> If the app is free but offered by a commercial entity, they pay nothing in commission or payment fees (13% X $0 ) but they reach 1 million installs, then they do have to pay €500k. If the business has no revenue - that’s a weird business model.
I think that "growth first, revenue later" is actually a fairly popular business model, and I can imagine a lot of my own app ideas reaching 1M downloads with <$500k revenue (in an ideal world, but still)
Agree with the insights in your comment about biometrics != security, but I'd like to take a moment to nitpick a slight inaccuracy-- Asian faces don't actually look similar to each other, but they do look similar to a person/model that has been trained mostly on white faces. If the facial recognition model had been trained predominantly on Asian faces, then white faces would look similar to each other instead.
Reminder that the outputs of AI don't reflect some deeper truth about reality, just an extrapolation of the training data. Garbage in, garbage out.
I listen to Data Skeptic and Practical AI. It's interesting content, and I find Practical AI has guests who are particularly relevant to what I want to learn, but you should keep in mind that the guests are usually trying to promote their own products, so everything they say should be taken with a grain of salt
As a student also trying to figure out how to use AI "correctly" to increase actual learning, I appreciate your effort. However, I'm not clear as to what the idea is. Like the parent commenter, it also seems to me like you're complaining about professors using ChatGPT to generate tests, but your tool seems to do the same thing. Is your tool basically a magic prompt that makes GPT generate higher-quality tests?
All the podcasts I've been listening to recommend RAG over fine-tuning. My intuition is that having the relevant knowledge in the context rather than the weights brings it closer to the outputs, thereby making it much more likely to provide accurate information and avoid hallucinations/confabulations.
I honestly think people like you are the target audience, but I don't feel like it's a huge population. And even for people like you, plugging your bank info into Twitter is only one step less than downloading Venmo and plugging your bank info into Venmo, so I don't think they're offering a whole lot of additional convenience
It's peer-to-peer in the same sense that Venmo is. From the user's perspective, you just plug in your friend's @handle and hit send, even if it's still centralized on the backend
Whoa, this is super cool! I can imagine if we had something like this for ChatGPT, we could use it to do some serious prompt engineering. Imagine seeing what specific neurons you were activating with your prompt, and being able to identify which word in your prompt was triggering an undesired behavior. Super cool stuff, excited to see if it scales
Darn, I was hoping the RWKV people had finally obtained reportable results. This is still interesting, though. Maybe we will see more alternatives to transformers soon
From an efficiency standpoint, I think having to harvest and modify the patient's stem cells is probably the biggest choke point, right? I would imagine that if you could inject something once and be done with it (I'm thinking like Zolgensma), you could mass produce it more effectively
Correct, but if we have good treatments for malaria (e.g. hydroxychloroquine, atorvaquinone) then I would argue that we no longer need that partial immunity
It's reasonable to expect small improvements in the risk profile, but I think blood cancer is going to be a side effect for any drug following this basic idea. You will always need some chemo to destroy the defective blood-making stem cells before replacing them with the genetically-modified blood-making stem cells, and any chemo that is strong enough to kill all of the blood-making stem cells in your body is necessarily going to have a risk of damaging healthy cells and turning them into pre-cancer cells. So the risk can be reduced but probably not eliminated.
You should be okay. I've found a couple of studies that have looked at groups of patients in the hundreds, with no worsening of hearing loss (and presumably, no worsening of tinnitus) [1][2]. Caveat is that both studies were done in kids, since they get more ear infections in general. Also, "topical ear drop medications containing gentamycin or neomycin do not appear to be ototoxic in humans unless the tympanic membrane (ear drum) is perforated" [3].
Yeah, especially with the PPU compensation scheme, all of those employees were heavily invested in turning OpenAI into the next tech giant, which won't happen if Altman leaves and takes everything to Microsoft
Feels like the same kind of language I use when trying to get ChatGPT to stop doing the thing. Just repeat my negative commands in various, slightly different wordings and hope one of them sticks
This is the idea of the book "Thinking, Fast and Slow" by Daniel Kahneman. LLMs follow what he describes as System 1, the intuitive heuristic that gets you through most of the day, versus System 2, the rigorous algorithmic thinking that you reserve for harder situations
Ohh so that explains why I saw `?model=text-davinci-002` appended to my ChatGPT URL the other day. I remember it was pretty slow to load then, too, so that's probably why they rerouted me
Yeah, my response to this is that I'm glad I quit Android app development a while ago to focus on PWAs. Just hope for better PWA support from Apple/Safari one day, even though it is explicitly against their self-interest
Sorry, are you saying that winget, chocolatey, and brew are equivalent to the mobile App Store? Because for the average user, that is not remotely true. The App Store is how you install programs on mobile by default. On desktop, however, the default installation process is to download an installer/executable from a website. Only very technically literate people (basically just software developers) will be installing things via winget, chocolatey, or brew
Hey, if it makes you feel better, a lot of biochemistry research ends up being completely meaningless fluff that is only published to inflate the author's reputation, but will likely have zero effect on the real world.
You can be cynical about anything if you want to :)
Does co-insurance help with this problem? e.g. if the patient is responsible for 10% of the cost of the pills, then the patient is motivated to choose the cheaper pills without the need for a long bureaucratic process with the insurance company, and the doctor just has to present both options
I'm thinking a loading spinner would be the best compromise? Notifies the user that their input has been registered, and also explains why clicking again won't have any effect?