Yeah, I might be naive, since they're professional mathematicians and I'm not, but their conclusion feels like saying that they "tested" the concept of infinity and found that it's simply unreachable. To me, the infinite monkey theorem is an abstract idea and not something that needs to be "tested", though it's certainly fun to run the numbers.
They didn't "test" the concept of infinity. They said that the results of the infinite monkey theorem are well known, and they wanted to see what happens in the finite case.
Not the OP, but if doing this at scale, I'd consider a quorum approach using several models and looking for a majority to agree (otherwise bump it for human review). You could also get two different approaches out of each model by using purely the model and external OCR + model and compare those too.
I’m working on a problem in this space, and that’s the approach I’m taking.
More detailed explanation: I have to OCR dense, handwritten data using technical codes. Luckily, the form designers included intermediate steps. The intermediate fields are amenable to Textract, so I can use a multimodal model to OCR the full table and then error check.
Jina did something related for extracting content from raw HTML and wrote about the techniques they used here: https://jina.ai/news/reader-lm-small-language-models-for-cle... .. in my tests, the 1.5bn model works extremely well, though the open model is non commercial.
(This isn't a political comment, honest! But..) After reading The Art of the Deal, I get the impression this is a lesson Trump took to heart early in life. Always using other people's money to build and do things which, in turn, seems to give him more power than he might otherwise.
Also explains how his assistants end up having to put themselves at greater and greater risk. The more they help him, the more trouble they face, and the more they hope he is able to bail them out.
I'm glad someone else has noticed this. The quality and availability of fruit and veg at mainstream supermarkets has fallen off a cliff since COVID. It all feels far "older" and less fresh than it did, even when it's new out on the shelf. Traditional greengrocers are still fine, but tend to stock more local produce, so I'm wondering if it's Brexit or shipping causing the issue rather than expiry dates.
I sorta have my own personal version of this. A script that goes through the firehose on https://hnrss.github.io/ and runs everything through an LLM to filter down to things I'm interested in which then end up in an RSS feed I'm subscribed to. It works fantastically. I'm not interested in popularity though, just the items, because a lot of great stuff gets submitted to HN that never gets anywhere.
One of my dad's oft-told stories is of having his grandmother "in the living room" overnight prior to her funeral. He didn't use a particular term for it but said he found it quite terrifying as a 4-5 year old at the time. (This would be in the 1950s in the UK.)
Postgres, notably, has not had these problems. There's a thriving ecosystem, despite the trademark, and many providers offer "Postgres" services without Postgres' core organizations or contributors getting their undies in a twist over it.
Also worth considering that EDB is backed by Private Equity, and there was some other very recent incident that seemingly directly resulted in OtterTune folding.
I think a difference there is that Postgres doesn't have a for-profit semi-attached to it.
There are certainly companies that do work on Postgres, but Postgres wasn't founded by people looking to make a business and its development isn't driven by one primary company (to my knowledge). Postgres started as an academic research project by Turing Award winner Michael Stonebraker. Berkeley released it under a BSD/MIT-like license. It just has a long history of being independent of any company that's the primary driver of its evolution.
That's not to say there aren't companies like EnterpriseDB, Neon, Citus, and others that haven't driven certain aspects of it, but they just don't get the same kind of control over the project.
Crucially, no one can really feel like someone else is making money off a project that's primarily their work. I think companies in the Postgres ecosystem all understand that even if they're a big fish in the Postgres ecosystem, they aren't coming anywhere close to having built 25% of the value in Postgres. It's hard to "get your undies in a twist" if you acknowledge that you've probably gotten more from the historical contributions than you've contributed - even if you're a stellar contributor today.
I agree with his thesis, if not the details. The first thing I do when encountering content is skim it, check out its form and shape, writing style, look at the headings, pull quotes, etc. then if I judge it worth reading, I'll go back and read it "properly". It could say 2 minutes or 20 hours and I'm still going to size it up because otherwise how do you know it doesn't just peter out in some silly way and end up wasting your time anyway?
