BigLadKarlLiebknecht [he/him, comrade/them]

  • 3 Posts
  • 13 Comments
Joined 4 years ago
cake
Cake day: November 6th, 2020

help-circle


  • Oooh I didn’t know he posted to Reddit. I used to follow him on Twitter, but he quit or got banned. His account is a goldmine for War Nerds…

    From 6 months ago:

    I am more worried now about the potential for nuclear war than perhaps ever.

    :yea:

    I think that a first strike might be possible on the US but not on Russia. Let’s assume an intelligence failure and that some Russian forces are able to generate to alert status in a covert manner. I think that it’s possible to initially do a limited first strike on both the East and West coast using stealth Kh-102 nuclear cruise missiles. These will not be detected until there is a NUDET. You could take out all the SSPARS (early warning radars), the White House, the Pentagon, Site R, and the SSBN bases at Kings Bay and Bangor immediately prior to a massive ICBM/SLBM strike. So now, the President and top military leadership in DC are dead. At this point, we’ll still have DSP/SBIRs/etc. picking up the missile plumes and USSTRATCOM will be trying to reach out to surviving leadership. Soon, the weapons begin to detonate on all NAOCs and TACAMOs on the ground and to take out the ground entry points for these systems. Yes, we still have the alert- and mod-alert SSBNs on patrol, but this scenario might be doable. Or might not. Hard to say.

    What scares me the most is a preemptive attack from Russian stealth Kh-102 ALCMs. I do not think that we would ever see that coming, and yes, there are ways to fool us that the ALCM carriers are spoofing ADS-B as airliners/etc.

    :putin-wink:

    I’m former intelligence community (Department of Energy and Defense Intelligence Agency). I have high confidence in the weapon systems reliability of Russian nuclear weapons and missiles. What is happening in Ukraine is not transferrable to the Russian Strategic Rocket Forces/etc.

    Almost refreshing to see someone with professional knowledge of these things refute the idiocy of “but lol the nukes won’t work”. Obviously it’s not refreshing as it means that 90-95% of the US population would be dead in 18 months, but still. There’s at least some people who take this seriously, somewhere in the empire.



  • Aside from the horrors of AI art, I wonder about the cognitive effects of more and more AI being present in our daily lives. For instance, GitHub CoPilot being a thing (in addition to a sneaky way to enclose the commons) means that engineers not thinking things from effectively copy pasted code through is entirely more likely. The way that Google Docs autocompletes text for me - I can’t fucking stand that. I can sense it molding my thought patterns already.

    It makes me think of getting a calculator as a kid, and my math skills deteriorating rapidly not long after. Or this paper, showing that navigation tools like Google Maps harms spatial awareness:

    The analysis of long-term effects through structural equation modeling showed that spatial aptitudes and accumulated experience of tool use independently affect wayfinding and spatial orientation and that the negative effects of accumulated experience were larger than the positive effects of spatial aptitudes.

    Not to sound like old-man-yelling-at-cloud, but when it comes to cognitive offloading I can’t help but feel that doing things like leaning on a bunch of shonky linear algebra for human expression and reasoning is going to have horrendous systemic effects that we can’t predict.




  • So one thing I’m trying to understand with the base and the superstructure is with respect to tech startups and the approach to engineering that they take. Basically:

    1. Start ups are in a constant cycle of chasing venture capital funding, needing to show constant growth of users/customers at fairly regular intervals
    2. This leads to a myopic focus on quality in engineering, as it is (incorrectly) deemed as a hindrance to delivering new features that would allow for increased growth. (In reality, the initial cost of focusing quality is quickly surpassed by the drag of poor code and technical decision making forever increasing as it is left unaddressed, but I digress).
    3. Further to that, early employees of the company are strongly incentivized in the short term, by relatively large stock option grants. These grants are generally exercisable after 1 year, and most folks will move on to another company after 2 years. As such, there is no incentive for them to fix the engineering mess they have created if the company is growing - someone else will be employed at a later date to solve that problem, for significantly less equity in the company. These later employees are effectively doing the work of repairing the poor work that the early employees left in their wake. A technical debt Ponzi scheme, if you will.

    So given all that - where is the base and the superstructure? Is it all base, as it is a combination of means and relations of production? Is tech bro “fuck quality” culture the superstructure here?

    Sorry if this makes no sense or is baby brained, I’ve been stewing on this idea of tech debt being a Ponzi scheme for a while and I want to be able to talk about it correctly from a materialist perspective. And I’m not the brightest at this stuff :comfy: