cafebeen

It's a known issue now! I have have all notifications silenced, so I haven't seen this behavior. Will reproduce and try look into a fix, thanks

Yes, I had a related experience of reading a book and observing what I thought were claude-isms, only to realize it was written in 2019. Some of the common tells are actually good writing practices, but I guess they are best in smaller doses.

Indeed, if only Apple still made an 11-inch macbook air...

I agree this should be built-in. Many apps have their own "focus mode", but I wanted something that works uniformly across all apps. There is a full-screen mode in macOS, but it stretches the widow and context to full width, which doesn't pan out too well in some apps (lots of empty space or lines that are too long to read comfortably). So this preserves the window width but extends the height when focused, which works well for my daily drivers. The "fade" is somewhat unusual too, which means everything else is darkened close to black, which I find visually suppressing things and reduces distraction.

By the way, repo has an animated demo in case you want to see how it works.

Would be great, and it's technically possible to spotlight multiple windows in this setup. But I wonder about the mechanics of picking the right apps. One way is to use named sets, e.g. dev=(iTerm2, Safari), each with their own shortcut. Another is to to focus one window and then use Cmd+Tab as an picker to choose which app to add next. I guess it depends on whether the arrangement is changing or relatively stable.

Author here, I made this macOS tool "Zen Mode" to provide a global focus mode that works across apps. There are plenty of things approaching this (HazeOver, Stage Manager, built-in full-screen mode, etc.) but nothing quite like I imagined. The basic idea is to have a shortcut to bring the current window front and center and fade everything else to black like a spotlight, and when you're done, the window returns home, and the lights come back up. A simple solution with about 250 lines of Hammerspoon Lua, built with Fable, though with many iterations to get the UX right. Happy to answer any questions or hear any feedback.

Yes, and one of my favorite anecdotes like this: at one of the greatest jazz concert ever recorded, Charlie Parker played a cheap plastic saxophone because he hadn't brought his own.

https://jazzfuel.com/charlie-parker-the-plastic-saxophone-th...

While not a cure, there are many known modifiable risk factors for Alzheimer’s, so to some extent we know enough to deflect / mitigate dementia: https://www.thelancet.com/journals/lancet/article/PIIS0140-6...

This is great and echoes my experience. Although I would add a caveat that this mostly applies to solo work. Once you need to collaborate or operate on a team, many of limits of self-hosting return.

I believe slash commands are all loaded into the initial context and executed when invoked by the user. Skills on the other hand only load the name and description into initial context, and the agent (not user) determines when to invoke them, and only then is the whole skill loaded into context. So skills shift decision making to the agent and use progressive disclosure for context efficiency.

Thank you for publishing that paper, which I think greatly helped address this problem at the time, which you accurately describe. I guess things have to be taken in their historical context, and science is a community project which may not uniformly follow best practices, but work like this can help get everyone in line! It's unfortunate, and no fault of the authors, that the general public has run wild with referencing this work to reject fMRI as a experimental technique. There's plenty of different ways to criticize it today, for sure.

This study was really highlighting a statistical issue which would occur with any imaging technique with noise (which is unavoidable). If you measure enough things, you'll inevitably find some false positives. The solution is to use procedures such as Bonferroni and FDR to correct for the multiple tests, now a standard part of such imaging experiments. It's a valid critique, but it's worth highlighting that it's not specific to fMRI or evidence of shaky science unless you skip those steps (other separate factors may indicate shakiness though).

In a practical sense, I would argue the scientific is primarily about churning out grants and papers.

To a large extent, I think this could be solved by labs having more long-term permanent research staff (technicians, data analysts, scientists) and reducing the number of PhD students. Many students would gladly stay on in that position instead of leaving, so it increases job opportunities. It would also improve the quality of the science because the permanent staff would have more historical knowledge, in contrast to the current situation where students constantly rotate in and out with somewhat messy hand-offs. The students could also then focus more on scholarly work, planning and overseeing research execution with the team. The problem is that the incentives are aligned to allocate students to doing all lab tasks, not long term staff. I think we could change this through changes to the requirements and structure of science funding mechanisms however, since ultimately that's the source of the incentives.

Thanks for your work great work on Claude Code!

One other feature with CLAUDE.md I’ve found useful is imports: prepending @ to a file name will force it to be imported into context. Otherwise, whether a file is read and loaded to context is dependent on tool use and planning by the agent (even with explicit instructions like “read file.txt”). Of course this means you have to be judicial with imports.

When asking customers how well they were helped by the customer support system (via CSAT score), I've found industry-standard AI support agents will generally perform worse than a well-trained human support team. AI agents are fine at handling some simple queries, e.g. basic product and order informatino, but support issues are often biased towards high complexity, because otherwise they could solve it in a more automated. I'm sure it depends on the industry, and whether the customer's issue is truly novel.

Taking a shot at this, one concrete definition might be that the business model is essentially white labeling, that is, the base LLM is rebranded, but task performance in the problem domain is not functionally improved in some measurable way. As a corollary, it means the user could receive the same value if they had gone straight to the base LLM provider.

I think this might be more narrow than most uses of the term “wrapper” though.

Yes, this feature might be a prime driver of user engagement and retention, and it could even emerge "naturally" if those criteria are included for optimization in RLHF. In the same way that the infinite scrolling feed works in social media, the deranged sycophant might be the addictive hook for chatbots.

Budgets are finite, and most science funding involves some decision making about how to allocate resources.

It’s more like keeping a fire extinguisher in your kitchen. Some people may never need it, but grease fires happen often enough to plan for it.

Greg Lemond wrote about fitness, "it never gets easier, you just get faster", which perhaps applies as well to programming.

I think you're both right, but you might be thinking of different tasks and libraries. If we're talking about solving a specific type of PDE, then in that scenario he's probably right that MATLAB will work better out of the box, but if we're talking about logistic regression on data from a messy CSV table, then python will work great and have better usability than MATLAB (IMO). He might also be thinking of versioning and packaging headaches in python, which are quite a pain the first time exposed to them coming from a walled garden environment.

Here's a link to the associated paper just out in Nature:

https://www.nature.com/articles/s41586-022-04492-9

Marek et al. Reproducible brain-wide association studies require thousands of individuals (2022)

"Magnetic resonance imaging (MRI) has transformed our understanding of the human brain through well-replicated mapping of abilities to specific structures (for example, lesion studies) and functions1,2,3 (for example, task functional MRI (fMRI)). Mental health research and care have yet to realize similar advances from MRI. A primary challenge has been replicating associations between inter-individual differences in brain structure or function and complex cognitive or mental health phenotypes (brain-wide association studies (BWAS)). Such BWAS have typically relied on sample sizes appropriate for classical brain mapping4 (the median neuroimaging study sample size is about 25), but potentially too small for capturing reproducible brain–behavioural phenotype associations5,6. Here we used three of the largest neuroimaging datasets currently available—with a total sample size of around 50,000 individuals—to quantify BWAS effect sizes and reproducibility as a function of sample size. BWAS associations were smaller than previously thought, resulting in statistically underpowered studies, inflated effect sizes and replication failures at typical sample sizes. As sample sizes grew into the thousands, replication rates began to improve and effect size inflation decreased. More robust BWAS effects were detected for functional MRI (versus structural), cognitive tests (versus mental health questionnaires) and multivariate methods (versus univariate). Smaller than expected brain–phenotype associations and variability across population subsamples can explain widespread BWAS replication failures. In contrast to non-BWAS approaches with larger effects (for example, lesions, interventions and within-person), BWAS reproducibility requires samples with thousands of individuals."

A good example from physics is this observation related by Feynman in 1974:

"[...] It's interesting to look at the history of measurements of the charge of an electron, after Millikan. If you plot them as a function of time, you find that one is a little bit bigger than Millikan's, and the next one's a little bit bigger than that, and the next one's a little bit bigger than that, until finally they settle down to a number which is higher.

Why didn't they discover the new number was higher right away? It's a thing that scientists are ashamed of—this history—because it's apparent that people did things like this: When they got a number that was too high above Millikan's, they thought something must be wrong—and they would look for and find a reason why something might be wrong. When they got a number close to Millikan's value they didn't look so hard. And so they eliminated the numbers that were too far off, and did other things like that..."

Alas, the possibility of comments like this are one reason that some researchers hesitate to release their code.

Even if there are issues, it's tremendously valuable that they shared this, so others can reproduce and build on their work. That said, most scientists would like to have better quality code in science as well, but I think the incentive structures and other demands that make it challenging...

Yes, there is surely evidence that intelligence is heritable, but it's worth giving the research on this a close reading. Current approaches can only explain 10% of the variance in intelligence (https://www.nature.com/articles/nrg.2017.104), plus the effects from the environment. My main issue with the article though was the assumption that a "best" set of genes exists without futher discussion (in addition to obvious ethical concerns)

Agreed that intelligence is a good thing, but the question is about whether there is a set of genes that are “best” for producing it. At the moment it’s unclear which genes those are, and even if we found some genes related to intelligence, they may have other negative consequences that make the idea of a “best” set of genes questionable

The bigger issue seems to be determining what "best" means. While there are a handful of diseases associated with specific genetic variants that could obviously be cured, the traits people would likely try to optimize (e.g. intelligence) seem to involve a complex combination of genes and environment, and an attempt to select for such genes could incur some potentially negative side-effects.

I find raincloud plots to be the best of both worlds, and they have been my preference lately:

https://wellcomeopenresearch.org/articles/4-63/v1

They can visualize common statistics like box-plots, the distribution shape like violin plots, as well as the raw data!

Perhaps a person's interests, income, or lifestyle are salient variables as well? I would guess that each person's circle of friends has some of these in common in addition to age (likely true for HN discussions as well). For a less biased picture, here are some statistics about tablet use by age:

https://www.statista.com/statistics/805143/us-tablete-users-...

and for a baseline, iPhone use by age:

https://www.statista.com/statistics/203063/iphone-users-in-t...

It's striking how uniform the distribution of tablet use was across age! There was a bump after age 25, though that was mirrored in the iPhone usage. This is really salient at the low and high end: age 0-18 was 20% of tablet usage and only 6% of iPhone usage, and age 55+ was 20% of tablet usage and only 14% of iPhone usage. Given all of this, it seems like tablets are quite broadly adopted devices