mbeavitt

Honestly I've been doing a lot of image-related work recently and the biggest thing here for me is the 3x higher resolution images which can be submitted. This is huge for anyone working with graphs, scientific photographs, etc. The accuracy on a simple automated photograph processing pipeline I recently implemented with Opus 4.6 was about 40% which I was surprised at (simple OCR and recognition of basic features). It'll be interesting to see if 4.7 does much better.

I wonder if general purpose multimodal LLMs are beginning to eat the lunch of specific computer vision models - they are certainly easier to use.

Having briefly experienced weight loss drugs - and the bliss of that constant “EAT!” voice in your head just going quiet - I’m pretty convinced most humans have a genuine genetic predisposition to overeating.

And when you zoom out to the population level, the “we’re all autonomous individuals” argument gets a lot shakier. Like yeah, at the individual level you have agency, you make choices, fine. But at scale? We are absolutely at the mercy of whoever has figured out how to tickle our monkey brains in just the right way to get us buying their fattening food.

if you read the article...

Instead of discarding stock, companies are encouraged to manage their stock more effectively, handle returns, and explore alternatives such as resale, remanufacturing, donations, or reuse.

I guess remanufacturing/reuse might be the intended solution if it's absolutely not to be worn.

> Don't take criticism from someone you wouldn't take advice from.

The funny thing I find about criticism is that you actually don’t have a choice about whether or not it affects your future actions. Criticism that I have dismissed has persistently come back to haunt me, perhaps via my subconscious.

what evidence do you have to back up this baseless claim? They openly publish their financial reports: https://wikimediafoundation.org/who-we-are/financial-reports

$178 million might sound like an extremely large amount of money if you're a member of the general public, but for a global resource kept up to date that serves hundreds of billions of visitors per year this is actually not a huge quantity of money.

Oh, I get it. Missed the joke.

Is it though? The algorithm you reference is meant to be a joke:

1) Write down the problem.

2) Think real hard.

3) Write down the solution.

This is not a useful algorithm in any sense, apart from that it might be thought provoking.

What's the "universal nike algorithm"? I didn't find anything on google.

You can't tease me with "Algorithm" in the title and then not actually define an algorithm... TL;DR: Read lots of widely acclaimed books that are close to the source of what you want to learn?

Merge conflicts are a blessing - they tell us exactly what the problem is...

I can't imagine the horror of tracking down a regression and finally fixing it, only to find someone else edited another section of the codebase, ruining all my efforts.

This is fine in text documents (to an extent, obviously references to sections of text that no longer exist can happen) because different sections are not as inextricably linked to each other.

Just curious - what do you see as being a viable alternative to the current git paradigm?

Ok but the lab has access to the unencrypted data? You haven't removed the requirement that the user needs to trust the lab with their raw genome markers. This entire operation hinges on the lab's trustworthiness, does it not?

Simulating the real world at increasingly accurate scales is not that useful, because in biology - more than any other field - our assumptions are incorrect/flawed most of the time. The most useful thing simulations allow us to do is directly test those assumptions and in these cases, the simpler the model the better. Jeremy Gunawardena wrote a great piece on this: https://bmcbiol.biomedcentral.com/articles/10.1186/1741-7007...

Yes you're correct - I more meant that paths either side of the river and other infrastructure was originally put in place for this reason, not that the body of water itself was man-made.

The historical reason is that canals were originally built to facilitate horse-drawn boats, and as such there are no fences. The modern reason is that people use the canals all day, for hopping on/off houseboats and rowing training - putting up fences just for a few cows would make many people upset.

FYI, I live right next to this area and it's almost as central as you can get in Cambridge, so it's super busy - people and cyclists coming through this field all the time, including myself. There are plenty of fences around the roads and land borders of the commons, but it's just the canals that have these special uses that require a lack of fences.

https://maps.app.goo.gl/PifHtYpg1u2SfojF9

Perhaps they want to include online discussions/commentaries about their paper in the training data without including the paper itself

You mention you are looking for a good resource for training listening - have you tried https://tonedear.com/?

Haha I was just thinking to myself "I wonder if anyone made any progress on that rav1d bounty yet?"

Ah, thanks, that looks cool. It's missing the core functionality I want though, which is to follow links to code on github. Maybe if this doesn't exist, I'll see if I can make something!

No, the desired behaviour is to follow a link to a github code section, e.g. https://github.com/splicebox/PsiCLASS/blob/04e2a118c1820f802... and then copy that into the editor with syntax highlighting and line numbers. I can copy it myself, but it would be nice to have a plugin to make things look nice. Additionally, it would be nice to preserve the link so that readers can jump straight to the relevant section of the codebase.

An example of the kind of behaviour I'm after is at the top of this issue: https://github.com/splicebox/PsiCLASS/issues/41

Can this be used to effectively create R environments? I’m desperate for such a solution.

This is super cool. You provide examples of “searches that work” - can you give an idea of the limitations here? What kind of searches won’t work?

check out [cgdb](https://github.com/cgdb/cgdb) - ncurses based gdb tui, with vim bindings. It's no longer actively worked on, but is decently maintained with bugfixes etc AFAIK.

Did you use AI to write this blog post? Be honest.