Curation notice: a good old fashion fact post. It's relevant, detailed, and legibly written.

Continuing the list...

• mother can't be separated from the baby for longer than it takes them to get hungry, and must handle every nighttime wakeup. In the newborn phase that can be every 1-3 hours.
• you can get around this by pumping, but this has its own costs. Pumping is uncomfortable to painful, time consuming, and then has all the inconvenience of formula feeding and then some (like maintaining a cold chain, and more steps requiring sterilization).  It's also a hardcore logistical puzzle to get as much stored milk as possible without underfeeding your infant in the moment. The best solutions are the least convenient to the mother. 
• and this is all pretty best case scenario. Lots of women or their babies have medical impediment or just don't produce enough milk, even on medication.
• some babies suck at breastfeeding and need a bottle. you could always pump and bottle feed, but as we covered, pumping has its own cost. 
• some women need medications that are contraindicated by breastfeeding.

None of this contradicts the evidence that breastfeeding is beneficial, or easier for some people. But the frame should be "this is (usually) a sacrifice that we want to quantify the benefits of, to figure out if it's worth it" not "hey, free value!"

Huh, yeah, does seem like Claude was the winner there. I reproed the intellectual disability answers and got the same results you did. I was able to get a better answer from Perplexity with a slight rephrase, but I hate having to play the rephrase game with AIs so that's a modest mitigator. And the answer was not internally consistent

I think the difference between us might be that I do primarily want a search engine, and perhaps my natural phrasing works better with Perplexity. 

I got my first hallucination shortly after posting this- it's definitely not perfect. But I still find the ease of checking a big improvement over other models. 

I haven't, but if you have I'd love if you posted the results here. 

I assume you're using claude pro? Because I found the top free version unusable. 

Could you post some questions you've run on both and their answers?

Projects I might do or wish someone else would do:

• Compare AI research assistants more intensively.
  • Compare individual Perplexity models
  • Compare paper reading ability across models. 
• Look into that fluoride paper

These results are all from the vanilla UI. Comparing individual models on harder tasks is on my maybe list, but the project was rapidly suffering from scope creep so I rushed it out the door. 

You can also retroactively support my work in the EA Community Choice grants. This is narratively for my covid work, but if you would like to label it with something else that's allowed. 

What happened with Approximate Entropy was that Chaos could be useful, it just wasn't as useful as a pure information theory derived solution. Wouldn't surprise me if that were true here as well.

Gleick gives Mandelbrot credit for this, but it wouldn't be the first major misrepresentation I've found in the Gleick book.

I know someone's gonna ask so let me share the most powerful misrepresentation I've found so far: Gleick talks about the Chaos Cabal at UC Santa Cruz creating the visualization tools for Chaos all on their own. In The Chaos Avante Garde, Professor Ralph Abraham describes himself as a supporter of the students (could be misleading) and, separately, founding the Visual Math Project. VMP started with tools for undergrads but he claims ownership of chaos work within a few years. I don't know if the Chaos Cabal literally worked under Abraham, but it sure seems likely the presence of VMP affected their own visualization tools.