wmwmwm

Joseph Heller methinks, but probably not too far away in embedding space!

I’m guessing Ramanujan: https://en.wikipedia.org/wiki/Srinivasa_Ramanujan

Also had to remember not to drag people out of cars and drive off with them when walking felt too slow in real life!

Book recommendation for The World Without Us which explores what might happen if humans vanished overnight: https://en.m.wikipedia.org/wiki/The_World_Without_Us

My aero engineering friend from university winds me up every time I see him saying that pi = 22/7 - I finally stopped getting angry, checked and it’s pretty good! I’m still glad he didn’t decide to design planes after he graduated though!

The domain name and the story seem a bit incongruous...

I had one due to a bulging disc in my neck and some pins and needles - it was fascinating, but I still remember them turning on the audio and letting me listen to my nerves trying to flex various muscles! My biceps sounded like a bit like white noise (good), while my triceps sounded like a plastic bag flapping outside a car window (bad!) The way they explained it some subset of my tricep nerves were firing harder to try and compensate for the ones affected by the disc bulge - but they couldn’t keep up the firing rate by themselves. Really interesting and somewhat explained my mediocre push-up performance!

You can throw python threads at it, but if each request traverses the big old datastructure using python code and serialises a result then you’re stuck with only one live thread at a time (due to the GIL). In Java it’s so much easier especially if the datastructure is read only or is updated periodically in an atomic fashion. Every attempt to do something like this in python has led me to having to abandon nice pythonic datastructures, fiddle around with shared memory binary formats, before sighing and reaching for java! Especially annoying if the service makes use of handy libraries like numpy/pandas/scipy etc!

Historically I’ve written several services that load up some big datastructure (10s or 100s of GB), then expose an HTTP API on top of it. Every time I’ve done a quick implementation in Python of a service that then became popular (within a firm, so 100s or 1000s of clients) I’ve often ended up having to rewrite in Java so I can throw more threads at servicing the requests (often CPU heavy). I may have missed something but I couldn’t figure out how to get the multi-threaded performance out of Python but of course no-GIL looks interesting for this!

Same for me on Firefox, though Chromium and Safari seem to work

Does anyone have any insight into why reinforcement learning is (maybe) required/historically favoured? There was an interesting paper recently suggesting that you can use a preference learning objective directly and get a similar/better result without the RL machinery - but I lack the right intuition to know whether RLHF offers some additional magic! Here’s the “ Direct Preference Optimization ” paper: https://arxiv.org/abs/2305.18290

There’s also a python api for getting the transcript from a given youtube video id so you can script the whole thing

I’ve had some pretty remarkable results pasting lecture transcripts from youtube into gpt4 and getting well formatted/relevant markdown summaries from meandering and mis-transcribed content! Needs chunking up but surprisingly effective. It can even generate youtube urls with the right timestamps if you ask it nicely

Not super related but a pleasantly surprising linux /dev trick I discovered yesterday was piping “hello” to /dev/udp/<ip>/<port> in order to check connectivity between two machines (the other was listening using nc)

I think it is actually meant to be "scruples" - so it sounds pretty accurate to me:

https://www.ibiblio.org/ebooks/James/Turn_Screw.pdf (see p4)

This is amazing, and scary (as a musician) but also reliably kills firefox on iOS!

As someone who’s been prompted every other night over the last couple of years for a fresh made-up story with an ever-changing cast of dinosaurs, superheroes and aliens by my five year old (who complains if the story arc isn’t to his liking) I’ve been very grateful for a few ChatGPT synopses recently!

On the other hand, I wouldn’t want to just leave him to it without me editing/steering the story as I tell it.

Something like “give me 3 different story ideas with these characters suitable for a five year old” followed by “i like number 3, please expand but make the dragon good instead of bad” has been pretty impressive, and is a relief when I’m out of ideas. Coming up with 180 meaningful improvised stories each year is a tall order!

I can recommend “Other Minds” by Peter Godfrey-Smith for a pretty in depth read about octopus intelligence. Fascinating stuff!

Actually, I think it was the initials of the original publishers

I love this story about Agloe, New York - a "copyright trap" that materialised into a real place:

https://en.wikipedia.org/wiki/Agloe,_New_York

When another map maker put the actual settlement on their maps, the original publishers cried foul, but then discovered Agloe had become real!

I remember several times putting my eldest son to bed when he was a few months old and having this intrusive thought that maybe the reason why he was yelling was because he knew he wouldn’t remember today when he woke up the next day and that every day was like being reborn! Darker than a Black Mirror episode for sure! It only went away when he started developing more of his own personality

(This stuff + sleep deprivation is a heady cocktail!)

I’m a fan of that book too - though hoping I never need it!

On a tangential theme, another one I read and liked at the same time was The World Without Us - all about what would happen to the cities and infrastructure if all humans suddenly vanished overnight. Kind of depressing but there’s a lot of interesting and non obvious stuff in there

Having just implemented a softmax() function for an online ML course, I think the python implementation here suffers from overflow if any of the elements of z get big(ish) - e.g. e^10000 is a big number! A spot of searching online suggests that subtracting max(z) from all entries in z makes it a lot more robust without changing the result e.g. https://www.tutorialexample.com/implement-softmax-function-w...

Although I haven't used LaTeX in decades to write anything substantial like a paper or dissertation, I recently had cause to use it to write up some homework for an online course I'm doing - mainly (pretty simple) maths proofs. After 20 years of not doing maths, my ability to faithfully rearrange handwritten equations from one line to another was even more rubbish that it was in my 20s! All sorts of dropped or flipped signs, terms and stray brackets would trip me up. Although initially sceptical, I found that doing this in LaTeX (with the ability to cut/paste to a new line and then manually 'refactor') was way more reliable. I think it tapped into my ability to do this semi-unconsciously when coding! (And with no dreaded page turns...)

However, it's still an effort to squint at all of LaTeX's curly brackets and long form symbol names, and LaTeX doesn't grok the equation of course.

Has anyone come across any tools out there that offer a similar experience for manually rearranging equations and somehow 'linting' as you go along? i.e. something that knows a bit more about precedence/structure etc? Even a intellij ctrl-W style shortcut (highlights chunks with increasing scope) would make it easier/quicker to do what I'm talking about.

I really like that you can use duckdb to sql query an on disk directory structure of parquet files with no preloading into RAM or other db formats. Super useful/quick and only one line of code! Instead of selecting from a table you just pass a glob pattern into the SQL - and since it’s a oneliner I can use it in adhoc notebooks too. It even has a to_df() method on the query result so you can get it into pandas for further manipulation

I met someone years ago who was an adjudicator for Guinness. I remember two stories in particular. One was a trip to Venezuela to judge a biggest bowl of soup (chicken IIRC), the other to Italy for a biggest doner kebab! (yes Italy not Greece). She said her biggest concern was that the doner kebab would fall off an underspecced trolley and kill someone, though I imagine that would probably have achieved a record of its own!

Yes, wouldn’t want to try and design it myself! I think the datacentre providers I was looking at spend quite a while doing CFD simulations to get it right!

I was recently researching how you'd host systems like this in a datacentre and was blown away to find out that you can cool 40kW in a single air cooled rack - this might be old news for many, but it was 2x or 3x what I expected! Glad I'm not paying the electricity bill :)

I remember reading it and wondering whether they’d have had more chance of making their fortunes building home computers out of their garages given the era! Great story though (I read the British edition - The Newtonian Casino)

As someone who finds himself thinking “wtf is money” every decade or so, the most recent book I read on this was pretty good - Money: The Unauthorised Biography by Felix Martin. It’s firmly in the “social construct” camp and makes some good arguments (to my lay mind!). There’s some really interesting stuff about the Fei stones https://en.m.wikipedia.org/wiki/Rai_stones