Won't this just encourage maintainers to sit on issues... just waiting for the bounty to increase? They might even purposefully introduce painful but simple bugs just to push consumers to pay to have them fixed.
Would be cool if bounties could have an expiration date. I’ll pay $100 if issue X is fixed by date Y. Probably more reflective of business value too, since if an issue isn’t fixed relatively quickly I’ll have to build my own workaround.
Polar could certainly be abused that way, but I think it's a theoretical problem more than a realistic one.
Open source operates in a marketplace today (just not an economic one) so market dynamics would flush this out imo. Intentional bugs, stale development etc would lead to user churn and/or forks. I'd argue faster within OSS where users can see how things are handled.
Having said that, I certainly think the risk of it and fraud more in general increases once an economic marketplace is formed. However, I think those problems are then "good" problems to have and easy ones to fix, e.g flagging repositories with unusually high ratio of repeat pledges, velocity of bugs etc.
Another kind of problem would be when someone pledges a big sum for an obscure feature, a first-time contributor writes a (technically perfect) PR for it. Maintainer does not want to merge it, because the feature does not align with maintainer's project goals. But not merging will now look like a dick move, as it denies the PR author the compensation.
With Polar today only the maintainer receives the pledges once an issue/feature is completed. However, we're working on the ability for maintainers to split that with any potential contributors. It's definitely an important feature and part of our model. Just something we decided to scope out from the initial launch – in the interest of launching early and building in public.
An important distinction between Polar vs. traditional bounty services though and regarding your point is:
Everything goes through the maintainer(s) and we're equipping them with the insights, capital and tools to reward their contributors. While bounty services allow anyone to post a bounty and anyone to pursue it. However, ultimately it needs to be reviewed, merged and maintained indefinitely by the maintainer(s). By ensuring it all flows via maintainer(s) it:
1. Rewards maintainers as well – key given their efforts
2.Guarantees maintainers are aligned & behind pledges first. Combined with in control on how to expose, delegate and reward it. Avoiding unnecessary friction, bad faith and overhead if such alignment is first made during PR.
Then the payment can go to a fork. If the maintainer really wants to keep his vision for the project and others are literally voting with their wallets for something else, why would that be a problem?
Bad incentive structures are everywhere though, including at corporations where people design "cool" architectures which fail, and then swoop in to save the day.
It shouldn't be possible for them to turn over the chats. E2E encryption is tables stakes for messaging, but they won't do it because there's profit to be had in understanding users through their private messages.
E2EE strictly means that it is encrypted between endpoints. It makes no promises about how the keys are handled, how the data is protected at rest, or how secure either endpoint is. TLS is a version of this, but also demonstrates how useless such a version of encryption is if either endpoint is malicious.
This is one of the reasons homomorphic encryption is interesting. It should enable this kinda of "total user control" over your data, but it's highly unlikely consumer systems will implement it.
It is possible to care about multiple issues simultaneously. Saying it's demented to care about one person being wronged because other, totally different people are wronged is a bit selfish. It's okay to have a purpose that you believe in, but the world isn't black and white.
If you're asking the wrong questions, the answers are going to be pointless no matter how much time or money is invested.
The problem for Meta is all of the questions asked were never congruent with reality and organizationally there's no incentive to work on coherent ideas because of politics. These layoffs are not a surprise to any realist.
Doesn't have to be actually difficult to comply for them to do this. It just has to be more expensive to comply compared to the cost of adding the allergen and the lost revenue from people who avoid due to the allergen.
It's like an auto maker evaluating the cost of a recall compared with just paying out settlements. There's no morality or logic involved, it's just numbers.
Very fascinating. If we train on every piece of knowledge, then statistically that should average around 100IQ right? It's not perfect analogy, but the model is shooting more for consensus, a bit hive mind.
It feels like to improve the quality of the model, we need to be more selective of training data.
It's not entirely clear what precisely IQ is measuring. It's not a stretch that averaging the output of a bunch of 100 IQ people yields something better than what any one of them could have produced. Perhaps "there's only one way to be right, but many ways to be wrong".
I'm glad you posted this. Sometimes I try to describe this gfactor to people, saying "the best indicator of intelligence is any other intelligence". Putting a name on things helps people understand I am not a total bullshiter.
Why do you say that this is a common belief that used to be true, but is now no longer true?
Written knowledge will probably skew towards higher than average IQ. For starters, you've removed everyone who can't write, and everyone who isn't comfortable writing will be underrepresented.
I don't think this is true at all and it suggests a view on what GPT is that I think is very incorrect. Not to rag, it's just a common POV.
GPT does not attempt to learn the average "text-generating system". It attempts to learn all of them. In superposition.
Just imaging how bad it would be at predicting a scientific paper if it spoke (and reasoned?) like the average person. Such a model would not survive training.
Training GPT does not make it behave more like the average, but instead widens and diversifies its probability distribution over all possible follow-ups.
To make this clear, ask it to take on a persona. It can pretend to be all sorts of people, invented or sufficiently catalogued. I've had fun having it play multiple roles from podcasts I enjoy. Or it can pretend to be a fictional FTP server at Disney where poor authors have stored their unpublished screenplays. You can ask GPT to run dialogues with itself, playing both sides of the argument.
There likely is some level of global persona at play, either through fine tuning or there being a general attractor basin of "helpful assistant" that we're all reinforcing. But there's no reason to believe that this looks anything like consensus or averaging.
I agree there is a lot of "averaging" going on, but here are two things that I think make it somewhat better than "100IQ". First, it seems like the people who talk the most about many narrow subjects actually have more interest and perhaps capability in them than the median person, and it will train mostly on the people who talk the most about particular token sets. Second, there is the unreasonable "wisdom of crowds" which works for some set of things (though it may no apply to LLMs).
So it's weird, but I expect LLMs to give me better answers in narrow well discussed fields than broad but highly argued fields. I guess it's still really important to know how to ask the right question.
Or, it’s a simulator capable of producing output that is plausible for any level of intelligence up to the capabilities of the simulator. It could be capable of producing text similar to anyone from 0-140 IQ depending on the prompt.
It will act “smarter” if the prompt indicates a smart person wrote the text.
It’s both really if you look at IQ tests. It’s part of the reason immigrants will often score lower, they lack the culture or language specific knowledge being tested.
IQ is a proxy for intelligence. People often use them interchangeably, but they’re not the same thing. Your statement is more valid for intelligence than IQ.
It’s not just that its the ability to reason in different modalities without specific knowledge, as well as others. Speed is certainly a part, but not the only part, of the IQ text.
