The nitrogen fixing organelle is the very subject of the OP's article.

> These independent lines of evidence leave little doubt that UCYN-A has surpassed the role of a symbiont. And while mitochondria and chloroplasts evolved billions of years ago, the nitroplast appears to have evolved about 100 million years ago, providing scientists with a new, more recent perspective on organellogenesis.

The new evidence is not that this symbiosis is new, but that it's functionally an early stage organelle dependent on some proteins from the host and in the process of merging.

> Much of the popular deep learning stuff is motivated by brain research

very little of it, except in the most vague abstract. Previous generations of neural network research was more attuned to biological plausibility.

the key algorithms: stochastic gradient descent and related with backprop gradients, and self-attention on a long context & softmax, are totally not biologically plausible at all.

> they will get exponentially better as they get more complex

they get logarithmically better with complexity, not exponentially

most likely a fat finger error, or miscomputation which can be easy with weird 10^-7 based prices.

for money laundering there are better options.

I have a different conspiracy theory. The actual full LLM is still working quite well, and un-lobotomized GPT-4V is still the leader.

But running the full LLM is costing them so much in compute and electrical power that they are inserting instructions and hacks to terminate the session and production of new tokens (which costs compute) to lower the average cost per subscriber. This is to make the financials look good right before the IPO that everyone knows is coming. The employees are going along with it as they might get a spectacular payout (like an average drone might be $10 million or more)

Actually of the "OG" papers, Einsteins on SR gets more citations than most old ones, particularly the English translation.

Special relativity does not explain ferromagnets, though otherwise electromagnetic theory was primary.

Einstein completed the one thing that Maxwell didn't get in classical electrodynamics, and then took the extraordinary leap that the recent Maxwell theory was fully correct as is and the successful and incontrovertible Newton needed modification---the opposite of every one else working on it at the time.

One cool point, Einstein and de Haas conducted an experiment that showed that the angular momentum in ferromagnetism (now known as electron spin) is real mechanical angular momentum, or correctly can be converted with usual conservation laws.

As scientists understand the natural world better, there is less that they are wrong about, and hence less disruption where previous citations drop away because they were profoundly mistaken.

Quantum mechanics and relativity and DNA and evolution are not going to go away.

There are many more people working on similar tasks compared to 150 years ago and so there is a more continuous stream of research, and because of that there are fewer large gaps to leap over with a singular disruptive paper, as there was already much more exploration. Not as much drama but better for society.

> Back when solar panels were expensive, tracking was definitely worth it. Now, the solar panels themselves are one of the less expensive parts, and it may well be more effective to just accept the lower power per panel and simply use more panels.

Except that tracking also helps produce power at early and especially late afternoon hours which is much more economically valuable. And the electronics aren't so cheap. Battery storage is still valuable but much more expensive than a panel which can still track to make power at 4:30 pm. Save the battery for 7pm.

> you just install additional storage and use both.

LFP batteries coming online now have superb cyclic properties (needed for energy storage). All batteries will have calendar aging decay in capacity.

In some future state it will be economical to replace only the cells and keeping the other equipment (electronics, cabinets, cooling, breakers), and there will be a robust market in used batteries for recycling. They'll put in a new generation which will last even longer.

> So you just calculate the capacity you need, pour concrete pads, place the batteries and power conditioning electronics, and have a structure around it.

OK, but nickel is pretty expensive compared to the materials in a lithium iron phosphate or sodium ion/Prussian White battery. That sodium-ion will probably be the mainstay of bulk energy storage and will have supply chain constraints, if any, different from lithium ion needed for vehicles.

Higher energy density means fewer cabinets, less cooling, smaller concrete pads, less wiring---all significant costs. Probably nickel-iron is sufficiently low energy density to increase costs, otherwise there'd be some commercial deployments. Today I think most new battery storage will be LFP and then sodium-ion when it is commercially mature.

The wording of the post suggests some sort of internal deception/treason, perhaps to steal assets/IP/people into an entity which Altman and friends owns and can IPO for themselves.

Not uncommon in Chinese subsidiaries owned by foreigners.

They took from Lenin the anti-democratic tyranny for class purposes, but chose a different class.

antarctica is Big

Billionaire Envy makes billionaire CEOs do idiotic things.

Mutliple examples:

1) Steve Ballmer was infuriated by Apple and Google, so he ordered Microsoft to try to make direct competitors (Windows Phone and Bing) which never really worked. He had a toxic workplace culture.

2) Mark Zuckerberg was infuriated by Apple owning the end-user device hardware and cutting off his ad money because of Apple privacy settings. So he went on this massive crusade to promote the Metaverse, not because he wanted the metaverse but because he wanted control of the end-user hardware platform and its raw social use data. He hoped if Metaverse hardware were as popular as phones he would be the #1.

3) Elon Musk is upset that OpenAI is succeeding and Sam Altman is getting all this attention. So he orders an OpenAI clone the way he thinks it should be. There is going to be no "understanding of the universe" from people experimenting with pytorch algorithms. I mean I like experimenting with pytorch algorithms but that's all it is.

The first two cases, Microsoft and Facebook started to recover once they re-concentrated on what they were naturally good at (business software & OS now on Azure, and social media respectively) rather than managing by envy.

In the third case, it's not clear what Musk is actually really good at any more. His primary talents used to be being able to raise money from investors for long-shot technical developments---and that's not an insignfiicant skill. For early SpaceX and Tesla that was really essential. But now his declining emotional maturity and publicly obvious management failures means that investors will stop believing in his promises.

I think he may be suffering from the effects of years of amphetamine abuse. He was always an egotistical prick but the amphetamines can amplify this.

> Gonna be real funny when we have nobody making any money to spend.

They will soon see people, correctly, as nothing but drains on resources and their wealth. Let's suppose now 5% of working age people are incapable of any useful work from profound disability. What happens when it's 80%, which will be the case once physical robots are embodied with good enough AI?

Like 85% of humans are as useful as the deranged hobo swearing under the bridge, and will never ever be able to offer any labor competitive with the price of robot AI slaves? 5% will be enough for all prostitution, 5% for super high end intellectual activities and the ruling elite, 5% for entertainment.