The Danger of Excessive Regulations
If it moves salute it. If it doesn’t move, paint it.
Here’s an article from the US Naval Institute that highlights an important issue. There is a natural limit to the number of rules and regulations an organization can sustain. At some tipping point, there are too many when they are in conflict with one another. This requires the individual in the organization to elect a path of compliance which is also a path of non-compliance. The credibility of the whole set-up degrades and the rules become less binding, generally, once people understand that non-compliance is not only okay, but expected.
‘Today, service members may find themselves in situations in which they disregard or violate a regulation, order, or directive.1 Because of an overwhelming number of requirements, they sometimes must choose which ones to follow—or may not realize there is one they should be following. This not only normalizes deviation from the requirement that is unmet, but also inculcates a culture in which it is common to “work around” requirements if they are perceived as unnecessary or extraneous. In a military that prides itself on discipline and standards, this could be catastrophic to good order and discipline and degrade warfighting capabilities.’
Dictator on Day One? Trump’s grand plans for the federal bureaucracy
Ironically, Trump would by stymied in his ability to cut the bureaucracy by … bureaucracy. Schedule F would convert tens of thousands of civil service positions into political appointments if these positions involved making policy. He would be constrained no more by the Liliputian ties of the so-called Deep State.
But first, he has to get a bunch of people to fill out some forms.
In a further irony, attempts to protect these civil servants has itself been hindered by the failure to pass legislation, making them vulnerable by the bureaucratic intervention to date on their behalf.
‘Biden’s upcoming inauguration and expected repeal of Schedule F undoubtedly slowed compliance with the order (and likely explains the apparent inaction of the vast majority of other federal agencies). But the haphazard agency submissions also reflect the enormously time-consuming administrative task of reviewing tens of thousands of positions. These lengthy bureaucratic hurdles would again have to be cleared if Schedule F were reinstated, which would require another months-long period for agencies to have sufficient time to finalise affected staff. This would delay the implementation of a Day One order by months, even in a best case scenario for a second Trump administration.’
Product Managers used to be creators. Now they are mostly bureaucrats
Bureaucracy comes for all of us sooner or later. In a tech company, it comes with scale, evidently.
‘At one point, Product Managers did what they did because they thought of themselves as creators. Sure, they did the impact assessments, the documentation, the product roadmaps, and the boring JIRA tickets, but they also revelled in doing things sometimes because it was cool, serendipitous, or simply beautiful.
‘Today, they pride themselves on being excellent bureaucrats.‘
We have entered the Compliance Wars. It is the tenth level of hell.
‘Republicans believe DEI is so deeply ingrained in higher education that simply banning it is not enough. And so they’ve enacted a new layer of oversight and bureaucracy to monitor administrators’ and professors’ actions, according to a Chronicle analysis.’
Technological Challenges for Regulatory Thresholds of AI Compute
Both the EU and the US impose thresholds for determining what constitutes a “systemic risk” in AI. In the EU it is 10^25 floating point operations per second (FLOPS). In the US, we’re one louder: 10^26 FLOPS. In the US, there is an additional constraint. The computations must take place in a single data center capable of 10^20 FLOPS.
However, new Transformers architectures and alternative approaches that promote the use of multiple, potentially smaller, LLMs mean that it’s possible to develop new models that never come close to the threshold. In addition, distributed computing suggests the possibility that the US compute threshold doesn’t come into play if no single node in a synthetic data center made up of disparate compute nodes hits the 10^20 FLOPs threshold.
‘Hence careful attention will need to be paid both to: (1) the rapidly emerging technological innovations around computing clusters and techniques and model architectures which impact compute requirements for training frontier AI models, as well as (2) the evolving regulatory thresholds attempting to delineate computing cluster and training compute thresholds for such AI models.’
Congressional Review Act: from sleepy law to potential regulatory wrecking ball
The Trump administration’s effective use of the Congressional Review Act was a big surprise in the early days of his presidency. The prior administration should have annealed its executive orders but neglected to do so, perhaps unaware of what had been an obscure and little used piece of legislation. The Biden administration isn’t going to make the same mistake. This explains the recent blizzard of regulatory activity.
‘As administrations try to withstand the potential for subsequent Congresses to "look back" at a specific set of the previous administration's regulations, the threat of the Congressional Review Act activity creates an incentive in the potential or real final year of a presidential administration to finalize as many regulations as possible to avoid the lookback period.’