HNHacker News
TopNewBestAskShowJobs

drtournier

106 karma · joined October 25, 2019

submissionscomments
drtournier··on Flashlight: Fast and flexible machine learning in C++
Flashlight vs. Scikitlearn ...
drtournier··on Why are so many successful founding CEOs software engineers?
I wonder how much classic survival bias plays a role in this.

Since most startups fail and the failure is usually due to lack of cashflow, startups that has the founder as a software engineer can have more bootstrap resources.

Which potentially increases the chance of these companies to cross the chasm.

drtournier··on New coronavirus mutation can reduce vaccine efficacy
It's an interesting remark! We still don't know about long term "immune memory" from our body. Assuming it would behave like all other corona viruses, I think that plasma would not have long term consequences. Actually, NEJM published a study showing some fair improvement for patients who take plasma in early stage
drtournier··on MIT study: Covid-19 vaccines may be less effective for racial minorities
Some interesting remarks to put in context before judging this MIT claim as true or false:

- Racial bias in medical research is real and methods for increasing fairness are urgently needed in clinical trials

- We don't know for sure how racial bias was balanced on most phase 3 trials, because most research protocols were not disclosed to the science community

- Since efficacy was measured by the sub population in the phase 3 trials who got infected, we can perhaps have other confounding factors who can affect more, or less, some ethnicities (for example, someone can raise the hypothesis that some cultures are more prone to use masks and keep distancing, reducing inherent risks, and that can be true)

- Sample sizes for the phase 3 studies so far very small (perhaps enough to assess minimal efficacy?) for ANY race/ethnicity. We hope that with the increase of the vaccinated population we can have a more precise estimate

- We are talking about 95% inside a confidence interval which is still uncertain lower/higher boundaries, given the small sample

- It is very dangerous to claim any statistical conclusion with small sample sizes

- Almost all "advanced machine learning AI methods" fail miserably. And they fail even worse with small datasets (which is the case). Examples are the handful of "covid pollsters" we got since the beginning of 2020, trying to guess how many people would die, or get infected.

- About vaccine efficacy in asians - we should monitor phase 3 results of CoronaVac (the chinese vaccine) trials (which are in course in China, Brazil and other countries), so we can account for these biases. I heard these results are coming in the next few weeks

drtournier··on Mmhmm
Quick thoughts about it. I hope the authors read these comments and make something more integrated. It's not because it's funny that I will waste my time duplicating slides that I already made using ppt/gslides/keynote.

Other remarks:

Good:

- Seems fun to use and build things, liked the humor Needs improvement:

- UI could be cleaner, the nagging about free premium takes a huge part of screen real state

Bad:

- No support to use google slides, powerpoint or keynote. If you think I will spend time redoing my slides in your tool, forget it. It just made me so pissed off that I deleted the app from my mac right away.

- Pricing should be transparent. No free tier (even with a watermark) is a huge bummer.

- Performance in older macs is subpar. My video slurred a lot in a 2015 Macbook 12'

← PreviousPage 2 of 2