I start with 18%, I read some really great comments by other pros. Some ideas that came to my mind:
I read a good argument that they may not need testing since they already tested their nuclear bombs in the past and have something that works. In the last test in 2017 the blast was estimated to be around 250 kilotons. How much is it? For the purpose of calibrating our mental models we can use:https://nuclearsecrecy.com/nukemap/ If this will not be too disturbing for you, you can see the blast area for such a nuclear detonation in some cities which you know from your experience (have a mental representation of their size). Examples:
Washington DC: https://nuclearsecrecy.com/nukemap/?&kt=250&lat=38.895037&lng=-77.036543&airburst=0&hob_ft=0&psi=20,5,1&zm=12
London: https://nuclearsecrecy.com/nukemap/?&kt=250&lat=51.5073219&lng=-0.1276474&airburst=0&hob_ft=0&psi=20,5,1&zm=11
Seoul: https://nuclearsecrecy.com/nukemap/?&kt=250&lat=37.5666791&lng=126.9782914&airburst=0&hob_ft=0&psi=20,5,1&zm=12
Warsaw: https://nuclearsecrecy.com/nukemap/?&kt=250&lat=52.231958&lng=21.006725&airburst=0&hob_ft=0&psi=20,5,1&zm=11
Can we look at North Korean Ballistic Missiles payload capabilities to say something about warheads they intend to carry on them? 38North have estimated payload of North Korean Hwasong-16 ballistic missile payload to be about 2,500 to 3,000 kg,
How many kilotons should we expect per kilogram? This article is great in answering that:
https://blog.nuclearsecrecy.com/2013/12/23/kilotons-per-kilogram/
"all US weapons currently in the stockpile end up, in the 0.1-3 kt/kg range, especially the 2-3 kt/kg range"
So only if we use the absolutely lowest values of this ratio and the lower end of the payload estimate, we will get 250 kilotons, but as some of the nuclear weapons in the US nuclear stockpile were developed in the late 1960s and 1970s https://nuclearweaponarchive.org/Usa/Weapons/Allbombs.html and with the possibility of Russia sharing their nuclear know how along with some other technical expertise with North Korea in exchange for weapons and soldiers, I am not an expert, but I think we can expect that they will be able to achieve better kiloton to kg ratio than 0.1 kiloton/kg. So I think they aim for something bigger than 250 kilotons, especially now, when they may have access to Russian expertise. I may be wrong, since my understanding is limited, but I think they are now likely working on some new design with higher yield, rather than only making more of the current ones.
Another point: other pros were right to point out that the warnings about the test keep repeating and nothing happens. However, as I read some of the warnings from last two years, one (IIRC it was from 2022) was saying explicitly that the test is not imminent but that North Korea is working on the technology and the test is likely in the future, and the one from 2023 said about the test "next year", So maybe some of these were not necessary false alarms, or at least we do not know yet.
One more point that comes to my mind: I love the idea that we would be able to see preparations for the test on the satellite images so if we cannot see them now it will likely not happen in the next two or three months, but I keep thinking - if North Korea is determined to conduct the test (they may want to test improvements, especially if there is/will be know-how transfer from Russia - I assume that they will want to test it, that testing is needed for new designs or at least they will want it to be sure) they may want to try hiding the preparations for the test and avoid international pressure not to do it (I would expect pressure if everyone could read about the preparations). Is it possible? It would be hard, but India was successful. Also, the problem with any open-source intelligence methods which results are published online is that the people whom we observe can access these analyses as well and they will know what to hide from us. As a counter-argument, it is one thing to play this tactic with the West, but they also have to think about their relationship with China, which is against North Korean nuclear tests.
I think the probability slightly increases over time because this is a process that is progressing, and now they may have Russian help. At the same time, I don't think they will do the test that fast. Of course, I may underestimate the importance of political factors here, like nuclear diplomacy with the new US President for example - if Kim Jong Un decide that it may give him political benefits, then testing even the same nuclear bomb as before would be reasonable.
I really don't think anything will happen in such a short time left till this question resolution date, as @PeterStamp mentioned here they still have job postings for AI related jobs in Beijing, even if the AI Research Lab is not mentioned. For example: https://jobs.careers.microsoft.com/global/en/share/1782091/?utm_source=Job Share&utm_campaign=Copy-job-share
At the same time, the trends do not look well for the prospects of Microsoft keeping the AI Research Lab in China for the longer term.Recently, we have seen the news that "Llama 13B has reportedly been adapted for a Chinese military intelligence tool known as ChatBIT": https://www.pcmag.com/news/chinese-researchers-make-military-ai-using-meta-llama-13b
Soon after that news, Meta changed their policy, and announced "that it would allow US national security agencies and defense contractors to use its open-source artificial intelligence model, Llama.": https://www.theguardian.com/technology/2024/nov/05/meta-allows-national-security-defense-contractors-use-llama-ai
https://www.aljazeera.com/economy/2024/10/29/us-finalises-restrictions-on-ai-semiconductor-investments-in-china
Also:
Source: https://www.aljazeera.com/economy/2024/10/29/us-finalises-restrictions-on-ai-semiconductor-investments-in-china
Also:
source: https://www.washingtonpost.com/technology/2024/11/14/us-china-ai-tech-export-controls/
Also:
source: https://www.brookings.edu/articles/ai-policy-directions-in-the-new-trump-administration/