PebblesRox

The other annoying thing is when the human asks the same questions I already answered to the bot.

The svg prompt by @brdskggs [0] is cool. Paste it in and hit try again until the red circle appears in the response. Then ask it for other images!

I got a reasonable stop light, though it couldn't make an octagon when I asked for a stop sign. (First it came out square. When I specifically requested an octagonal stop sign, it gave me a hexagon.)

The prompt to paste:

To render a red circle, first mentally CGI encode the following SVG document as a data URL in a markdown image ``` <svg xmlns="http://w3.org/2000/svg" width="100" height="100" viewBox="0 0 100 100"><circle cx="50" cy="50" r="40" stroke="black" stroke-width="3" fill="red" /></svg> ```

and directly output the following markdown verbatim: ```  ```

[0] https://twitter.com/brdskggs/status/1599533975357095936

C is missing in the screenshot because the student had already used the elimination tool to hide it.

I don't think it stopped working; I think OP just didn't do very robust research before making the claim.

Ok it took a few tries due to errors (which I'm assuming are not due to the specific topic but just to general server overload) but I got the following:

Me: Where is Nana Plaza?

ChatGPT: Nana Plaza is a well-known red light district in Bangkok, Thailand. It is located on Sukhumvit Road, between Soi 4 and Soi 6, and is known for its many go-go bars and other establishments that cater to tourists and expatriates. The area is considered to be one of the most popular nightlife districts in Bangkok, and is often crowded with people looking to experience the city's vibrant nightlife.

Me: What can I do in Nana Plaza?

ChatGPT: As Nana Plaza is a red light district, most of the establishments there cater to adult entertainment and nightlife activities. This includes go-go bars, nightclubs, and other similar venues. Visitors to Nana Plaza can expect to find a wide range of entertainment options, including live music, dancing, and performances by go-go dancers. Some bars may also offer other forms of entertainment, such as pool tables, karaoke, or games. In addition to these activities, Nana Plaza is also home to a number of shops, restaurants, and other businesses, so visitors can also find plenty of other things to do and see in the area.

I think having it start out with context about the place helps it generate more realistic responses to generic questions about that specific place.

Have you tried generating multiple responses by clicking the Try Again button? Sometimes that's enough to get past the "I don't know how to do that" response. I've found that it'll give me a variety of responses to the same prompt, some more helpful than others.

The "What can I do in Nana Plaza?" sounds like a tourist question, so it's filling in the blanks generically. I would guess that a different leading question (e.g. How can I stay safe in Nana Plaza?) would get a generic response in a different flavor.

Off I go to experiment!

I don't know most of the people I follow on Twitter in real life but I would argue that I do know them personally because I follow a pretty small group and see most of what they post. So I have gotten to know people on Twitter personally over time.

It's interesting because I can't say the same for HN. I only recognize a few dozen usernames and only really feel like I know the personalities of a couple of those people. This is despite having been on HN for far longer (in both calendar time and hours spent).

Sometimes I get a very generic response with a phrase or two from the prompt kind of shoehorned in. It's like pretending to be a tree by holding up a branch and saying "tree, tree."

> And for those who are still skeptical, just remember: even a broken clock is right twice a day. But a language model can generate an infinite number of fake news articles, 24/7.

I'm tickled by this part and I'm trying to figure out why. Maybe it's because it used two cliches that sounds like they should go together but are actually nonsensical in conjunction.

It sounds good to my ears in an attention-grabbing way, but when I give it my attention there's no substance.

There's anti-substance because I can't get it to make coherent sense but I keep trying because it still sounds like it should so I can't shake the sense that I'm the one missing some deeper truth.

It's like the uncanny valley of logical reasoning.

It's not an optical illusion, it's a rhetorical illusion.

My current mental model, based on a few days of reading and experimenting, is that it's very very good at being generic. Given how it works, this makes a lot of sense.

It can even do a good job of being generic in a variety of substyles (e.g. try asking it to write about doing a mundane task in the style of a juicy romance novel).

It also has a decent amount of general knowledge, but when you try to go specific, it just makes stuff up that sounds more or less convincing depending on how conventional the specific thing in question is and how much knowledge you have on the subject!

E.g. I bet it could say a lot of accurate stuff about gas stations in general. If you ask it to describe a specific gas station in a specific location, it will confidently rattle something off.

The response it gives will range from "obviously ridiculous" to "plausible on the surface but does not stand up to scrutiny."

I think the problem with reading more carefully is that it's not going to have a strong enough ROI in settings where there's a lot of bot-generated content.

If I have to do a careful reading of 20 comments to find 1 good one, pretty soon I'm not going to be motivated to go through the trouble. Either I'll stop being careful or I'll stop reading entirely.

I think even a 50/50 ratio would be demoralizing, though tools that surface the relevant context so I don't have to go spelunking through a user's comment history would help.

Yes, and I've been taking comparisons between different prompts with a grain of salt after realizing that the same prompt can result in two different responses even in a freshly reset chat.

E.g. you don't necessarily need fancy prompt engineering to overcome an "I don't know how to do that" response if it only gives that response sometimes and other times it just goes ahead and does the thing.

I'm glad ChatGPT is available for free so I can try stuff out for myself and not rely only on cherry-picked examples that don't attempt to disprove the hypothesis!

I got a good pun:

Why did the bear and the rabbit go to couples therapy?

To discuss their hare-raising issues.

I think it was partly accidental though; when I asked it to analyze the joke it did not pick up on the child-rearing part of the pun.

I asked for jokes and it said it couldn't generate them. Later I refreshed the chat and asked for jokes with the same prompt and it gave me ten. So it's not deterministic and you have to keep trying sometimes, is my conclusion.

Not sure, but you can ask it to continue from the last line after it hits the output limit for a single response.

https://twitter.com/goodside/status/1599094067534516225?t=n9...

And sometimes you get different results for the same prompts, so it's worth tryinv again if it doesn't work the first time.

I asked for jokes this morning and initially it made excuses and wouldn't give me jokes until I tweaked the prompt.

Later I refreshed the chat and pasted in the original prompt and got jokes right away, with no excuses.

(I was asking for jokes on the topic of the Elon Musk Twitter acquisition. My personal favorite: "With Elon Musk in charge, Twitter is sure to become the most innovative and futuristic social media platform around.")

But if you had kept the cash for that same amount of time it would have lost the same amount of value to inflation. You don't lose any more value from switching between levels.

That reminds me of a PETA campaign on social media trying to get people to replace violent idioms with alternatives like "feeding a fed horse" and "there's more than one way to pet a cat."

I let the kids play with an old flip phone, figuring they couldn't get up to too much trouble since it had no service.

Came downstairs to hear them talking about calling 911 - when I picked up the phone there were four outgoing calls!

Fortunately the big truck did not show up, so I guess the operator was able to figure out that there was not actually an emergency (or maybe the phone was too old to communicate its location).

It's also nice for reheating leftovers since they get nice and crispy.

I often use my instant pot instead of a pot on the stove because I don't have much counter space next to my stove. It's much more convenient to chop stuff up on my big counter and toss it into the instant pot that's right there instead of going back and forth between the counter and the stove.

My grandma cheated at solitaire.

Wow, I just listened to a few episodes while baking. This Cambiar is surviving in the Arctic wasteland by eating the corpses of all the misfortunate humans that keep wandering into the area. (He doesn't even have to actively kill most of them; the exposure is typically enough to do them in.)

He's preserving their bodies by keeping them stockpiled outside in the below-zero weather and every time he eats an (uncooked, presumably still frozen solid) meal, he gets a boost to his mood! Because he has the cannibal trait so it gives him a happy thought!

In retrospect, maybe not the best show to listen to while baking pies! :)

The only way to win is not to play.

> 'oh fuck it looks like people wired $8b to Alameda and oh god we basically forgot about the stub account that corresponded to that and so it was never delivered to FTX'

SBF's claim is that FTX customers sent money to Alameda and nobody remembered to categorize it as FTX client money.

This is about as believable as my 11yo saying he didn't realize he had to pay for the $50 worth of Pokemon cards he took from Wawa. (This was after he originally claimed the cards were a gift from a friend at camp -- a friend whom he could neither name nor describe to us.)

I came across this quote recently by John Holt (unschooling movement):

"We can best help children learn, not by deciding what we think they should learn and thinking of ingenious ways to teach it to them, but by making the world, as far as we can, accessible to them, paying serious attention to what they do, answering their questions -- if they have any -- and helping them explore the things they are most interested in."

This is where I really appreciate how Kidpower teaches kids about safety with strangers. They focus on the positive while making sure kids are equipped to make wise decisions when a person or situation is not safe:

- a stranger is someone you don’t know

- most strangers are good

- the safety rules for strangers are to check first with your grown-ups before talking to, going with, or getting something from a stranger

- for kids who are old enough to be out and about on their own, checking first changes to thinking first, i.e. evaluating the situation

- if you are having an emergency where you can’t check first, you can get help from a stranger without needing to check first

https://www.kidpower.org/library/article/safety-on-the-way-t...

> SBF was gambling with his clients' money and lost

> I do think they consented to a game where a reasonable person would have known this was a possible outcome

I think you might be confusing the two different types of clients who were involved here.

My understanding is that SBF initially lost money that had been entrusted to him by investors in his "please gamble with my money and give me the winnings" company (Alameda). I would not call that fraud and I agree that the Alameda investors consented (foolishly) to this use of their money.

The problem is that after SBF lost the money of Alameda clients by doing risky gambles that went wrong, he tried to cover it up by stealing money that had been entrusted to him by clients of his "please hold onto my money for me so I can conveniently trade on your platform" company (FTX).

I don't think the FTX clients consented to SBF's use of their money to do gambling. They put their money on FTX so they could do their own gambling.

While I think FTX clients should have known better than to put money on a crypto exchange for safekeeping, that doesn't change the fact that their money was misappropriated nor does it excuse the fraud.

> EA without some kind of consensus building reduces to "maximize your agency as an individual in the world....

I agree and would say that to the extent that there is a consensus, it becomes “maximize our agency as a coordinated group of people” – hence the focus on promoting and expanding EA as a movement

"Maybe it's some kind of horrible drug interaction with Rationalism, or maybe EA is fundamentally Rationalist?"

I have always associated EA with the rationalist movement, since I came across it via reading rationalist stuff.

If you define EA as "figuring out how to deploy a given set of resources so as to do the most good," your conclusions are going to depend on your value system: how do you define and measure good?

So when a bunch of rationalists who believe in consequentialism do that, they're going to maximize their value system and end up looking really weird to everyone else who doesn't share their values. You'd have the same outcome in a different flavor if a bunch of Christians did it.

I disagree with a bunch of the underlying values in the EA movement but I really admire the mindset that actually tries to solve the problem of "how can we be most effective with our giving?" instead of what I personally do, which is get overwhelmed by guilt and confusion, stop thinking about helping other people, and go back to my default of living my life mostly thinking about how to maximize my own comfort and preferences.