My last AGI Friday has a collection of notable new slopcore music videos, plus a $100 stump-the-AI challenge: agifriday.substack.com/p/mem…
I wrote about the chasm between here and AGI: agifriday.substack.com/p/cha… (conclusion: confidence interval still very wide, like 3-30 years or wider)
First-world problem: dozens of people are suddenly signing up for AGI Friday, seemingly largely from Twitter, and I can't figure out how to find the tweet that's doing it 😭
Oh yeah, I started a weekly newsletter about AGI timelines called AGI Friday. agifriday.substack.com/
PS, found it! nitter.gq/nic__carter/status/197…
as a sidenote, one reason I really like Leo is because he maintains a kind of "pragmatic cautious optimism" around AI. he thinks it's a civilizationally important project and a matter of urgent national security. he thinks AGI and maybe ASI is possible, but he isn't a doomer. in fact, he's helping bring this future about (more on this later).
most "AI futurists" in silicon valley are doomers, unfortunately. they unhelpfully think AI is going to literally kill us all. I'm not going to try to "debunk" the doomer mentality here but I think it's far fetched and more of a spiritual/metaphysical position than an empirical one. (secular doomsday cults are the main thrust of religion-without-religion these days, whether climate or now AI.) the doomers also don't really have good suggestions for dealing with the apocalypse they foresee. they just want us to collectively stop building AI, or something. this isn't a very practical suggestion but they are very good at making noise, so they get disproportionate attention. there's a very big market for prophecies of doom and these people take advantage of that.
the second camp is the skeptics mainly found on the left that intone things like "AI is just a stochastic parrot" camp which we can trivially dismiss since they are just obviously wrong. these people tend to think all tech and markets are bad and wrong and think AI is a bubble. not much to say here, they are just wrong.
the other major camp is the e/accs that are kind of the inverse of the doomers; they want to push forward AI no matter what as a kind of political statement. they tend to be a little cavalier with regards to genuine AI safety risks and seem more motivated by owning the libs.
the last camp is simply the people building AI that have a strong economic incentive to downplay the risks. Leo isn't in this camp either. he's sanguine about the risks and the fact that this technology deserves a sort of manhattan project of its own. our ability to deal with it IS existential and he acknowledges that.
daniel reeves describes Leo's philosophy as "AGI realism" and it's the same camp that I subscribe to (even after reading every piece of doomer literature ever written).
The word "the" in phrases like "the faster we go, the sooner we'll get there" is not the word "the" as we know it. It's a homophone! It used to be spelled "tha" or "þa" and is basically used for if-then statements. "Tha more tha merrier" is "more ⟹ merrier" or "more ∝ merrier".
It’s been 2 years. I’m still ridiculously excited about self-driving cars, just that @elonmusk’s timeline here was fairly fantastical.
Now 4 years and a baby step taken (summoning from across a parking lot). I'm still rooting for this and don't want to sound all snarky about it or anything.
Daniel Reeves retweeted
Last week we fixed an issue that was causing our feedback pop up to pop up too often! 😅 On the brightside we did get some feedback from a customer that our "user support is stunning and second only to beeminder"! #buildinpublic
Daniel Reeves retweeted
New paper with R. Freeman, @pennockd, @dreev, B. Waggoner leo.ugr.es/isipta21/pmlr/fre… "Towards a Theory of Confidence in Market-Based Predictions": can you create a margin-of-error for a probability? How does that relate to volatility? Paper asks as many questions as it solves!
Daniel Reeves retweeted
one little fact about the world that makes me sad about once a month is when people who are really doing fine and could raise a great family but are much too aware of all their sundry emotional issues tell me they never plan to have kids
Daniel Reeves retweeted
We're looking for someone to help out with our user support. Let us know if you're interested or know someone who might be good for the role! help.beeminder.com/article/1…
What's the principle called where penalties have to be multiplied by the reciprocal of the probability of getting caught, otherwise technically-rational agents just take the calculated risk of committing the crime?
I feel like coding is like: You keep adding duct tape & chewing gum to satisfy different requirements & bugs that come up, then you gradually whittle it back down and end up with a few simple lines and it's bewildering how you were too dumb to've just written that from the start
I have a new appreciation for how the Axiom of Choice is kind of a wild leap of faith that leads to preposterosities like Banach-Tarski: doc.dreev.es/axiomofchoice
