dhfromkorea

Tempus AI | AI Scientist (Staff, Senior) | REMOTE (US) | Full-time | https://www.tempus.com

Tempus is a leading precision medicine AI company with an exceptional data flywheel in oncology: multi-modal data (pathology, DNA/RNA, spatial -omics, longitudinal outcomes, etc.) for millions of patients and deployment channels to the majority of the hospitals in the US.

As an AI scientist at AI Products group, you will build frontier clinical AI products for clinicians and life sciences partners, on top of our portfolio of foundation models in oncology. If you get excited about proving a novel product is feasible at all, or building predictive biomarkers for a new drug out of a small cohort, and you want impact beyond the benchmark results, this is the group for you.

Many of us come from Paige AI, the creators of Paige Prostate (the first FDA-cleared pathology AI application). We look for future colleagues who are smart, driven, and good at working out hard problems together.

Staff: https://www.linkedin.com/jobs/view/4387110060/

Senior: https://www.linkedin.com/jobs/view/4453706650/

Please apply and email donghun(dot)lee@tempus(dot)com

Referrals are very welcome.

This is really cool. I searched for "Korea" and the third result was a page of Richard Stallman's website about his visit to Korea in 2000, from which I discovered he wrote about an array of big companies and why not to use them---pretty interesting. Last time I met him was in Finland. Such a consistent, profound character.

> "SoftBank routinely selling assets to its Vision fund."

Can anyone help me understand the real reason Softbank does this and whether investors of VF are okay with it? One case I know of is Coupang [1], a Korean E-commerce company, recently "priced" by VF at $9B. Last year, Softbank sold its 20% shares in Coupang to VF at a 30% loss (down-valuation) and then VF poured in an additional $2B to Coupang. It appears Masa did double down on Coupang but why bother to sell its long position at a loss? Not a domain expert on this, so I wonder what is really going on...

https://www.bloomberg.com/news/articles/2018-11-20/softbank-...

It would have been more interesting if they could share deeper, nuanced arguments for setting wealth as the objective function. Sure, wealth past a certain point admits a lot of freedom---but what next beyond that point? But the utility function over wealth (x: wealth, y: utility) is sub-modular in nature. When the curve gets flat, what metric must one use to measure her life's worth? Is there even one as such? Having followed their opinions for quite some time, I know they must have some good opinions on these...

> The idea that priors are somehow ruining an "objective" model is just absurd to me.

I think some caution can be justified to a certain extent (not the blind "emotional" objections). When establishing priors in a low data regime, one must necessarily be careful. It's a knob whose mass can change a lot in the inference conclusion. That said, if we trust our belief about the region the available data do not inform us well of, why not utilize our domain knowledge/belief?

one vote for BDA. For programmers who learn better by implementing things, this book [1] is also good:

[1]: https://www.amazon.com/Bayesian-Methods-Hackers-Probabilisti...

The employment rate has increased by somewhat unnoticeable amount over the past 10 years [1]. The increased housing supply is mostly due to cost factors (tax breaks from the previous administrations, low interest rate) and the demand due to decline of small cities (demand gets concentrated around Seoul, the capital city; people buy houses around Seoul while living far away in another place on rent). As for the office/factory/commercial space, there has been a sharp increase in supply, and the vacancy rate (spaces that cannot find tenants) is 10-20% in major cities. Folks still buy buildings with a yield of 2.5-4% because the interest rate is so low. I am currently based in Cambridge, Boston. There are empty stores that have been looking for tenants for more than a year.

[1]: http://www.index.go.kr/unify/idx-info.do?idxCd=4013

I am not convinced. The attribution to the simple zoning system is surely not a complete proof, if any. Consider a counter-example: South Korea that has a very similar zoning system. Yet the housing price around Seoul soared roughly by 60% the past 3 years---so did rents (to a smaller but still significant rate). All the while the apartments supply has been at record high in 10 years. That zoning policy may be a factor but clearly not sufficient.

While switching to the simple option may help the likes of Uber defend themselves on the legal front, the switch may not be that simple to implement and will be costly on the business front (if not making them cease to exist). First, the switch will likely lead to an increase in ride fares and a decrease in user experience. Second, it will be increasingly difficult for the Uber's management to justify its massive operating losses, when they cannot say to their investors "hey, we can always increase the price at will".

I was just curious, not expecting anything particular---I wondered if Uber's business and its problem space are fundamentally bad as some of the comments here suggest. Re: your opinion, I think it's a good news for Uber. If it just takes Lyft going out of business for Uber to surface above water, I would bet there's a realistic chance.

A question: if you can make one assumption about Uber (or fix one thing about it) such that Uber will be sustainably profitable in the future. What will it be? Why do you think the assumption implies sustainable profitability in high probability?

Really nice! One suggestion would be to have a list of resources for slightly more theoretical materials, so curious students can be exposed to deeper parts of ML (of course, such materials need links to applied tutorials like this). Perhaps can be done through the pull requests of the community.

I thought so.... until I took a real analysis course. Most calculus tricks build upon on a "small" set of big ideas. The big ideas (such as compactness, convergence, continuity, diff/integration..) are limited in numbers to convince oneself on, yet are generalizable tools to think about A LOT of complicated mathematical phenomena in a concise, clear way. Should they fall short to evaluate a certain math phenomenon, they should do so unambiguously rather than opaquely. Real analysis is a start to developing such tools.

The environmental externality must be internalized strongly and consistently throughout the "developed" and developing economies [1]. Participation from the two groups is a necessary condition for addressing the air pollution issue.

Meanwhile, the situation in London is similar in South Korea [2]. Most likely, worse: South Korean government is afraid of further regulating the domestic industry, when it should. Over the past decade or so, Chinese government has moved a lot of their factories to the East coast which faces SK in the West. Both SK and Chinese citizens must acknowledge this is a mutually-interested issue and take concrete steps to collaborate while minimizing finger pointing. I am convinced this is a tractable scientific and social/diplomatic problem. One proposal is to set up a joint government-backed venture/think tank that incrementally tackles this issue at a tech/policy level. Such a small step but no where near it...

[1]: https://en.wikipedia.org/wiki/Externality

[2]: https://www.theguardian.com/environment/2019/mar/13/social-d...

RIP, Mr. Bogle. Does anyone know how Bogle concretely thought of the existence of samples like Warren Buffett who have outperformed the market over decades, granted the outperformance may not continue in the future [1]. I presume Bogle's view on this would be nuanced.

[1]: https://qz.com/1216260/warren-buffett-doesnt-beat-the-market...

Oh, now I see. Thank you for explaining your point succinctly.

Though I do not have data to back up my argument, I still reckon the Chinese glyphs/scriptures would have had a better chance of survival.

While I think your point is valid, its disadvantages outweigh the advantage, at least since paper/papyrus was invented.

Being spread out to double in length (double being an arbitrary multiplier) would still be inferior to being dispersed to two physical locations (redundancy). I think this is where don't put all your eggs in one basket holds true.

Plus, important docs must have been actively maintained by hired librarians(?). With human maintenance involved, less in volume could have been an advantage for it is easier to move around and maintain the docs. Ofc, when left out in the wild, it is a different story.

Personally I do not like Chinese character system as it has so high a barrier to entry for learners. I love alphabets, Korean Hangeul, or Japanese Hira/Katakana for this matter. Have you tried learning any of those? :-)

That may be true. And perhaps the same argument could be made for Chinese characters?

Could you elaborate on why the alphabetic system is intrinsically more efficient than Chinese characters in terms of recovering messages from partial loss of texts?