cameronfen

You are right it's looking more like a dud based on what has come out the last couple of days but there have been some replications that demonstrate partial success, Superconducting at 110K and ambient pressure. See Wikipedia's list of replications. https://en.wikipedia.org/wiki/LK-99

Yea for sure. I think most physicists agree with that. It’s looking more likely than not that this is not a room temperature superconductor, but I think it seems like scientists are optimistic that this paper is not junk and pushes the frontier closer towards that goal.

3i a London-based P/E firm owns Action the DG of Europe. Action has no dollar store competition or even much mass-market competition like Walmart or Costco. Action is 60% of NAV and market cap. Growing revenue at 33% (22% same-store sales), income maybe 60%. I'll post a write-up at some point. The thread is here though atm:

As I’m sure you know, if you have robust enough growth this could change the fair value multiple.

From the wikipedia page of LK99, you do have two replication attempts that have been successful at replicating some aspects of superconductivity. One at temperatures of below 110 Kelvin and then a sharp drop in resistance at 230-250K (as mentioned above) and one replication showing diagmagentism, but no word on resistance yet. Two other labs show null results but this might not be surprising considering the success rate of the original LK99 paper was only 10%. It's crazy all the labs are in China and India. Maybe that's where the world is heading.

The material is actually pretty easy to make somewhat poorly, but the replications done seem somewhat promising. But I agree that to make these things at the purity and quality that is necessary for mass production is difficult. I think the LK99 paper even stated that they could only get superconducting on 10% of the samples they made. I don't think it will take years to reproduce the findings though. After all the betting markets contract for a no success is if no one can reproduce by 12/31/2023. So I imagine the experts consider that reproduction shouldn't take close to that long. It seems like you know the physics details better than me, but my guess is the simulations weren't as heavy handed as your friend. a) because they don't have an interest in the result either way and b)because they probably have better skills and familiarity with this field that only comes with many years of research.

Andrew was quick to point out, other superconductors also work at 110K but not at atmospheric pressure iirc. Not a scientist.

Here is a tweet from someone who knows way more than me that there is something here:

Replications are already in progress and some are complete see the Wikipedia page: https://en.m.wikipedia.org/wiki/LK-99 Most of the experimental results have not replicated but theoretical replications indicate good properties.

This seems like the moderate position. I don't know if it is correct: It is a superconductor at atmospheric pressure and temperatures a good deal higher than current records, but not at room temperature yet. Lots of opportunities for further analysis in related compounds. (maybe this is the wrong tweet, will try to find the right one).

Renewables should be good before someone developed fusion with the help of superconductors. I own some Infratil.nz which is long a good deal of renewables (although I was long before any of this stuff came out)

I’m sure there are but I don’t know enough to confidently say something like AMSC that’s popular with the retail crowd will win or lose. ATKR seems reasonable though.

what do you mean?

Ha, its your billionaire origin story.

Yea, I looked at Sun Communities. Trailer parks have really good economics, as you have a captive customer where you can just raise rents over time. There also is the role up angle too as the business used to be quite fragmented as once in a while you can buy trailer parks from mom-and-pop owners for good prices. However, a lot of that has played out. It's very hard to build trailer parks in the US as municipalities don't like them. I bought some LIC.ASX and INA.ASX. They are the Australian trailer parks (actually LIC.ASX is going more into retirement communities) much earlier in the development cycle, with more land to build properties and more room for pricing power. SUI has invested in INA.ASX. The problem is Australia's real estate looks shaky. You can find information about the industry from podcasts: https://open.spotify.com/show/39m7rwaGmGyAfTfPtjwlh7 and related shows. Looked at this a while ago so I forget, why do you think SUI is mismanaged? Obviously haven't done as well as ELS.

Sure but from what I know, computer replications can indicate feasibility. Looking at commodities, this is roughly the cost per mole of LK-99. I'm not sure I want to buy miners to play this: Phosphurus 0.30 per gram 123.88 gram per mole 1 Mole per solution 37.16 Price Copper 0.01 per gram 63.5 gram per mole 3 Mole per solution 0.59 Price Lead oxide 0.10 per gram 223.2 gram per mole 1 Mole per solution 22.32 Price Lead sulfate 0.40 per gram 303 gram per mole 1 Mole per solution 122.29 Price

Apparently, a room temperature superconductor has been proposed and initial replications seem to support the original paper: LK-99 requires lead and copper. I'm in no way an expert but probably bullish on battery power: EVs and Renewable Energy (although maybe also bullish for fusion, which is probably long long-term bearish for wind/solar). Could benefit quantum computing and maybe even regular semiconductors (but that I'm less sure about). Goodness, I hate how this makes every meme stock idea better. Bearish for fossil fuel companies as well. Maybe considering investing in lead. I probably can't access lead futures, but it might look at miners with large exposure to lead. Anyway, I guess it's kind of cool to think about. Anybody else have any thoughts?

I mean if you are at the intersection of the metaverse and helping people find their true love, that thing should sell itself!

Wait so despite all the BS, this company is profitable? Even if those numbers are fraudulent, why sell the whole 9x P/S bit if it reports profitable numbers? Tangentially, are pump and dumpers just getting worse at their jobs?

Do you have to pay $50 to view the forum? I feel like that walls off new customers because they don’t know what they are missing. Should be $50 to post, if that’s not already what you do.

I don't have any idea what is this egg people are talking about, but I first thought it was about EGG.AU. Trading at 3.5x FCF, 7% dividend, 10% buyback. Management is on record saying it's cheap at $1.75 and cheap even at $3. Insider buying. Business is marketing consulting (and advertising arbitrage) so not really a growth business, but not going away soon. Sorry for derailing this conversation to the topic of equities. I'm sure @John Hjorth will give me a tongue-lashing for being off topic

This might be an opportunity for others to catch up, but doing this kind of research is essential for the profitability of GPT-4 etc, even if quality goes down, you can always bring it back up. People expect LLM costs to be free per query (ad-supported or subscription-based). If you can't find a way to reduce the costs of your LLM, you could have power users eating up significant profitability for each prompt. For example, for state-of-the-art replies, you should use a technique like Chain of Thought prompting or its descendants. This requires prompting the LLM multiple times, evaluating which prompt is the best (checked via an outside program like a compiler, calculator, knowledge graph etc.), and then using the best prompt as a basis for trying again and again. One prompt could query an LLM 20 or more times. If you just want to take share and you don't care about profitability you don't optimize LLM size, but LLMs are expensive so figuring out what the smallest LLM is that can satisfactorily answer your request is important. Accuracy can go from 10% to close to 100%, but it's really expensive: https://www.promptengineering.org/tree-of-thought-prompting-walking-the-path-of-unique-approach-to-problem-solving/#:~:text=Tree of Thought Prompting%2C thus,and focusing on promising ones.

^ @SharperDingaan so when these models are trained, you have code that attempts to filter “bad” data. Keep in mind that attack, is really difficult as there a 100 billion to 1 trillion tokens in the training set after being filtered. Additionally once the model is trained, you have to use reinforcement learning with human feedback. This also helps to scrub bad “habits” the model has picked up. Basically you have people rate model outputs, and you modify the model to produce outputs that humans rate highly as desirable (whatever desirable is defined as). So long story short, there are multiple steps of training making this adversarial attack on training data very hard to pull off.

One thing you can use is plugins. For arithmetic use a calculator plugin that comes with GPT-4. Basically the way this works is GPT produces a prompt and then runs the calculator to verify the prompt is correct. This is also how code interpreter works. History is more difficult because there is no black box program that is evaluatIng a given historical output is correct. You can try using a knowledge graph plugin, if one has been developed for GPT. Knowledge graph basically scrapes Wikipedia etc to get a graph of relationships (ie this caused that, or this leader started that event etc.) You can also try chain or thought or more advanced multi step prompting.

I wrote this on Seekng Alpha: https://seekingalpha.com/article/4608680-a-data-scientist-explains-large-language-models-and-implications-for-businesses. Let me know if you are looking for something else.