5,023 karma · joined December 28, 2015
https://mobile.twitter.com/THISWILLWORK/status/1593376111563...
Twitter is full of employees with 5-10 years of experience sharing that they are leaving. There is no doubt many people with critical knowledge and skills that are not easily replaced are gone.
https://masto.host/the-25-percent-commitment/
We will likely migrate to our own cloud setup once size gets stable.
We have no idea why this happened, and we'll work to fix it. Bear with us please.
Here's my rationale: Twitter is kind of invaluable if you are an academic working in AI (and in many other fields as far as I can tell), it's a great way to discover new work as well as promote your own (and yeah there is silly drama sometimes, but it's rare). Many academics genuinely love it, including myself. Those same academics generally loathe or at least dislike Musk, and worry about the future of the platform, and don't care that much about non AI stuff on there. So if just the part of Twitter you care about it migrates to a Mastodon instance, such academics would likely consider it. And if you are a person or entity with a decent following on Twitter, it does seem feasible to encourage the community to at least try to move over. Maybe, maybe not, but I figure it's worth a try (I am planning to announce it on Twitter later this week).
It's unclear what paper they refer to, as it is not linked. This article (https://www.ncbi.nlm.nih.gov/books/NBK361016/) has many citations to papers that cover this stuff. I think the important take away is in "Antidepressant drug effects and depression severity: a patient-level meta-analysis" (https://pubmed.ncbi.nlm.nih.gov/20051569/):
"Results: Medication vs placebo differences varied substantially as a function of baseline severity. Among patients with HDRS scores below 23, Cohen d effect sizes for the difference between medication and placebo were estimated to be less than 0.20 (a standard definition of a small effect). Estimates of the magnitude of the superiority of medication over placebo increased with increases in baseline depression severity and crossed the threshold defined by the National Institute for Clinical Excellence for a clinically significant difference at a baseline HDRS score of 25.
Conclusions: The magnitude of benefit of antidepressant medication compared with placebo increases with severity of depression symptoms and may be minimal or nonexistent, on average, in patients with mild or moderate symptoms. For patients with very severe depression, the benefit of medications over placebo is substantial. "
Short version: if your depression is mild, antidepressants may not work better than a placebo (though placebos are not exactly nothing). If it's severe, there is more of a chance it'll really help,
From my personal experience, I can definitely attest to them helping, though they are not by any means a magic bullet.
But yeah, in general the article assumes that GPT-3 will have lots of applications that make it super easy to make a useful product with very little extra effort, and that is just not true. Twitter demos are easy, robust and useful products are not.
It may be argued we need more human-like learning mechanisms. Then again, if we need internet-scale data to achieve human-level general intelligence, so what? If it works it works. Of course, the comparison has some value in terms of knowing what can be improved and so on, especially for RL. But I wouldn't call this a 'troubling problem'.
This sort of website would be incredibly light weight (just API calls, no data storage), and the users would not be tied down to it; as long as they retain ownership of the login credentials to whatever cloud provider they choose, even if this front end shuts down the fediverse instance won't. And if it's open source, it would be easy to migrate to another such 'frontend management' website.
I have thought about this quite a bit, and it seems like a great idea - is there something I am missing (aside from it perhaps being nontrivial to set up a Fediverse instance via API calls to cloud providers)?
Of course, a clear issue is that there would still be (small) ongoing costs even for tiny instances, so it would not be the same as Discord or reddit in terms of having a 'free tier'. But the pitch of a 'personalized social media website' seems like a pretty cool idea and i'd be willing to pay a bit to try it out.
These guidelines are deliberately broad and allow for many exceptions wrt politics.
Really you need image+text->image instead of just text->image generation. Some examples of relevant papers: "Conditioned and composed image retrieval combining and partially fine-tuning CLIP-based features", "IMAGE GENERATION WITH MULTI-MODAL PRIORS USING DENOISING DIFFUSION PROBABILISTIC MODELS". There was a more recent one I saw on Twitter I don't recall the name of. I wouldn't be surprised if these kinds of things work well by a year from now.
Also, the ideas page has lots of weirdly low quality stuff (eg heath ledger joker does not look like heath ledger, baby yoda does not look like baby yoda, Totoro tattoos are generally bad, etc.). And then there are things like this: https://www.tattoosai.com/tattoos/62ecce4651c8873ccf2bbe67
The pricing model of a one time payment is also quite weird... I get that this is a quick project so it's not polished, but as I would not start charging for it.