Agreed, but this will probably be limited to domains where there’s better products
116 karma · joined August 13, 2018
Again - to each their own. But what people use google for GPT doesn’t replicate anyway (and what the Google business was built on) - which is commercial info retrieval.
To each their own.
If indeed a similar disagreement happened in OpenAI but this time Hinton (Ilya) came on top- it’s a reason to celebrate.
Not saying that isn’t impressive - but it also usually isn’t what Hnews community values. So I’m just a bit perplexed.
The point was fairly clear even if he just uses their names, the usage of quotes if quite bizarre.
If he wanted to get that point across he should have called them “board members” which is a clear insinuation of puppetry.
The company is not profitable and miles away from being profitable, I’d go as far to say it doesn’t have a path to profit.
Outside of the copilot use cases that MS is leading - GPT is both cost ineffective, and not that terribly impressive - it’s built on foundational technologies developed elsewhere and is not miles away from similar models built at Meta and Google/DM. At the point it was launched and started generating terribly inflated buzz that formed the AI balloon - both Meta and Google has similar scale and parameter models already running in their stacks.
The only thing he did is package the product nicely and put it out to masses (an ethically dubious move that couldn’t have been done by big corpos for PR reasons - explicitly because it formed a misinformed balloon). He did that at huge cost, even though the product is largely useless outside of some eyebrow raising and incidental gimmicky use cases.
All of the actual product work (i.e copilot and distillation that GOT brings) was done by other companies.
What is everyone drinking and how can I get on that? Is he getting credit for bringing sth that was widely known to the AI community to the masses (and thus starting the AI arms race) hence bringing in more mainstream capital funding? I’d argue itms not a good thing that technology as powerful as foundational AI is now being debated and policy formed on by people who don’t know the first thing about ML, I think we skipped a couple rungs on the natural evolution of this - which is why the whole AI safety debate started.
He did all of that because he wanted a moat an an edge over the competition (including trying to regulate the competition out of the running). This is like Apple-level shenanigans- something that HNews usually despises.
I genuinely don’t get where the impressiveness is coming from?
Is it being non comp-sci that automatically invalidates proper usage of your actual name? Or is there another key by which their names are less worth?
They are also both fairly established in their respective fields - which - yes - isn’t hard comp-sci, but if you think tech companies should have purely comp sci board leads, I’d call that incredibly naive.
They were also presumably vetted by the other board members - unless you think they are serving a different purpose on the board (diversity targets?) - which if so - puts the comment from red flag into mysoginist territory.
Personally I don’t see anything in their CV’s that would disqualify them from executing their function on the board, and I wouldn’t call them incompetent in being able to assess whether a person lied or not (which even in your theory - Sam would’ve done). You don’t need to be an ML/AI expert for that.
This review is a lot less than a smoking gun though. The author goes into large detail on the quantification of associations and correlations in the first section (association review) - and limits towards communicating only qualitatively that there was an effect when getting into the longitudinal and experimental reviews (the causation-proving ones).
Reading through the actual google review doc, this is present in the review as well. Very little quantification, and when quantification is given it's the wrong type of quantification. Even in an experimental study, showing mean comparisons rather than distribution comparisons is misleading, even when statistically significant.
This is a bit of a problem, as it doesn't go into the effect size of the causation. The author is quite cleary rooting for causation to be there. This is not uncommon in science (a strong conviction into an unproven hypothesis) - but when this happens, we have to be extra vigilant in looking at the results.
EDIT: Also, for sth to be a major cause we would need to demonstrate that the effect of social media is in the top quantiles of all known effects. The most we can get out of this review (and I'd even challenge that) is that there is enough proof that social media is a causally proven contributor to mental health deterioration. Nothing presented indicates anything to say that it's a large contributor or that it's a major contributor, it just isn't discussed at all.
Anti-discrimination legislation has already made black-box algorithms illegal if they are deciding on anything that a user might take objection to - so for most use cases this is not a big change.
As for - the recommender systems will have to not be based on profiling - unless we're talking about removing recommender systems based on data altogether - it will be interesting to see what the legislation considers profiling. If I tie your recommendations to the last viewed piece of content (content contextual recommendation), is that profiling? It's arguably worse for the user and for society more than profiling recommendation. If the recommendations are based on your explicit categories is that not profiling? Yet it's the principle used in news aggregators for the last 30 years.
The wording is going to be important here.
And this is a true dichotomy. The "autocorrelative" effect doesn't need perfect correlation, just some correlation.
Full disclaimer - I was a sociological researcher before I started working in IT - and would (I can appreciate the irony given all of this is about DK effect) rate myself as very significantly above average in terms of methodological rigour and mathematical skill compared to other social researchers.
One thing that is taught to social researchers - although I've seen it much less with psychologists - is that social research is fundamentally different from natural sciences in that it is accepted as fundamentally subjective. Now, a radical such as myself will tell you that all research, including natural science, is not entirely objective due to very subjective navigation of selection bias, but putting that to the side - this is an extremely important point when evaluating social research.
Coming back to your original point - I would agree with the points you object to vis-a-vis original DK Effect paper, however, as a social researcher, I am always already coming into reading that paper knowing that I'll have to take it with spoonfulls of salt. There is no need to write the paper in a way that puts in many of the disclaimers you might expect, because we are institutionally taught that these disclaimers apply.
Having said that - one of my peeves with social research, and why I ultimately went away, is that a lot of garbage goes on and gets through peer review. There is almost no proper testing of quantitative instruments and methods. Which is why I agree with your point that it rightfully isn't taken seriously - but I would object to your assertion that it shouldn't be taken seriously. Especially amongst IT professionals who are already going to have a bias against non-STEM. Point out the shortcomings and apply a different interpretive lense, rather than discounting the field completely - as social science can be better and taken seriously if it was held to a higher standard, even with the methodological shortcomings we have today - but it is very often discounted wholesale, which I don't think is going to incentivise the bubble that is forming around it to reform and get better.
What I couldn't imagine back then is that Twitter is going to be the main platform (I was using a nascent FB and early 00's bulletin boards as the examples of the system). And Twitter is absolutely bonkers in terms of how it polarizes opinion. If I didn't know any better, I would think that it was specifically designed to achieve this purpose. If I needed to design a system that polarizes opinion more than Twitter, I would be hard pressed to do it better.
1. The 160/240 character limit forces you to reduce topics to their most impactful tenets - which strips out all of the nuance and grounds you into an extreme position. For the one expressing the opinion - it will always look more benign than for the one reading it, because the writer can't fully strip it's perception through bias of known context. 2. The users have "gotten around that" by starting threads - and Twitter has started optimising the feed to show threaded views - but threads are not generated "at once" (there are tools for that, but most users don't use them). They are generated post by post - and in a fairly complex topic (such as, oh I don't know, the war in Ukraine) - this will mean >10 tweets as part of the thread. Each of those tweets is replyable and because you can like parts rather than the whole thread - it will show in people's feeds piecewise -> which will create discussions on parts of the point, not the full context. 3. In turn - these piecewise reactions are forming a base for someone else's reaction, where you've gone even further into the rabbit hole. By this point - an argument has already started - and because the 240 char limit doesn't allow for nuance - the argument looks like a black and white argument - as there are no points that are in the "grey zone" that a reader from the side could connect between the opposing sides and perceive any hope of consensus. 4. Users from the outside see it as B&W - so they enter with B&W assumptions in - and assume an aggressive tone - completely derailing any hope to avoid creation of opposing echo chambers.
I can see how they got there - and I can see the choices they made - 160 characters was a good choice for user engagement and revenue - as the flywheel will allow for more ad insertions - and if you optimise for quick reactions - your algorithm is going to take you into a zone where people don't read entire threads - because the algo isn't rewarded for people idling about and not reacting/answering. In terms of what it's doing to public opinion - it should be a textbook example of how growth-oriented product design - when not evaluated hollistically, can create unwanted social effects.
EDIT: Forgot to link to original point of the article /facepalm. The reason why it's never going to achieve higher penetration - in my view - is that a lot of people get adverse emotional reactions from how the algorithm works - so you end up having people who have to be there (media) - and people who are quite comfortable with emotionally loaded engagement on social networks - which amplifies all of the functions stated above. And those people are, fortunately, in a minority. However, the fact that people don't tweet, doesn't mean that people don't read the tweets - as twitter is public without an account and major news outlets usually link to twitter threads. This drastically amplifies the reach of the distorted echo-chamber battles - which makes the overall effect on public opinion even worse.
I mean it's perfect in how horrible it is.
Ultimately - it's the user choice where they're going to do it - and if the user wants to reward the company, they could subscribe through other means. Most users don't know this, obviously, but it is a possibility. And if we reach them through a different acquisition channel, the users won't subscribe through Apple, they will subscribe elsewhere.
However, the customers who subscribe through Apple are usually a lot more valuable to us (more affluent, stable payments, more engaged), and even with the 30% cut, having Apple as an acquisition channel is valuable. It is not worth it for us to go out of Apple's ecosystem.
Apple know this - and are just playing the economics. They are aware the added benefits to the user and the marketing pull they have in their store. The next step for App Store is going to be changing the 30% to fit the business model of whatever they're marketing - if they're rational. There was already suggestion that for streaming apps (very high operating cost, very low margin) they might drop to 15% - because they'll try matching the economics to the barely positive tradeoff for the businesses using the app store.
I see nothing wrong with the Apple model from a business standpoint of a multi-platform business. When something is specifically an app business it becomes more tricky because essentially in order to acquire the customer they have to install the app so the app store becomes your main acquisition channel. But even there you could argue that your business model wouldn't exist if Apple wasn't providing the customers, OS, app store and marketing. So it makes sense for Apple to take a larger cut (in relative terms per acquisition volumes).
If I was a betting man, I wouldn't bet on this being a significant change, given the world is moving more towards privacy (yes, including US).
And even if it was supported by research, I would think about the long tail. What does this mean for my user engagement in the long run. This list might satisfy them now, but it necessarily leads to a narrowing down of the content pool in the long run. I would ask my marketing sciences unit or my data science unit, whatever I have, to try to forecast or simulate a model that tells us what would the dynamic of user engagement be with intervention A and intervention B.
I feel this is one of the biggest problems of program management today. Too much reliance on short-term A/B testing, which, in most cases, can only solve very tactic problems, not strategic problems with the platform. Some of the best products out there rely much less on user testing, and much more on user research and strategic thinking about primary drivers in people.
If you were to use this approach - you might see that actually, the product you have with choosing to optimise for short-term engagement brings less user growth and less opportunity for diverse marketing - which, it is important to note, is one of the main purpose of reach-building marketing campaigns.
I would say the way this whole problems is phrased shows that the PM, or the company indeed, is only concerned with optimising frequency of marketing campaigns, rather than the quality, reach and engagement with marketing campaigns.
Obviously, hindsight 20/20 and generals after battle and all that. I'm still pretty sure I would've thought more strategically than "how do I increase frequency of showing ads".