dontreact

In a very real sense a lot of people had no choice but to stay in company towns. There are certainly similarities between a company town and Facebook, but choosing not to use it is not a matter of survival for anyone.

I agree that Facebook, like company towns, should be regulated given the scope. But I think we should evaluate bannings like Alex Jones more from first principles such as whether there is more harm in allowing him to stay on these platforms or in banning him rather than being strictly adherent to the principle of free speech.

It’s certainly not true that Facebook has a monopoly on communication (here we are communicating without Facebook involved)

I do have some agreement that when a platform gets very large, it makes sense to have some regulation around the censorship decisions they take. Specifically, rules around transparency make the most sense to me. On the other hand, a blunt regulation such as saying that large tech platforms are not allowed to do any platform or country-wide moderation doesn’t make as much sense to me. My experience with all unmoderated online communities is that they devolve into something extremely unpleasant, and forcing Facebook to push ALL of the moderation work down to specific groups or users themselves is not a workable approach to this problem.

I also disagree with government level censorship. At least in the American right, there is some political pressure and discourse around pushing free speech as a principle from the government level (where I agree with it) down into the platform level (where I do not agree with it).

It seems like you are not one of those people advocating for that, but this is why I would push back on the statement “No one who advocates against hate speech laws doesn’t want you to have that freedom”

I was responding to the parent comment and tried to make clear that I too am against government mandated censorship. I’m talking about platform level censorship because I thought it would be interesting to the discussion to clarify these two different levels. At least in American discourse about tech platforms I often feel these two levels are conflated.

I agree that government-level banning of hate speech is problematic for the reasons you describe.

However, as a member of a minority group who grew up being the butt of many racist jokes, I would like to have the freedom to participate in an online community where hate speech is moderated. It’s fine to me if other people want to participate on platforms where hate speech isn’t banned, but I will stick to ones that are less unpleasant for me personally. I don’t think I’m alone in this point of view.

Seems like he left and got into Crypto? https://www.linkedin.com/in/doronichev/

It seems a bare minimum is releasing enough data that the public can assess whether it is better on average. Right now, Tesla releases data that show that when autopilot is engaged, it is safer on average. However, this data is not adjusted for the fact that the miles where users choose to activate it are on average safer. So really some sort of base-risk adjusted comparison would be the bare minimum I would like to see them publish. They really don't share much data.

That's how long it takes to get a ruling, but how long will it take them to enforce it?

Why do you think this is a good thing?

Wow, I get the same messages about golfing with Kevin and Mike! (My name is also neither of those)

Seems like there is a wide gap between a single company organizing a union for their employers and a monopoly on the labor supply. I would be more convinced by your point if this wasn't one of the first tech unions (according to headline).

Ok really dumb question but: how do you plug in a normal plug into a battery like this? I'm nervous about doing it wrong and then frying my musical equipment. The nice thing about the Anker is not having to worry about this.

My friends tried hooking up a car battery to an inverter and something went wrong (not sure what) and they fried 3k of musical equipment.

So is this the competitor to Fast, which also had a big round of layoffs recently? (edit: ah yes I remember now it was a complete shutdown)

My current understanding is that both of these businesses were premised on the end of Amazon’s one-click checkout patent. Is this proving that to be a faulty premise?

Makes sense for who? If all competition becomes winner takes all, then most humans become losers.

My sense is that in general, and especially in software, the world is becoming more of a winner take all place. This is not a good thing.

And history repeatedly shows that even if you get a dictator of the first kind, it’s an unstable equilibrium of a system that eventually leads to a dictator of the second kind.

Lots of good points here about how opening up the algorithm itself won't illuminate much and is technically extremely difficult without potentially devaluing the company since you would also have to open the data system behind the algorithm.

However, open sourcing any and all manual interventions over the algorithm + the guidelines used for evaluation and/or labeling (if any is done), would help to build a little bit of trust.

Not that much though, but it would be a start.

* CRUD: Faiss indexes are static, if you want vector search over data that users can edit freely, it's not easy to use FAISS for this * Non-vector filtering: Often you need to do a search based on some structured query + rank the results by vector similarity. FAISS can only do vector similarity.

This is by far the best explanation. The other responses in this thread are either very incomplete, or incorrect.

Just a fun technical note:

CLIP would be possible with any language and vision encoder.

It turns out that transformers were the most efficient encoder choices available at the time but most likely the approach would have given interesting results with a resnet and some sort of convolutional language encoder. In fact the paper has a resnet as one model for the vision side.

I’m curious on the HN community opinion on whether Cruise is ahead of Waymo in SF, or if Waymo is just more careful and has spent less time training models specific to SF, but will expand to a fully public program sooner

I agree with all of this, but ML will evolve a lot over the next 30 years of your career. Being able to read the papers as the field evolves is useful, and many of the papers (especially the ones that shift the field) will assume knowledge of calculus.

I disagree with this. Being able to read and understand the math in the paper of the algorithm you are implementing is useful, and calculus is common in ML papers.

You may have some short term success without understanding the algorithms at all, but as the field changes and you are no longer in school, being able to keep up at least somewhat with papers is very useful.

I agree that the day to day is mostly about formatting data though!

AI in medical imaging has been very limited and ineffective thus far.

A) I think it's too early to make a call here. The first big paper for AI in medical imaging with modern methods came out in 2016, and we are just now getting large scale prospective studies where algorithms outperform specialists: https://www.thelancet.com/journals/landig/article/PIIS2589-7...

My institution pays for an expensive lesion detection / mammography CAD software package but it generates so many false positives I don’t use it in my practice. Overall having to review an extra n pseudolesions per scan results in decision fatigue and increases my miss rate, at least in my experience.

B) Totally agree that old CAD has not been effective, and this has created a huge trust gap for the new generation of technology. The new generation of technology using deep learning is also easier for anyone to create so it will take some time to separate the good from the bad.

There are several issues with AI when it comes to radiology that I can’t address with a single comment but given your question my biggest criticism would be that we don’t have a sensitivity problem in medicine. What we have is largely a specificity problem, but that’s inherent with imaging which is not a ground truth representation of a physical entity but rather how this tissue attenuates photons (for mammo).

C) This claim seems overly broad. If we can improve sensitivity it can certainly be a good thing. That being said, improving specificity in an assisted read scenario seems almost impossible in my experience, because if a doctor is convinced something is cancer, no output from an algorithm is going to get them to downgrade. We did an analysis on exactly what was going on when our algorithm would give a lower, non-recall score vs. a panel of radiologists, and we found that the most common situation was missing that a lesion was a scar. https://www.nature.com/articles/s41591-019-0447-x

It turns out that radiologists will in general not look at a lesion from all different possible reconstructions in 3D, because it's too time consuming. But sometimes there is only one reconstruction/view where the scarring is obvious. AI obviously does not have this issue, so in principle it is certainly capable of improving on specificity and this has been seen in many peer reviewed retrospective studies.

As I mentioned, the issue is how to make this happen in clinical practice where assisted read is the only option that will be tolerated for deployment for at least the next 5-10 years. In that time period, I think the only thing we can really hope for is a boost to sensitivity without impacting specificity. I believe eventually enough strong evidence (prospective observational multi site studies) will accumulate, and the algorithms will improve enough that for something like mammography, at least a portion of the reads can be done without a radiologist. This is probably the only way we will see improvements to specificity AND sensitivity in my opinion.

Yes it's not trivial to balance these things and this is why there have been randomized control trials that look at long term end points where both the costs of false negatives and false positives come into play.

It's also why breast cancer screening is not recommended for everyone, just women over 50.

For women over 50 there has been a decades long sincere effort by the medical community to take this all into account and the consensus is that screening makes sense in that population.

For starters, false positives are an inherent part of medical diagnosis. Interpretation and diagnosis is probabilistic based on ROC curves balancing different sensitivities and specificities balanced with disease prevalence as well as the significance of missed/delayed diagnosis.

I totally get this and it's technical terminology.

I think my takeaway from trying to explain why this isn't a big deal on this thread is that calling a tomo or mammo "positive" is just a recipe for confusion.

I understand why it's done this way historically, but the idea of a mammo or tomo being "positive for cancer" makes no sense since there will almost always be either an ultrasound a biopsy or something else before a diagnosis is made. It's a test for whether a more invasive/expensive test is worth it, not a test for cancer.

Curious what your opinion is on AI for DBT? I think there is great long term potential here, even more so than the potential that we have seen so far in mammo (such as https://www.nature.com/articles/s41586-019-1799-6), because it's easier for AI to thoroughly look at things in 3d and spot new patterns that are not obvious to human eyes.

Yes, and this is how the standard of care treats things. You never proceed from positive mammo screen to treatment. You always first do some other steps like ultrasound or biopsy.

Nice follow up!

Yes, NLST was actually a real randomized control trial for lung cancer screening, done before it was known whether or not it was helpful. The trial eventually stopped once they had enough outcome based evidence.

The statistic you mentioned is not used in assessing the effectiveness of screening. It does closely relate to what is though to be the causal mechanism that leads to effectiveness.

You can still try to estimate this using something called QALYs which are life years adjusted for quality of life.

A breast cancer scare, is certainly harmful, but it's important to remember that it is going to be a short episode as it becomes clear that the tumor is not growing. So you have maybe a few months of scariness.

On the other hand cancer is brutal, and death is permanent. Years of life cut short. The years leading up that are extremely sad. The impacts of death extend beyond the person to their family and friends.

When people try their best to measure these things on balance, breast cancer screening is overall a good intervention.

Yeah I just linked that study because it was looking at QALY which is an aggregate metric so it should in some sense take into account the effects of overdiagnosis even if overdiagnosis is not explicitly handled.

Here is another study looking at all-cause mortality, which again does implicitly handle overdiagnosis. https://www.ajronline.org/doi/full/10.2214/AJR.14.12666