ComputerGuru

I (lifelong side sleeper) started using a (dedicated) knee pillow about a decade ago and I am completely torn about them. On the one hand: amazing. Better spinal alignment, zero sensitivity or soreness where the knees meet, better sleep, easily the best $10 I ever spent. On the other hand: I can no longer tolerate the weight of my knees on each other for even a minute. Can’t sleep or even just side-lay for more than thirty seconds without a pillow or a folded blanket or something between my knees without it aching like hell and being too uncomfortable to bear. From speaking to others that adopted knee pillows, it seems to be the same: once you use them, you can never go back.

So now I marvel at how I used to be able to sleep without them and ponder over why we either have no fatty pads (or muscle, even!) on the inside sides of our knees or have reduced sensitivity in that region, since we’re clearly meant to be side sleepers (it’s the healthiest position, assuming you can find a rock or log to place under your head, or even just your hands folded together).

There is no good reason to believe language-specific models are going to be any meaningfully smaller, just worse. Same as English-only models vs those trained on a multilingual corpus.

Just to play devil’s advocate: you can’t compare Qwen to a (proprietary/closed source) hosted model and deduce that Qwen is overthinking, as Qwen gives you the full reasoning/thinking trace while all the proprietary models now give you only a summary “to prevent distillation”, making it hard to properly compare apples to apples here.

The tokenizers are included in the open s̶o̶u̶r̶c̶e̶ weights releases; you wouldn’t be able to use the weights without the corresponding encoder/decoder, in fact.

Look into tursodb for this.

What quantization level is that? Because official endpoints are slow.

I would have the same concerns but note that backup generators aren’t just run when needed but also have a weekly/monthly quota of hours they need to run for both testing and keeping them in shape.

What is MMD? That is not an svg.

Does the page actually load for anyone? I get stupid spa skeleton spinners.

It gave the svg for the wheels in the reasoning trace then forgot to include them in its final answer. Lol.

They could certainly still achieve that even if they had a slightly more expensive tier with a (pinky promise) "no training" bit set, or even if it were just an opt-in most users wouldn't bother with.

If only they would let you opt out of training use, it might actually be a viable option.

> I pay about $0.075 USD per kWh

Around here electricity companies quote prices like yours but that is supply only while transmission, taxes, and fees are again as much on top. Is that really all inclusive?

It’s not hard to use libavcodec to transcode, though.

Bandwidth is almost never a concern for databases. The right number to worry about instead is latency.

Very good point, but also unfortunately requires getting into the internals. If you can tolerate it just use fastupdate off.

In practice you’ll never notice the difference between joining on a bigint vs uuid for most OLTP purposes and you’ll be able to generate the uuids in your backend instead of in the db which a) can be very helpful if you’re pre-generating linked records and inserting them pipelined rather than sequentially, saving a lot of round trip traffic, b) reduces concurrency bottlenecks on the db server.

More usefully, a uuidv7 can take the place of both the id and the created_at field (if precision is sufficient), and can (depending on your security comfort) also take the place of the uuidv4 public id field.

With uuidv7 the data is naturally sorted so you don’t run into the issues with btree worst case scenarios that you would with a uuidv4 id, and you can even take it a step further and use BRIN instead of btree indexes for a massive boost (also applicable to serial ids, though).

The seqscan option being a binary on/off is a bit of a misnomer; what it actually does is set the cost associated with a seqscan to be astronomically high so that other options will be preferred over it. An index on age when you're looking up by name won't be used whether or not seqscan is enabled.

Some comments and corrections:

* Use uuidv7 not uuid in general (typically v4)

* in addition to minimizing locked records, make sure your locks are ordered deterministically across all queries (eg by id asc, always) or you’ll deadlock (but postgres has a really good deadlock detector so you’ll more likely just error out if you’re lucky)

* always use explain (generic_plan) to be able to a) copy-and-paste your queries with placeholders for parameters as-is, b) see how your query will actually be optimized when Postgres doesn’t have visibility into the specific parameter values

* use set seqscan = off when testing your query plans esp when tables are empty or nearly so so you can see if indexes will be used when seq scans become less cheap

* everyone defaults to btree indexes which are heavy and increase index bloat. Consider using a hash index instead if you just need to look up by column/id but not sort or get values greater/lesser than a param. You can’t create unique hash indexes but you can create exclude using hash constraints for the same effect (except no multicolumn unique index support)

* learn about GIN (and GIST) indexes. They can speed up common queries without needing new syntax, something people coming from MySQL might not expect to be possible; i.e. you can use them to speed up Plain Jane like ‘%foo%’ queries without switching to FTS.

There’s no need to get all complicated and fancy or introduce more dependencies. For most people, a cron job calling pg_dump_all piped to zstd and copying the output to s3/ftp/whatever is plenty good enough.

Obviously past a certain point carting around full backups becomes time/dollar prohibitive, but this can take you very far.

It runs easily on a vps at your scale, even the same vps serving your app. That used to mean having a modicum of sysadmin knowhow but it’s straightforward these days, especially if you just use a premade docker file.

So 3.6 Flash is a somewhat of an admission that Google miscalculated by charging 3-5x for 3.5 Flash what it did for 3.0 Flash (3x input and output costs plus large token inefficiency changes) despite only modest improvements?

3.5 Flash Lite is only a hair cheaper than 3.0 Flash, but I think 3.0 Flash is a massively more capable model?

If you mean the new windows 11 emoji, they are some of the ugliest design with I’ve ever seen, and I say this as someone that appreciated the previous windows 10 style despite the hate it got.

It’s insane because it literally breaks everything published with the old emoji to date.

Why did the cat get an evil smirk? It seems like adding rain to the umbrella potentially breaks the existing meaning? What’s the emoji that went from three colored circles on a stick to three completely ambiguous widgets?

> Emoji with the darkest skin tones can be difficult to see in dark mode — a problem for anyone who sends or receives them. That’s why we built an AI-powered contrast tool that analyzes each emoji at the pixel level, flags when the contrast ratio is too low and suggests high-contrast solutions that are implemented by designers

What part of this is AI? What part of that couldn’t have been done without it? Do they just mean they vibecoded it?

> Apple can make them smaller and put them on the device

Someone can, but Apple has essentially admitted defeat and handed the reigns over to Google.

Author also discussing it on their Twitter [0]. I’m reserving judgement for now, but this has lots of potential usecases even apart from replacing postgres.

[0]: https://x.com/glcst/status/2077759127682486561?s=46

> This is great

Absolutely spectacular, indeed…

> and long overdue for outdated devices!

I would pause here and reflect. There is zero chance this performs better (speed, latency, memory, battery life, etc) than Safari on iPhones since the latter has been heavily optimized over the decades for this exact use case. Older devices not even getting iOS updates are likely the devices that will work worst with this software in its current form.

Of course if Apple had allowed third party browser engines from the start, it would likely be a much more even playing field!

Why do I have a sneaky feeling Safari for iOS isn’t fully written in a fully memory safe language?