the unabridged Oxford English Dictionary.

The relevant definition here is neither legal nor technical, but from common usage, where recognizing a face, if not qualified, is taken to mean recognizing an individual by their face, not recognizing that you are seeing a face.

A little later in the class, you'll learn that even completely false hypotheses can be logically consistent.

There is one question here, and you have failed to address it.

As far as I know, compatibilism takes no position on whether the universe is deterministic. It is, rather, an antithesis to the thesis that determinism is logically incompatible with free will (or, from a more deflationary perspective, it offers an explanation for how we could feel we have 'classical' free will even in a deterministic universe.)

You got me interested in the question, and one of the first things I found out is that there is Zipf’s Law, and then there is Zipf’s Law of Abbreviation.

Monkeys-randomly-typing is one of many processes which will indeed generate sequences conforming to the former, and it is perhaps the exceptions which are most interesting.

The latter law observes that the former generally applies to sequences generated for communication, having semantics and usually a grammar. While this may be the expected finding, there is value in having this expectation empirically verified.

It is not primarily a matter of difficulty. The problem with choices like these is that they seem to be predicated on a rather simplistic, formulaic view of software development - a view that experienced developers will recognize as flawed.

In this case, one tacit assumption is that a given developer will typically adopt just one of these approaches. Another is that they can meaningfully and objectively be ordered along either of the axes purportedly being measured.

The analysis you quote smells of vapid pop psychology. What empirical evidence exists for supposing that the result is reliably objective, let alone relevant?

I agree, but would add that it is just one of its problems. Even before we get to the questions, we are told that one of the outcomes will be a score on a “friendliness” axis. What would it mean to be computer-friendly, and why would it matter?

But here, the context is relevant, and the analogy does not change this.

I take it that you are referring to the movie Memento? I had not heard of it, but I'll put it on my watch list.

I take your point about the non-necessity of dynamic learning for AGI.

Two months earlier there had been the so-called Skandalkonzert in Vienna, conducted by Arnold Schoenberg: "...it was during Berg's songs that the fighting began. At the trial, Straus commented that the thud of Buschbeck's punch had been the most harmonious sound at the entire concert. For Berg's work the Skandalkonzert had lasting consequences: the songs were not performed again until 1952, and the full score did not appear in print until 1966."

https://en.wikipedia.org/wiki/Skandalkonzert

I too can hear distant trains at night, especially if it is a still, clear night creating a low-level inversion to channel the sound.

There are several places in Britain (and elsewhere, I imagine) where beam engines have been preserved and are periodically run using live steam. the engines themselves are quiet by modern standards, though I believe the machinery they drove often produced a racket.

The next generation? Our generation is already there, apparently.

You cannot divine the motives of a person with self-delusional tendencies by assuming that their behavior was rational.

You are also merely technically right. It would require an intentional suspension of one's theory of mind to not recognize the extent to how Tesla's own marketing of its products, and its determination to hide pertinent information in cases like this, is intended to perpetuate the popular misconception of its capabilities.

> ...we have to live with the [the world] we got.

There was nothing inevitable in how we reached this situation, and no reason to let it continue.

Agreed, but this puts numbers on it.

It is exciting to see this course addressing the biology space and the chemistry space, but the final frontier is the space space.

The airport, being, to no small extent, a shopping mall, has incentives to get you there early.

One thing that would tend to discourage frequent travelers from determining their optimal pre-flight margin would be to have somewhat arbitrary and frequently-changing security procedures and requirements. A second way to encourage long loitering times would be to under-staff vital functions such as Air Traffic Control.

EWR appears to have nailed these strategies.