Bluesky knows full well it's not operating a real federated service

Better tell that to Blacksky, Eurosky, etc.

The vast majority of people stay on the primary PDS, relay, etc namely because Bluesky hasn't proven itself to be some evil overlord pursuing insidious goals. If that were to ever occur, people would just migrate. Unlike with ActivityPub (Mastodon), ATProto allows for true migration. Your content isn't tied and linked to a specific server - it's more like a URL on an arbitrary domain, and you can just change the "domain" (the PDS). Everything is timestamped and cryptographically signed, so if you download a backup of your content, you can just reupload it somewhere else and it continues to remain linked into the whole ecosystem.

More to the point, primary Bluesky servers have gone down and third parties like Blacksky remained operational, very much demonstrating that the network is federated.

Also, re: this from the header:

" and by the end of October last year, it had reportedly seen a 40% drop in daily mobile active users over the past 12 months."

... is cherry picking. If you graph users, you'll see that - like most sites - new users tend to arrive in big "spikes", triggered either by events at other social media platforms, or major news cycles (such as elections). Most new users to a site are not "sticky". Some drop off in days, some in weeks, some in months, etc, but this slowly levels out, and the rest are "sticky". With Bluesky, usually half or so of new users stick around, which is an unusually high percentage. If you measure from a new-arrivals spike, of course you see a "dropoff", but you see that for any site. The question is, how is the long-term trend of users that stick around? If you cancel out the spike pattern, Bluesky has a long-term population of around 600k daily posters / 1M daily likers / 300k daily followers.

What you can say is there haven't been any big new user spikes since late 2024 / early 2025. That said, there kinda was some serious news going on in late 2024 / early 2025....

The extra fingers will come in handy when climbing.

Imagine being like that and spending that much of your day thinking about other people's genitals.

1) You said "AI hype is dying", not "AI companies aren't profitable" Undergoing an insane sustained ~10x/yr exponential growth trajectory over 3 1/2 years , and that rate even accelerating now, is in no way "dying"

2) It is absolutely not normal for companies undergoing rapid rates of growth, let alone such an insane rate, to be profitable. Scaleup generally means you lose money hand-over-fist, as scaleup is extremely expensive (no less so in this field!). And yet:

3) Anthropic may actually pull it off this quarter.

Their margins are like 40%. That's all users combined, not just paying users. Inference is cheap to serve; compare what the closed commercial operators like Anthropic charge vs. what the open source models (an actually competitive for-profit marketplace) charge, for models of equivalent size. The closed models rob you blind. But people pay it because their models are the best.

(BTW, the main thing that's driving it now isn't random people asking questions on the website or in an app in their phone. It's software developers).

Zoophiles with a thing for large cats must get so frustrated by such ads.

Remember, the AI hype is dying

Better tell that to Anthropic's annualized revenue growth.

Dec 2022: $10M
Dec 2023: $100M (10x in 12 months)
Jan 2025: $1B (10x in 13 months)
May 2025: $3B
Oct 2025: $7B
Dec 2025: $9B
Feb 2026: $14B
Mar 2026: $19B
Apr 2026: $30B
May 2026: $47B

Here's the crazy thing: initially, SpaceX / X.AI didn't acquire the servers. Tesla had the contract for them, and then Tesla just gave those rights to X.AI. Tesla is a public company, while X.AI was Musk's private company, founded so that he wouldn't have to share any AI profits with Tesla.

Now, Musk's excuse was that Tesla's datacentre wasn't ready yet, and it cost Tesla nothing. But of course, there was a massive backlog on servers; the rights to early delivery of servers was incredible value. It let X.AI jump to the head of the line. Tesla could have sold those servers or the rights to them at a huge markup.

If you don't like your honey having been regurgitated, there are alternatives for getting honey...

Drones are specifically the male bees. Most bees are not drones.

That's not what I said. I said that the word "drone", as in a mindless unthinking being, is derived from drones, as in male bees. I did not say "all bees are drone bees".

And an individual bee has limited memory. They even forget which hive they are from after a while if they don't return to it.

You are confusing "forgetting" with "disruption".

Bees have both a geospatial "mental map" (based on landmarks, the sun, etc) and a chemical fingerprint (they recognize their nestmates' smell). Concerning their geospatial memory, not only is it not poor, the main problem with it is that it's too stubborn. If you move a hive 20 meters away, the bees will fly back to the same empty location where their hive used to be and wait there. They don't adapt well to change because they have a long-term memory of "the hive was here".

If a beekeeper wants to move a hive, they have to trigger an "orientation flight" to get the bees to learn the new location (this typically involves locking them inside their hive for several days to disrupt their routine). During an orientation flight, the bees will learn the new hive location, and then they'll subbornly remember that location long-term, even if you move the hive again.

As for recognizing their nestmates, this is again based on smell. A bee being isolated for days or weeks will still be recognized by guard bees at the entrance and welcomed in. However, guards will sometimes let in bees that don't belong to that hive as well, if e.g. they're passive and laden with pollen and nectar; they haven't "forgotten" their scent, they're just "forgiving" of mistakes if there's a reward to be had (bees sometimes make navigation errors, esp. if all nest boxes are similar in shape/colour or due to wind, and enter the wrong hive)

I'll repeat: bees do NOT have a short memory. This is a myth. It's not true. The very example you gave is actually an example of bee memory being too rigid.

It's IMHO amazingly impressive how dense information can be stored within neural networks. Even a comparably tiny LLM can store more information than the human brain, despite the brain's theoretical storage being far higher due to its vast number of connections (ANNs are better at information density, we're better at learning from limited datasets). The tiny LLM will crush humans at a quiz in virtually anything except said human's particular areas of expertise. Storing information as a superposition of states across a large number of neurons and connections (whether we're talking artificial or biological) is an immensely space-efficient way to do so, and the human mind is nowhere near the limits of information storage capability.

There is no technical reason why a given organism, such as a bee, could not achieve far denser information representations in order to be able to do more with its limited neural capacity (though there are always tradeoffs). One of the reasons that ANNs learn slower-but-denser is the use of a very low learning rate with a very large amount of data that covers the same topic from many different angles, giving the weights ample time to explore different possible circuits in parallel and seeing which ones predict reality the best ("learn everything all at once" vs. "learn this thing NOW"). Bees aren't tasked with learning anywhere nearly as diverse things as a human is and spend all day doing the same basic job (the same information "from different angles"), so it seems quite possible that their greater "information specialization" as they go about their day may be able to lead to denser representations of said information.

BTW, at risk of a tangent (your comment about non-neuron cells playing roles), it's been really interesting to me seeing how a key difference between artificial and biological learning has been clearing up. In biological neural networks, weight cannot flip sign (Dale's Principle). In the general case, a neuron is either excitatory or inhibitory (usually a small number of inhibitory neurons per cluster of excitatory neurons); it can't change from one to the other even if learning would favour that. At a first glance, that would seem to cripple learning capability (and definitely does if you implement that in ANNs). But what appears to actually happen in biological neural networks is a sort of horizontal learning, co-dependent synaptic plasticity, between excitatory and inhibitory neurons. Instead of merely weakening an excitatory connection down to zero and then being able to go no further, learning simultaneously weakens the excitatory connections and strengthens the inhibitory connections. The excitatory neurons are the primary drivers of information storage and processing, but the inhibitory neurons adjust the baseline to give them the flexibility to express negative net activations as needed.

It is a myth that individual bees only retain information for half an hour. Depending on the memory at hand, bee memories can last days, weeks, or even the remainder of their foraging life. They have to remember things, because the timeframes a hive operates on are much longer than half an hour, including night time and being kept inside by inclement weather for days or even weeks at at time. Individual bees also learn much more than can be conveyed through waggle dances, such as what colours and shapes of flowers are yielding best in a given area at what time of day (bee learning is essential to them being able to function as generalists, able to handle any mix of plants at any latitude).

Also, the hive doesn't just blindly accept whatever any bee says. Each bee functions as an individual in a society. When a bee waggles in the "town square" (on the comb), other bees gather around to "listen" (detecting oscillating shifts in the electric field plus tactile contact and sound). But whether a bee actually decides to make use of that information depends on whether they're having good or bad foraging success. Only a small fraction of bees on average (usually a single-digit percentage of watchers) will decide to make use of the information. And if another bee "disagrees" with a waggle dance - for example, if they've been there and found nothing, or worse, found dead bees, predators or a rival hive), they can make a counter-buzz to argue against it. The arguments can get quite "heated", with many bees taking part.

We think of bees as mindless drones (literally, we took the very word!), but they're all individuals each acting on their own. There are simply various rallying factors that keep them together (for example, the scent of the queen, the desire to live in a warm hive, etc). The information communicated within a hive is limited; bees overwhelmingly rely on their own mind and memory, and perform their tasks as individuals.

I mean, bees attack small hive beetles by building prisons of propolis around them, that's something :)

Meanwhile, X.AI has asked the EPA for permission to release up to 100 million hyperfecund pesticide-resistant male mosquitoes in California and New York over two years to own the libs.

*head bangs in approval*

Oh, forgot to link the dry density for you: here you go. 341kcal/100g. Aka 3,41kcal/g.

Which, like I said, should be obvious, since they're almost entirely carbs (~4kcal/g) and protein (~4kcal/g), and they're, as noted, dry (12-16% moisture). It would be quite the trick indeed to get something that is dry and and is almost entirely comprised of things that are 4kcal/g to be 1,38kcal/g! ;)