- Run
@IAPSai
to secure benefits and manage risks of AI
- top forecaster
Understand+mitigate AI risk โ safety+trust โ AI adoption+innovation โ good world ๐
This is the most amazing graph. On this scale, only the industrial revolution has mattered. The rise and fall of Rome, several different times losing ~10% of population, major shifts in the social order and we just keep humming along in what looks like a flat line until BAM
My wife, Samantha, passed away this afternoon. Her life wasn't always the best, but she filled it with such love and kindness to everyone. She touched so many people and I'm really happy to have been a part of that. She was my best friend and I will miss her very much.
I love how much Texas basically hates their own solar and wind industries but they can't stop building simply because they actually manage to just not be insufferably bureaucratic
Other states could do this
Hollow gospel of "effective altruism" - give away millions of dollars to restore sight to blind people
True altruism, not hollow at all - Do nothing to help other people, complain endlessly on the internet about other's efforts, never offer any suggestions for how to do better
People don't like it when we assign kinda made-up numbers like 8% to the risk of nuclear war within the next year, but honestly what the heck are you supposed to do with nonsense phrases like "unlikely but not altogether implausible"?
@Willyintheworld
A larger component here IMO is a strong default away from admitting that there is no reasonable answer, instead hallucinating something.
It consistently gets questions correct when there is a correct answer.
Look, I really didn't want to write about this, but here we are.
I'm very upset with Nick Bostrom.
His original email was terrible, racist, and offensive.
His apology was absolutely idiotically executed.
Here I explain why I feel this way.
Do any financial advisors here on this app know what percentage of your net worth it is advisable to invest in the fact that Biden is definitely going to be the Democratic nominee?
PredictIt is somehow selling this outcome for 83c on the dollar.
Still crazy to me that one of the earliest concrete public warnings about COVID came in late January from an anonymous forecaster on a forecasting platform.
On January 24th, the number one predictor on Metaculus, an anonymous user who goes by the handle traviswfisher, revised down his prediction on the question โwill the world population increase every year for the next decade?โ. (1/4)
Effective altruism is three radical ideas I want to protect
Until someone creates a social movement that better embodies radical empathy, scope sensitivity, and the scout mindset, I'm going to continue to value effective altruism highly.
To give an idea of just how much SOTA exceeded forecasters' expectations, here are the prediction intervals for the MATH and Massive Multitask benchmarks. Both outcomes exceeded the 95th percentile prediction.
Being poor in the United States is not fun and should be fixed, but it's important to still recognize that poverty in Ethiopia and other low income nations is much much worse.
More people need to hear good news on climate:
* Previously 4ยบC warming was the median path. Now, thanks to hard work on policy/advocacy, the chance of that is ~5%
* Median now 2.7ยบC
* Chance of 6ยบC warming from ~10% to <1%
Keep up the momentum!
From:
Regulators will block scientific and economic progress to save nine birds, and then go to KFC for dinner.
I'm begging for some sanity when it comes to how we protect animals and balance these trade-offs.
Whatโs more important:
1) Connecting the entire world via telecommunications satellites & becoming an interplanetary species
Or
2) Protecting 9 bird nests in a tiny corner of Texas
Gato๐a scalable generalist agent that uses a single transformer with exactly the same weights to play Atari, follow text instructions, caption images, chat with people, control a real robot arm, and more:
Paper: 1/
Effective altruists: Collectively we've raised $500M to address global health and are aiming to do $1B/yr by 2025. 23% of us are vegan and another 23% of us are vegetarian. Recently a lot of us are worried about AI and pandemics too.
Weaver: Why are EAs so greedy?
Guy goes to a doctor.
"Doctor, the world makes no sense."
Doctor looks at him and says: "Treatment is simple! Forecaster convention is in town. Go talk to the top forecasters. That will make things make sense."
The man bursts into tears: "But doctor, I am a top forecaster."
Pretty crazy this isn't more widely known:
In 1961, a nuclear bomb was accidentally dropped by the US on the US. Three of the four failsafes failed. Only a single failsafe averted the deaths of millions of people.
Good point - Sequoia and others invested hundreds of millions in FTX without uncovering the fraud or massive lapses in governance despite direct access to due diligence. Understandable that EA wasn't able to see this coming.
Your incredibly obvious reminder that MANY people besides FTX's charity recipients had HIGH personal stakes in guessing whether FTX was stealing customer deposits; and the supermajority of those people, who were much deeper into cryptostuff, did not publicly guess this right.
Bragging time: I'm good at forecasting (at least 98th percentile)!
After getting 20th place in the 2022 ACX forecasting competition, I got 12th place in the 2023 competition!
My scores were better than the entire
@metaculus
community (of which I was part)!
OpenAI has a had a year's worth of corporate scandal in four days --
Thu: Lead safety researcher leaves, posts thread saying the company isn't prioritizing safety
Fri: Equity controversy
Today: Scarlett Johansson controversy
Yikes.
I'm happy that I've gave away a third of my income this year.
I know not everyone can do this, but more people can do it than they think, and more people should really sincerely try... 33% or 10% or 5%, but try! There's so many problems we can solve!
Our 2023 giving recommendations are in! We estimate that donations to the programs we recommend can save a life for roughly $5,000 on average. Read our blog post to learn how you can make the most impact with your charitable donations this giving season.
Crazy to think how worried the world used to be (and sometimes still is) about overpopulation when it now seems like underpopulation looks much more concerning (relatively, and at least for now).
Elon is right
- the total fertility rate in the US has declined from 2.12 in 2007 to 1.64 in 2020
- half the worldโs population lives in countries with below-replacement-level fertility
- China now has a fertility rate even lower than Japanโs
Exclusive: Aspartame, one of the world's most common artificial sweeteners, is set to be declared a possible carcinogen next month by a leading global health body, pitting it against the food industry and regulators
Iโve been investigating the contamination theory of the obesity epidemic as laid out in
@mold_time
โs โA Chemical Hunger,โ and found that several of what seem to be the strongest arguments for the theory turned out to be dubious. ๐งต
Good Evening, "Robert F. Kennedy, Jr." was a one month political science field experiment conducted by Harvard University to test the impact of name recognition. We are now complete with our study. Thank you for your time.
Once I got over the shock of Lina Khan talking about p(doom), I realize it's kinda bonkers we're talking about a 15% chance of global catastrophe by AI as optimism. I don't want to accept that high of a risk!
damn this blew up. let me plug my soundcloud
...Ok, I don't have a soundcloud but I do have a job opportunity! If you're a researcher and want to get paid to save the world and not deal with academia BS, apply here!
Chickens dwarf pigs / sheep / goat / cattle, but...
...fish dwarf chicken
...insects are on par with fish
...and everything is dominated by shrimp
So actually animal welfare is basically shrimp welfare
TBH "complex but wrong" is also fairly common and underrated. Way more common than "complex but right".
And I think "simple but right" is underrated as well.
Gave GPT one of those (not-scientific) online IQ tests. It got 6 of the 10 questions correctly. The website says this "indicates an IQ score between 105 and 120". So I guess the AI is human-level now?
As a huge Taylor Swift fan I hesitate to say this, but I think she is getting too powerful. She's had multiple unexpected capability advances this year and is now even creating minor earthquakes. If we extrapolate this over the next decade, it points to us losing control.
Reminder that people who think there hasn't been technological progress in the past twenty years are under-rock-dwellers that have somehow never heard of LLMs, CRISPR, mRNA, semaglutide, the revolution in Solar energy, or 5G.
Not to mention your laptop is ~32x faster now
Supporting more Russian-to-US immigration does sound like one of the best things we could be doing for both the US and for hindering Russia. Why don't we do much strategic immigration initiatives?
Imagine you're about to get on an airplane and 48% of the engineers that built the airplane say there's a >=10% chance that their plane might crash and kill everyone.
Would you get on that plane?
That's where we are at with AI right now.
Find it hard to stay up to date with the large volume of content on the EA Forum and Lesswrong?
My assistant Zoe Williams is now reading the EA Forum and Lesswrong so you don't have to! Here's the first weekly summary
This is crazy.
Go to the IPCC 2015 estimate for solar. Use their most *optimistic* forecast, not their actual expectation. Use the value for *2030*, not 2022. We've already exceeded that.
We're so good at solar.
Just looking at the 'Shared Economic Pathways' used in the IPCC reports.
The most optimistic scenario โ SSP1 โ has global solar output at around 1,000 TWh by 2030.
In 2021, we *already* produced this amount from solar. A decade earlier.
Disagree, vision pro is the seventh best and the ranking is:
1.) LLMs
2.) CRISPR
3.) mRNA vaccines
4.) semaglutide
5.) Perovskite solar cell
6.) Faster internet
7.) Apple Vision Pro