Long horizon events like this on Polymarket stabilize around a % odds corresponding to time value of money. You can get 4% buying risk free CDs for that horizon.
Split brain experiments show that a person rationalizes and accommodates their own behavior even when "they" didn't choose to perform an action[1]. I wonder if ML-based implants which extrapolate behavior from CNS signals may actually drive behavior that a person wouldn't intrinsically choose, yet the person accommodates that behavior as coming from their own free will.
Noteworthy: this is powered by a new ultra-low-latency userspace TB5 driver offering an ibverbs/RDMA interface over thunderbolt 5[1]. Non-inference HPC workloads could see a massive benefit as well.
I figured one of the really useful applications of LLMs would be in processing transcripts of city government public meetings.
First I was doing it for myself, but I figured others might be interested to read these summaries as well. I use a few prompts to semi-automatically create reports of the most discussed topics for various board/commission meetings.
Ad free, done in my spare time. Just figured people might be interested.
This could be useful for e.g. bazel, where I’ve regularly seen deleting the bazel caches take on the order of 10 minutes because of absurdly large and repeated directory trees (caused by things like runfile trees containing the python interpreter).
It’s somewhat true in the sense that banks need to keep some fraction of their deposits in cash as reserves (the “reserve ratio”). So if they lend money it needs to be backed to that extent by their deposits.
Hash tables are (usually) faster to do all sorts of operations than tree based maps, as most operations become a simple function to calculate a tree’s hash followed by a table lookup. Of course, they’re unordered, so if you need to iterate in order, or find all keys < a certain value, or things like that, tree maps can be better for your algorithm.
Also, TreeMap uses a red-black tree to implement the map, which is a basic type of binary tree. Depending on the data you’d like to store, other kinds of tree-based maps can have better performance characteristics. A map based on a Splay Tree[1] speeds up repeated accesses, so it could perform well if you had keys that were cheap to compute an ordering but expensive to compute a hash, and your access pattern has good temporal locality.
I swear by type hinting in Python, but it is pretty frustrating how many common patterns don’t work or require very convoluted use of typing.
It’s an uphill battle to convert a team over to using hinting because of how awkward things can get and how easy it is to just pretend the feature doesn’t exist.
Note that it’s a qualified claim: 3x as efficient as the “average propellor of a small boat”. The tech looks awesome in the video though, would recommend giving it a watch.
I think the intent would be to filter out SEO spam essentially, and get you a small primary-source website. Like searching for "san francisco restaurants" gets you some much more interesting results on MS than on Google.
One particular thing that has always deeply concerned me about Amazon is the cognitive dissonance: they aren't a store but a platform, and yet they can exert total control over the store in an anticompetitive way when it suits them.
They banned selling the Chromecast and Apple TV [1]. Isn't that curious? They don't have to exert quality control over the Amazon marketplace because they just Connect Sellers to Buyers. But if that seller happens to compete with an Amazon product, sorry! No-can-sell.
Fault injection is a really interesting avenue for exploiting things like game consoles, where bootloader exploits tend to be the holy grail and where dumping decryption keys is huge for emulators.
It was used on the Nintendo 3DS, for example, to get the CPU to take an exception during bootROM execution [1], which combined with some other flaws allowed dumping the ROM with all of its keys.
If anything, projects like these can be interesting time capsules in the future, both preserving the city itself and providing some clues about which landmarks were most detailed, and which parts of those landmarks were so detailed.
One thing that comes to mind is UC Berkeley's "Blockeley"[1] server a bunch of students built in the same vein. Some very specific parts of campus are insanely detailed; the V&A Cafe even has furniture where most buildings are hollow facades.
It's also interesting as a virtual meetup environment... at Berkeley, the official graduation this year was even held in the Minecraft server.