dangom

The choice happens implicitly rather than explicitly. If Claude tries an approach and hits a wall, it'll try a different approach. If an API call keeps not working, it'll choose a different API. It a tool is broken, it'll use something else. If it can't find docs nor read the code, it'll try to implement functionality from scratch. If you give it messy tools with confusing docs, you'll notice Claude not calling them as you'd expect, and instead trying something simpler instead.

I think the content-centric model you are describing has been alive and thriving since at least the late 70s in Emacs.

Do you have any references or examples that describe how this simplification would come about? Would love to learn more about it.

Depends on what you mean by "improvements". Is it coordination? Is it sustained increased blood flow? I would imagine that different bike exercise regimens could induce more variation in fitness than the comparison dance vs exercise alone.

One can always find positive and negative outcomes related to any intervention to a biological system. Fasting is no exception. The question is when and where is it beneficial, and what are the trade-offs. I'm sure if one has a clean, healthy diet, and consistent sleep and routine, it likely does not matter in the long run at what time one decides to eat or not eat. If the effect size were noticeable we'd have seen it already in smaller samples.

If one is overeating, or eating garbage all the time, then I'd hypothesize fasting to be beneficial by giving the biological system a break to try and bring itself back to a better steady-state without so much forced external input.

This keeps coming up. Is this depth story not simple to fact check with simulations? Are we missing something?

Nice. Love vim too, but letting go of org mode is too much of a negative to justify a switch. I know I could use Emacs just for org mode, and vim for everything else, but that seems like even more overhead.

I appreciate that people are looking into this, and I think more research of this kind should be funded. However, one needs to be cautious when claiming an "impairment" in fear regulating regions from a measure of cortical thickness with MRI. For once, there is no evidence that a slightly thinner or larger cortical thickness means better or worse regulation of any kind. Second, morphology studies are susceptible to many sources of biases which are not really addressed in this study. For example, anything that affects hydration levels or causes a redistribution of blood and cerebral spinal fluid volume can lead to significant changes in measures of cortical thickness, since they will change the contrast between gray and white matter that drive cortical thickness measurements.

Changes in thickness have been found even when comparing people scanned in the morning and in the evening [1]. Any drug intervention could be expected to cause physiological changes that could act as confounds. Also, in the discussion one reads: "Interestingly, no lasting effects of combined oral contraceptives (COCs) use were detected when comparing the four groups." This suggests that the changes could be driven by physiological changes instead of permanent changes in brain circuitry.

It would be great to see a follow up study controlling for potential confounding effects (for example, measuring baseline perfusion, blood pressure and controlling for time of day effects and usage of other drugs), and expanding the study with functional tests that involve fear regulation.

[1] https://www.sciencedirect.com/science/article/pii/S105381191...

https://www.youtube.com/watch?v=6BhV-o77RqQ

Does anyone know why the decision to ban it? Was it a move that had a relatively higher injury rate compared to other high difficulty moves?

Because we can safely assume that under normal atmospheric conditions and within the distances we are talking about, light travels in straight lines and the SUV is too high thus blocking some of it from reaching the drivers eyes. I can't pull the sources now, but there was a study discussed here on HN a couple months ago showing an analysis of how this plays out.

Name your frames and the problem is solved. I for instance love the model because I no longer have to be tracking down millions of filenames - just one for each project.

Saw this on a separate thread the other day:

"Get some weights. Pick up the weights. Put down the weights. Eat healthy food." There is no real secret. No need for a gym to do those.

But now more seriously - I'd suggest just getting a pull up bar. You can then do pull-ups in addition to push-ups, squats and some core work, and that'll hit all of the main muscle groups. You can do that at home.

You may not see any progress, but that doesn't mean there isn't any. You are perfecting the moves you are making, and strengthening your bones, ligaments and tendons, which just don't develop nearly as fast as muscles.

Staying at a plateau for a while is sometimes great to avoid injury.

If the AI objectively does a better job at summarizing the meeting, then maybe someone who cares about the project should use the AI instead of taking notes by hand.

Exactly, that there is an end to the rabbit hole is a limitation of today's models. If something does not exist, it should be generated on the spot. GPT5 should check for the existence of an API and if it exists, test and validate it. If it fails tests or doesn't exist, create it.

The point of the article is that we don't like being idle. We'd rather spend our idle time "pretending" we are being productive, and tools for thought are what we use for that.

Being actually productive (quality > quantity), I argue, is a process that takes physical time. Absorbing information, internalizing it, and summarizing it with our own understanding requires a lot of energy. This process cannot be massively accelerated. Same as with physical fitness, one can operate close to optimum and see and maintain great results, but one cannot operate better than optimum given one's own physical constraints.

For intelectual work, defining what "operating close to optimum" means is much harder because the quantity of output is usually the metric, and that varies so much from discipline to discipline and person to person. I believe many of us are already operating close to optimum (reading and writing, attending meetings, presenting our work), so there is no point in investing even more towards productivity. But the falacy is that because we don't have a proper metric for productivity, we believe investing even more is worthwhile since it increases output, and so we perceive ourselves as better.

I don't see AI changing the picture for us because the problem is not what we are doing, but how we perceive to be doing it. That's what's up with tools for thought and personal wikis.

Sure, just like working out continously is better than rest. There is no better or worse, you need both. Ideas take time to materialize.

Part of the appeal for "Tools for Thought" is that by using them we feel we are taking action towards being productive, regardless of whether that turns out to be true of not.

The falacy comes, I believe, from the combination of two facts: 1. much of the intellectual work we do these days simply takes time. No amount of writing can accelerate that beyond our biological limit of learning, so we might as well just sit and think. 2. Just sitting and thinking is considered unproductive and regarded as lazyness, so we believe we should be writing even more instead.

In that regard, using tools for thought may be pointless, since all we need is time to think. But perhaps that pointlessness serves a purpose. Like a guardrail in a highway, tools for thought are not something we "really need", but they're there to at least keep us on track in case we were to drift away while our minds move forward.

This idea is presented by Jeremy Howard on literally their first Deep Learning for Coders class (most recent edition). A student wanted to classify sounds, but only knew how to do vision, so they converted sounds to spectrograms, fine tuned the model on the labelled spectra, and the classification worked pretty well on test data. That of course does not take the merit away from the Riffusion authors though.

I feel the limitation is not that these references don't exist, but that GPT didn't go ahead and finish its job by also creating the content in the references on the fly.

How big a lens does one need to focus sunlight onto a cooking pan to boil water and cook pasta without emissions?

Funny salary ranges are disclosed literally every else and it's never been a problem except in the US.

Podcasts, concerts and books are something we review subjectively. Scientific articles one would expect would be reviewed objectively.

Any references for those studies?

Makes sense, but if there is a slack channel where people are supposed to be responsive and answer technical questions, then that's value beyond just material and course work. And given there is a community, if you set the price too low you start to get spammers and people that are not truly interested in the content, bringing down the value of the community in the first place. I think the price is fair.

Another killer feature: the ability to actually sync your org notes.

Looks very pretty indeed. Will give it a spin.

For those interested in alternative dictionary apps for the English language I'd also recommend checking out the advanced english dictionary [1] as well. It certainly checks all the boxes the author asked for and then some more.

EDIT: Just noticed the author was kind enough to share the source code. That's super cool - kudos for doing that.

https://apps.apple.com/us/app/advanced-english-dictionary/id...

What happened to the xi-editor?

What is the best research article on UI that you'd recommend to someone new to the field?