pushrax

just send sighup to nginx and it will reload all the config—there's very few settings that require a restart

It's true that turbopuffer does vector search, though it also does BM25.

The biggest difference at a low level is that turbopuffer records have unique primary keys, and can be updated, like in a normal database. Old records that were overwritten won't be returned in searches. The LSM tree storage engine is used to achieve this. The LSM tree also enables maintenance of global indexes that can be used for efficient retrieval without any time-based filter.

Quickwit records are immutable. You can't overwrite a record (well, you can, but overwritten records will also be returned in searches). The data files it produces are organized into a time series, and if you don't pass a time-based filter it has to look at every file.

LSM tree storage engine vs time series storage engine, similar philosophy but different use cases

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6612475/ is a review of Britton's research discussed in the article. It presents several points of evidence with a coherent argument for why meditation brings benefits while an excessive level of meditation may cause adverse effects.

Proper brokerages will pay you to lend your securities, transparently disclose their programs, and require consent. For example:

- https://www.interactivebrokers.com/en/pricing/stock-yield-en...

- https://www.fidelity.com/trading/fully-paid-lending

Some brokerages lend out spare cash (and pay a transparent interest rate, and are subject to strict reserve requirements) but the majority of assets controlled by a typical brokerage are securities. FTX assets were all subject to their leverage strategies, with no specific reserve.

JPEG XL implements lossless image compression, but that's definitely not the most interesting feature. It also implements lossless JPEG recompression. So your existing JPEGs can be served with ~20% less bandwidth, without quality loss.

Unlike AVIF, JPEG XL also has advanced progressive delivery features, which is useful for the web. And if you look at the testing described in the post, JPEG XL also achieved higher subjective quality per compressed bit, despite having a faster encoder.

Production audio and video recorders generate or intake an SMPTE timecode signal, and stamp recordings with this timecode.

This timecode format is a timestamp with seconds resolution plus a frame count within each second. To properly sync, all the timecode generators must use the same framerate. In other words, the audio recorder’s timecode framerate needs to match the camera.

I wrote 100ns, and meant an analog repeater. Full digital regeneration is probably in the 1-10μs range.

In theory having fewer repeaters improves latency, probably in the range of 100ns per repeater. I don't know how much of a practical effect that has, likely very minimal with modern implementations.

Either way it's more sensible to build high throughput microwave networks given the tiny amount of shortwave bandwidth we have.

Shortwave radio can be transmitted around the curve of Earth by ionospheric reflection and refraction so fewer repeaters are needed. This allows crossing vast oceans where microwave infrastructure might not be possible.

As you say the downside is available bandwidth and throughput.

Video description:

“Chosen by fair dice roll. Guaranteed to be random.”

Worst case you just add more layers, like we do with sound absorbers now.

If some kind of moth wing inspired metamaterial can shrink the width of broadband sound absorbing panels by 10x that would be quite useful still.

Reaper and FabFilter getting on board would definitely help make this standard gain some momentum.

The list of “Following companies and projects are already evaluating CLAP for their host and plug-in software” is pretty impressive and personally accounts for a large amount of my workflow. Hopefully it gets big enough for Native Instruments to implement, Kontakt could benefit significantly from some of the new features.

> I'd be bothered with my own stupidity if i did not cheat.

What? I'd be bothered with my own stupidity if I felt I couldn't get a good grade without cheating. I put in a minimal amount of time/effort in university where the subject matter didn't interest me, but always did fine without cheating.

According to the sources I've read, it uses a tiled rasterizing architecture but it's not deferred in the same way as typical mobile TBDR that bins all vertexes before starting rasterization, deferring all rasterization after all vertex generation, and flushing each tile to the framebuffer once.

NV seems to rasterize vertexes in small batches (i.e. immediately) but buffers the rasterizer output on die in tiles. There can still be significant overlap between vertex generation and rasterization. Those tiles are flushed to the framebuffer, potentially before they are fully rendered, and potentially multiple times per draw call depending on the vertex ordering. They do some primitive reordering to try to avoid flushing as much, but it's not a full deferred architecture.

> football player style contracts for engineers, where you are tied in to a team for N years, and with a requirement that another team has to pay big money for your contract if they want you to transfer?

i.e. golden handcuff equity grants with vesting schedules? Top performers in highly demanded areas can have some or all of their remaining equity bought out.

It definitely ties the value of the contract to the stock price. In a way the company is leveraging its stock - significant declines hurt talent retention, and significant gains help it.

In casual English “or” means “xor” most of the time.

I’m not sure I understand this argument. Why can’t users simply use the mouse wirelessly even if it’s possible to use in a wired way. Are you claiming that people would end up leaving the mouse plugged in, like they do with the trackpad? If the point is that using a wireless mouse is a better experience, then why wouldn’t people use it wirelessly except to charge?

I have an Apple keyboard and trackpad that I use at my TV wirelessly but have benefited from the ability to use while charging a few times when the battery died. Doesn’t stop me from using it wirelessly 99% of the time because the experience is better.

By the way, I have an older phone I leave plugged in all the time by my bed, the battery life on it has deteriorated but it works perfectly for music.

Apple’s own wireless trackpad and wireless keyboards work while being plugged in.

I think the sleek visual and ergonomic design specifically resulted in the charge port being put on the bottom, which resulted in their decision to disable the mouse when it’s charging. Not the other way around.

All else equal, having child related expenses later in life rather than earlier will only leave you richer due to time value of money. Assuming you plan ahead and save.

It’s more common in CS than engineering in my experience, when CS is part of the math department.

Agreed, I self hosted for many years and eventually gave it up. Maintaining high IP trust is tricky.

A fun hack is to run a captcha-free fake forum and allow random bot signups to generate outbound mail volume and trust.

Their recent laptops use Apple silicon and are anything but an overheating mess. There are plenty of things to complain about but heat has been solved.

I’ve destroyed two MacBook Air screens, also because the lid isn’t particularly rigid. The MacBook Pro is definitely more durable, no issues after 12 years of owning them. I’m now much more careful with my MBA.

I will seek when the option is presented. It’s either tiny or nonexistent though (e.g. for gifs, or if the author didn’t enable the video controls, which default to disabled in many browsers). I’m happiest when a video loads paused, and has a giant play button on it.

How are you scrolling? When I’m browsing and reading my hand lives on the mouse. On mobile it’s even easier to click to play. In either case, clicking on a video takes less than 1 second.

I do agree that extremely short loops can be a different story, but 20 seconds is by far over that time for me. 5 seconds is around the threshold.

I’m sure this preference varies by person.

I agree completely. When there's a looping video below the fold, I'd rather start it with a click, instead of having to watch it up to twice just to see the clip from beginning to end.

Are you sure it was actually using the index you expected? There are subtleties in index field order that can prevent the query optimizer from using an index that you might think it should be using.

One common misstep is having a table with columns like (id, user, date), with an index on (user, id) and on (date, id), then issuing a query like "... WHERE user = 1 and date > ...". There is no optimal index for that query, so the optimizer will have to guess which one is better, or try to intersect them. In this example, it might use only the (user, id) index, and scan the date for all records with that user. A better index for this query would be (user, date, id).

This entirely depends on what kind of index is used. A sorted index, such as a B or B+ tree (used in many SQL databases), will allow for fast point/range lookups in a continuous value space. A typical inverted index or hash based index only allows point lookups of specific values in a discrete value space.

That has happened before, but very rarely, and I highly doubt will ever happen to SponsorBlock.

The reality at scale is that all cloud VMs are preemptible and unreliable. You can pay more to increase the probability of uptime but inevitably have to recover from failure sometimes.