RubberSoul

Do you know if anyone has looked at what kinds of task lists are better handled by Ralph versus giving the LLM the list and asking it to implement everything in one session/prompt? In other words, how do you recognize when Ralph is the right approach?

I ask because I've tinkered with some Ralph loops and found I can achieve similar results by handing Claude (Opus 4.6, 1M context) a PLAN.md file with a long checklist of tasks and telling it to sequentially complete them.

I don't seem to have the problems that Ralph solves. But like any tool, it would be good to have a mental model for when to reach for Ralph in the future should a need arise.

> To make this enforcement cost effective, it needs to recover more than the $80 Billion budget increase annually, no?

Improved enforcement can deter tax avoidance in addition to directly recovering money.

This is, of course, a major philosophical debate. I will just say that there could be reasons this is the government's business.

One issue with modern food production for consumers is that it can be difficult to determine whether a particular ingredient is safe or healthy. Information has public good qualities, and having each consumer make an individual determination about whether each ingredient is safe/healthy is costly. The obvious way to justify government intervention here is that it's more efficient for us to pool resources and empower a third-party to produce the information we need and ban/label products that are likely unsafe. The decisions won't always be right, but the approach might be much preferred to the alternatives.

Plus, conditional on socializing a lot of medical costs, it might make sense to put restrictions on behaviors that affect individual health.

Great question. The answer might be not that soon, and the effects will be confusing. Search "Solow paradox," "productivity paradox," and "IT productivity paradox," and you'll encounter a long debate in economics and management research about the role of computers and ICT spending in productivity.

The goal is to estimate the proportion of accounts that are bots. Let that number equal p. The variance of the proportion is p(1-p). The highest that can be is (0.5)0.5. Then, the standard deviation of that is the square root, which is 0.5.

Now, we want to know the standard error for our estimate of the bot proportion. That is sqrt(p(1-p)/n). Suppose 50% of accounts are bots (I assume that would be very high), then our estimate of p would be 0.5 and our standard error with a sample of 100 would be 0.05. Hence, our 95% confidence interval is roughly 0.4–0.6 in the worst case (with a sample of 100).

If the proportion is under 0.1 (let's assume 0.05), then the standard error would be sqrt(0.05(1-0.05)/100) = 0.022. Our 95% confidence interval in this case would be roughly 0.01–0.09.

These seem like large ranges to me. Hence, I would expect them to use a larger sample too.

I recently had Intuit delete all my personal data, across all their products. Doing this was exhausting. Their online process for deleting an account did not work. Contacting customer support resulted in unhelpful back and forth. Finally, I gave up and filed a complaint with the CA Attorney General. Within two weeks, I got a call from someone at Intuit who quickly resolved the issue and promised to delete all my Intuit-owned accounts (it seems to have worked).

I tell the story because I think it illustrates a lot of reasons companies can get away with bad behavior:

1. Only a subset of consumers recognize the bad behavior and know who to contact.

2. Most consumers are not motivated to complain. In the above story, most people would probably stop when the Intuit website doesn't work (I usually give up too). There is a free-rider problem. My complaint could benefit lots of people if a company changes its behavior, but I alone incur the cost of complaining.

3. When a consumer successfully complains, companies can sometimes quietly make the problem go away for the one consumer and avoid regulatory action. Intuit called me, resolved the issue only for me, and we both move on. With the issue resolved, the regulator has less reason to continue investigating.

4. Even when regulators get enough complaints and go after bad behavior, they are up against powerful attorneys and lobbyists. And even if the regulators win, the company probably has lots of substitute ways of achieving the same goals that weren't contemplated or prohibited by the settlement/law/etc.

Not sure I follow your question.

An example of something that's not human subjects research would be emailing people who have websites and asking about their privacy policy.

An example of something that is human subjects research would be emailing people who have websites and asking what inspired them to start a website.

I realize that may seem like a subtle difference, but it's an important distinction from an IRB perspective. For reference, and because a lot of people seem confused on this thread, here's what the human subjects research training at my university says about this...

"...some research that involves interactions with people does not meet the regulatory definition of research with human subjects because the focus of the investigation is not the opinions, characteristics, or behavior of the individual. In other words, the information being elicited is not about the individual ('whom'), but rather is about 'what.' For example, if a researcher calls the director of a shelter for battered women and asks her for the average length of stay of the women who use the shelter, that inquiry would not meet the definition of research with human subjects because the information requested is not 'about' the director. If the researcher interviewed the director about her training, experience, and how she defines the problem of battering, then the inquiry becomes about her - and therefore 'about whom.'"

That's not the definition of human subjects research. Not everything that involves a human responding to questions is human subjects research. A lot of comments in this thread are uninformed about the relevant definitions.

You can think the study is poorly designed, unethical, etc. But if they're not obtaining data about a living individual, then it's not human subjects research.

This is incorrect. It's only human subjects research if the researcher is obtaining data about a human. This is the "about whom" requirement. A classic example is calling a business and asking someone about the products and prices they offer. That's not human subjects research.

My high school history teacher recommended his book to me when I was junior. I loved it, and even wrote to the author regarding a mistake I thought I found in the book (pretty arrogant of me). Loewen replied generously. I don't remember the exact exchange, but he was so encouraging. It meant a lot to me in high school that he replied.

I got this email and started to transfer my library only to learn it requires also creating a YouTube channel. The message says...

> On YouTube Music, playlists are stored on your channel. To continue transferring your Google Play Music playlists, create your channel now.

> All your Google services will display your channel name. Learn more

Huh? I don't really know what a YouTube channel is and probably don't want one. Does this mean my playlists will be public? What does the second part about all services displaying my channel name even mean? Clicking "Learn more" does not help. I'm having a hard time figuring out the privacy and other implications of transferring my music library. Why does listening to the music I uploaded have to impact all my other Google services? I don't want a YouTube channel, I don't want to be social, and I don't want other people seeing the music I listen to.