#DHL and #AnPost seem to have lost my workstation. Sent it from Munich to Ireland. AnPost says they are returning it because label had an issue. DHL thinks it’s still in Germany. If it’s returned to Germany I can’t get it shipped back because I don’t live in Germany anymore.
We cannot change our past, but we can look critically at our present and work to shape a better future.
I think we have the mathematical tools to argue for the efficacy and outcomes of interventions in these situations, and the people conducting the experiments are well-equipped to use them.
The demand should be for rigorous analysis, not “you must use this approach”. We are not in primary school.
Have we some “explanation” or “reasoning” as to how instabilities are caused?
Could you explain how info is propagated across layers in your approach vs densenet in some more details over the conversation?
How do your approaches scale when we need to do RL on say LLMs?
2/2 (for now) Thanks again!
Beautiful work, thank you for the overview!
I haven’t done RL work in half a decade — I thought delayed updates with exp moving average of target networks helped deal with these instabilities. I assume it’s not enough given your paper.
1/N
If this is indeed it, then we have good reasons to be optimistic, no?
Maybe I am just seeing things, but it looks like we are past the upper inflection point of the sigmoid.
Do we have sims of different scenarios?
Don’t worry, scientists thought of that as well. There is a whole science of communicating with future people that don’t speak our language.
From hostile architecture, to death symbols.
The very active isotopes of fast-breeder-reactor waste decay over 300 years and need containment.
The very inactive isotopes take thousands of years and are indistinguishable from background radiation. They pose a problem only when ingested or inhaled.
So personally, if the choice was living close to a nuclear reactor or storage facility vs living next to a battery factory or a solar factory, I’d take my chances with the former.
Your risk of developing cancer is hundreds of times higher tanning in the sun on a high UV day than living close to a containment facility.
For fast breeder reactors, the half-life of the more active isotopes is 300 years, then it gaps to thousands of years for the longer dated ones.
The longer dated isotopes are so inactive they might as well be background radiation. In fact, for them to do damage they need to be ingested or inhaled.
So the important thing to know is that longer lived isotopes are less dangerous than shorter dated isotopes. Inversely proportional actually [1].
[1] en.wikipedia.org/wiki/Technet...
The US does not recycle nuclear waste or use breeder reactors.
It also implies not being dogmatic about particular solutions or fear mongering about things we don’t actually understand and shoving our head in the ground like ostriches.
It is in our best interest to think in 5 years, in 10, in 20, and in 50.
That implies reducing waste, improving efficiency, reducing reliance on complex chains, funding sciences, math, physics, material science, chemistry, biochemistry, biology, computational sciences and so on.
7/8
But because the constant supply chain issues reverberated across markets, resulted in difficult to predict outcomes, and threatened stability.
It follows then that the right thing to do to protect our society is reduce the fragility of the system.
6/8
As much as I hate to say this, the cynic in me believes that we did not make the decision to go off oil not because of climate change leading animals to extinction and threatening ordinary population.
5/N
It is integral to identify all the ways that said supply chains can go wrong, and ameliorate the situation. The oil crises are exemplars of this.
4/N
of modern society.
It follows that any country that aims to think in decades about its own sovereignty should seek to minimize the effects of disruption on its supply chains.
Betting it all on any one particular technology is naive if not outright stupid, and when evaluating solutions, 3/N
Just as we don’t know the future in sciences we don’t know the future in terms of geopolitics.
Covid, the Russian invasion of Ukraine, the aggression of CCP towards Taiwan, the Israeli and US war on Iran, the US’ unnecessary and myopic tarrifs exposed the fragility of the complex supply chains 2/N
When we as the broader society fund sciences, we don’t put all our eggs in one basket. We fund as wide as we can, with some areas a little more or less.
We don’t know what the future holds, and thus we try to explore as much as possible.
The responsible approach is to plan in decades. 1/N
Nothing you have said around here is of substance.
Just assertions that can be dismissed with the same effort you spent writing them, and a “gotcha” that lost you any and all credibility.
Of course your argument is also a whole lot of bullshit because you are not arguing in good faith here.
We seal the uranium and all fission products in lead contains that can withstand collisions with cars and don’t allow radiation to leak out.
After breeder reactor processing, the waste is primarily fission products that decay to safe levels in 300-500 years, leaving only seven very long-lived isotopes (200,000+ year half-lives) at very low activity levels.
www.nature.com/articles/s41...
Nuclear is a far more stable and predictable source too. Contrast this with the issues that arise in dealing with over and under production, potential cascades due to lots of smaller power plants, limits or failures of energy storage, and all your “simplicity” arguments go out the window.
Standard nuclear reactors are already a solved problem and we don’t need to care about complex supply chains or weather systems or integration of lots of moving pieces, nor do we care about batteries.
Lmao.
I am not even opposed to solar and renewables, I just find it fascinating how lay people got convinced that nuclear is not safe when we have all the data and numbers, when the tech pushes forward even without all the funding given to renewables, and that they are without their own sets of problems.
bsky.app/profile/quee...