2,593 karma · joined September 29, 2015
Facebook mining and profiting from harvesting political user data shows a direct conflict of interest between keep data private and making money. When Facebook is used as a platform by hate groups, I can't see how they would meaningfully profit from it. Saying that encouraging or ignoring it would increase userbase is very short sighted, at the bare minimum it would caused the group being targeted to leave.
These situations feel more like the debate of whether ISPs are a public utility and the EARN IT act and section 230. Holding Facebook 100% accountable would set a very dangerous precedent. Facebook is not an entity and without showing hateful posts/comments were highlighted while alternatives were not, this feels like an argument against math.
Popular posts reflect the interests of the vocal population. Assigning morality to the math used in an algorithm is a failure to understand the purpose or function of it. "Inciting" would means the algorithm promoted a hateful post despite the underlying signals saying it shouldn't be promoted.
https://www.culawreview.org/journal/extending-corporate-tort...
Queuing videos is okay and when streamers have to take a break to fix equipment or use the restroom they keep it up. As long as chat is actively moderated, you can do anything for the most part. I don't know if there's an unwritten rule about having to show yourself on screen at least once though. I imagine a lot less people would show up or stay if they couldn't see hear or see the streamer.
There's things like study with me streamers where they just read a book or study in front of the camera. There's even sleeping streams where some people sleep on camera. Sometimes they are doing a 24/48 hours stream challenge.
Ludwig, a very well respected creator, did a subathon where each new sub extended the time he continuously streamed by a little. He ended up streaming for 30 days.
I think prerecorded is the only real taboo culturally. There's streams where a creator will say, chat send me things to watch, and they just watch with chat for hours and react. There was a time during which streamers were pushing the limits of DMCA striking and some streams watched seasons of master chef and other popular shows on stream.
This article does a poor job exploring the nuance of the situation or even the diversity of content people make.
There's two situation that really highlight the uniqueness of the relationship between the audience and creators. One is the category of streamers that has been on the rise in the last few years, IRL streaming. The other is the concept of a sub-a-thon. I am surprised neither were brought up in the article.
For most people, being able to have a strong boundary between work and free time is necessary for the sake of mental health. My favorite content creator is an IRL streamer. She takes a camera out with her during the day and visits different destinations, usually spending a couple of weeks at locations. The amount of hours a streamer requires is pre determined by a contract with Twitch. There have been days where she leaves for the day at 4 pm and goes around places, like in a travel show, but essentially every minute is live streamed with an interactive chat. Some streamers talk to the entity that is chat like they would a friend, asking "am I going the right way?". The continued amount of time spent experiencing things together blur the line for whether it's actually parasocial. It's one thing when you hear a podcast and then imagine you are friends with a creator because you know a little about their history. It doesn't feel as one sided when the creator reads out and responds to your messages frequently, enough to where the creator remembers and requests for viewers by screen name.
This creator also has a discord and interacts there very frequently and has watch parties. There's channels to share food you ate or stream ideas. The creator engages with the discord like anyone else would, they also happen to be the content people make memes with. There's some validation they feel from their community, and the thing is most times the community responds to fishing for validation by being contrarian regardless of context. I enjoy this streamer's content because it feels like the way in which we process the world is similar. This understanding came from the creator being vulnerable with chat and opening up in a way you would a friend. Some creators do put on a persona, but some don't.
In that situation I don't know what it would mean to log off. I didn't understand initially, but it's clear that leans on their community strongly. The creator feels a sense of obligation to finish hours, but they seem to appreciate and enjoy their community and not feel pressured by it. This creator feels no guilt about delaying a stream or not follow her own schedule she has previously shared with chat. And I am glad, because I think it's a good way to be healthy around not having people build up expectations. A part of the interaction is still transactional, and with part of the twitch audience still maturing emotionally, donating or buying a monthly subscription to a creator can make people feel some ownership or entitlement.
When creators frame it as, only spend money on me if you have money to spend, and the money will not change how I see you, keep away building unhealthy dependencies. Not all creators can maintain that kind of distance, and sometimes there's nothing the creator creator can do.
Amourath, in the last week, received a gift of 70,000$ in cash along with some self defense tools and a prepaid bodyguard service. Unlike most similar stories, that's where the story ends. There were no conditions or strings attached to gift. I don't know how many people there are in the world that can drop 70,000$ on a person without it affecting their life. In this situation, I am not sure what the creator can do, it's not something she requested or wanted. She said she was grateful and hasn't contacted the person although they did leave a number, there was nothing dangerous about accepting the gift that would dox her. She wasn't even sure if she wanted to keep the gifts when she initially received it. The only reasonable expectation of Amourath is that she refuse gifts if she finds out it came at an unreasonable cost to the person who chose to gift.
The other thing is subathons and the apparent cause of burnout. Subathons are how they sound, a creator sets a time to extend their stream by for every new subscriber they get. Usually it's a few seconds. Ludwig, a very prolific creator ended up streaming for 30 days straight. He netted 1.4 million$, a large amount of which he donated or paid his team of moderators who worked to manage chat during the subathon. He even slept on stream, his moderators would make content for chat to engage with during that time.
Ludwig has spoken about how fondly he enjoyed doing it, he didn't really need the money and he potentially would have done it for longer if it wouldn't lead to negative consequences with friends and family. He is successful because he loves the work.
Content creation is often a lot of work for very little rewards, the people who end up successful are the one who would have done it for free. Burnout isn't necessarily about the platform itself. Any situation where a passionate person can obsess about their craft or where it can get repetitive can lead to burnout. The creator takes a break by making content or a company or doing work somewhere else. It very much depends on the person, and it can be argued that there isn't a great support system for content creators right now, but it's not a Twitch issue since most creators make their own schedule.
Twitch is absolutely is real work. Successful creators I've met usually end up apologizing if we spend more than a couple hours together saying, "sorry, it's hard to turn my brain off from thinking about new ideas".
Taking these things together with the facts in the article lead to a differnt final conclusion, to me.
The goal is different so it's not a direct alternative, but it's the closest that can be run on consumer hardware. In some ways models that run via Kobold are a lot more practically useful because they have a semblance of memory, but they don't have the breadth of knowledge that ChatGPT does. You can go back and edit a response and models will be able to help generate pages and pages of writing. It's also not limited to chat mode.
I run it locally but I have heard a lot of success stories about running on Collab as well and there are GPU and TPU notebooks maintained in the repo. It can use a lor of different models like OPT, Fairseq Dense and even older models like GPT-J. There are fine tuned models for NSFW content as well, my understanding is that those models were motivated to move away from AI Dungeon that censored and read the text of user's stories.
The models on hugging face come in all sizes, 8bg VRAM is enough for a 2.7B model, without taking a time penalty and splitting compute across GPU and RAM+CPU. There are options ranging from 350M [1] to a OPT 66B in the FB/Meta AI repo released on May 3rd [2], the 66B parameter one is openly available and the full 175B parameter model is available via request. GPT2 and GPT-Neo are supported too. I found 2.7B and 6.7B impressive personally. 66B Models would take over hundreds of gigabytes of vram to run at a similar speed to how ChatGPT works. I think that's part of the reason why it's not as popular, very few people can run even the smaller model at a reasonable speed.
OPT is motivated by only training on open access material. The example bias demonstrated about the how the model complete "the man works as a" vs "the woman works as a". I thought having more modern works to train on would help avoid strong biases because culture has shifted toward being more moderate, but that doesn't seem to be case.
"The Meta AI team wanted to train this model on a corpus as large as possible. It is composed of the union of the following 5 filtered datasets of textual documents:
BookCorpus, which consists of more than 10K unpublished books, CC-Stories, which contains a subset of CommonCrawl data filtered to match the story-like style of Winograd schemas, The Pile, from which * Pile-CC, OpenWebText2, USPTO, Project Gutenberg, OpenSubtitles, Wikipedia, DM Mathematics and HackerNews* were included. Pushshift.io Reddit dataset that was developed in Baumgartner et al. (2020) and processed in Roller et al. (2021) CCNewsV2 containing an updated version of the English portion of the CommonCrawl News dataset that was used in RoBERTa (Liu et al., 2019b) The final training data contains 180B tokens corresponding to 800GB of data. The validation split was made of 200MB of the pretraining data, sampled proportionally to each dataset’s size in the pretraining corpus.
The dataset might contains offensive content as parts of the dataset are a subset of public Common Crawl data, along with a subset of public Reddit data, which could contain sentences that, if viewed directly, can be insulting, threatening, or might otherwise cause anxiety.“
[1] https://huggingface.co/KoboldAI
[2] https://huggingface.co/facebook/opt-66b
https://ai.facebook.com/blog/democratizing-access-to-large-s...
Amazing how much it was able to get right while coming out in the same year consumer did, and the philosophical questions it asked continue to be relevant. I rewatch it every year. It's available for free with ads on Youtube, in Japanese with subtitles, which is the recommended language to watch in. There have been many analysis written online that I recommend reading into after finishing the movie. It's a little slow and gives the viewer time to think. There's a independent show with the same characters that is sort related but not a sequel in any way. The 2017 remake is not a good reflection of the story nor does it stick to the original story in any way.
https://www.youtube.com/watch?v=iHil4Y4r3Wk
https://en.wikipedia.org/wiki/Ghost_in_the_Shell_(1995_film)
Chat With Traders Podcast has Quant traders and software engineers from time to time. Hearing technical people talk about work and answer questions to a non-technical audience is useful, because podcasts often don't have PR talk. There are exception to that, here's a 3 year old episode with Sam Bankman-Fried https://www.youtube.com/watch?v=gSDk5PAJss4 https://chatwithtraders.com/?s=software
DataSkeptic is an AI, data science, stats, advertising podcast I have found relatively recently. The have had multiple seasons, Seasons labeled, Ad-Tech (current), Physically Distributed, k-means, Time Series, Consensus, Interpretability, Fake News, Natural Language Processing, and Artificial Intelligence. I really liked the episodes I have heard so far. https://dataskeptic.com/
I also enjoy old developer conference presentations like ones from Game Developers Conference and Defcon. There are projects to put saved articles through text to speech in batches on the computer, but I have set up classic TTS on my phone instead. Not all articles will work well for conversion though.
My experience is that a, sometimes complicated, series of button presses for a specific amount of times resets the counter and it continues printing like usual. For the model I have the reset procedure is just holding the power button while turning it on for a specific number of seconds.
There might have been an update to add iOS printing to it via some software bridge, but I never checked. There are multiple tutorials online about setting up a Raspberry Pi (even a zero w) as an AirPrint server and interfacing with the printer via CUPS. This allows using a lot older printer models as well that don't have built-in wifi.
https://www.tomshardware.com/how-to/raspberry-pi-print-serve...
https://github.com/w3c/webauthn/issues/1640
The fact that passkeys are supposed to replace password and TOTP feels like a security risk. I understand that the phone uses a security enclave or TPM along with a pin/pattern/biometrics to attest a valid 2FA unlock, having it be tied to a mobile phone feels incorrect. I understand that's how it's set up for the vast amount of people already, but shifting from potentially separate devices for TOTP and Pass to a single device for everything seems weaker.
I understand that power users are not the target, but the issue of vendor/device lock based on attestation and not allowing DIY keys. Moving from a piece of information I know like a password, or even an SSH key/cert I print out, to a "password" of a private key even I am not allowed to read feels less secure. It's not a matter of convenience, it's a question of giving up control of key generation and syncing to a third party. From a standpoint of opsec, not having control of your private keys is a much bigger weakness relative to a compromised password. Changing passwords is trivial, but changing a device secured private key is impossible.
Passkeys don't fix the bribes or 5$ wrench loophole [1]. Unless you force users to only use biometrics and pin/patterns are disabled, passkeys will have the same issues that passwords do. In a world where getting people to understand 2FA is hard, it feels short sighted to depend on Bluetooth protocol dependent authentication. Trying to get people to shift from the convenience of a password to the inconvenience of needing to mess with Bluetooth is going to be its own headache. When the topic of needing a superhuman memory to remember passwords comes up, I think of https://xkcd.com/936/ . The conspiracy theorist in me can imagine companies like Netflix forcing passkey reauth every 15 days in order to combat things like password sharing. I respectfully disagree and believe passkeys will serve mostly as an inconvenience for the world. I am struggling to see the use cases where passkeys are meaningfully better than passwords and 2FA unless the use passwords is completely discontinued. And phones would become an even bigger single point of failure. A broken phone would be the same as losing all of your 2FA keys, there would be no password to help demonstrate account ownership to support.
Until open source, free roaming, implementations of passkeys exist that expose the private key to the end user, it's effectively a locked-in ecosystem. I still haven't seen a good reason to replace tried and tested pass+2fa with passkeys. A simple push to force non-sms 2FA is an existing solution to the problem passkeys is attempting to solve. A compromised password is much less valuable now than it was even a few years ago. Even if you miss when a password was compromised, companies that care about security have started sending those "a new login was detected" emails that would alert you of any concerning activity.
I have heard stories of people not being let in and having to contact friends who work at Facebook and have access to the internal support queue. I know my friend had to do something similar at Google and have a friend make a internal support ticket when they were locked out of a gmail account. That experience with Instagram is the reason I moved away from using DUO for everything to Aegis where I can copy a password encrypted backup of my 2FA secret keys like I would any file. I am grateful I learned that lesson by losing 2FA to a single account.
It seem more like how I have to enroll my device to Duo if I want to use it for 2 factor to log in and Duo using a proprietary implementation locking me into their software. As it stands, despite this being an open standard, Google has made it clear it has no plans to add support to linux for "Local user verification" or "Passkey sync".
The idea that Google will one day push users to only use passkeys and phase out passwords isn't unfounded. It's clear that Google would be motivated to make such a push. The reason people are concerned is because using Google ties the passkey to your Google login. From what I recall, Google also used a proprietary method to generator TOTP/HOTP making it impossible to use third party authenticators to log in to Google. Google Authenticator is open sourced but hasn't been updated in over 2 years. I believe Aegis supports Google Accounts so it can be used instead of Authenticator.
I understand this from the more basic SSH authentication using a generated public/private key. I am also aware of Yubikeys, which are also open, but I have chosen to continue using passwords because I want complete control over my private keys.
This is informed and healthy concern for a new implementation of security protocols.
The current standard of storing salted passwords makes the uniqueness requirement of passwords meaningless. If companies don't properly salt passwords, then uniqueness and length became an issue again. Salted passwords are effectively the same as appending an SSH key to a chosen password and letting websites manage them. With properly salted passwords, even if everyone chose the exact same password, a compromised password database would be meaningless.
This is the reason why NIST changed its guidelines about how often a password should be reset. It also changed the guidelines for complexity rules of passwords.
"Q-B05: Is password expiration no longer recommended? A-B05: SP 800-63B Section 5.1.1.2 paragraph 9 states:
“Verifiers SHOULD NOT require memorized secrets to be changed arbitrarily (e.g., periodically). However, verifiers SHALL force a change if there is evidence of compromise of the authenticator.”
Users tend to choose weaker memorized secrets when they know that they will have to change them in the near future. When those changes do occur, they often select a secret that is similar to their old memorized secret by applying a set of common transformations such as increasing a number in the password. This practice provides a false sense of security if any of the previous secrets has been compromised since attackers can apply these same common transformations. But if there is evidence that the memorized secret has been compromised, such as by a breach of the verifier’s hashed password database or observed fraudulent activity, subscribers should be required to change their memorized secrets. However, this event-based change should occur rarely, so that they are less motivated to choose a weak secret with the knowledge that it will only be used for a limited period of time.
Q-B06: Are password composition rules no longer recommended? A-B06: SP 800-63B Section 5.1.1.2 paragraph 9 recommends against the use of composition rules (e.g., requiring lower-case, upper-case, digits, and/or special characters) for memorized secrets. These rules provide less benefit than might be expected because users tend to use predictable methods for satisfying these requirements when imposed (e.g., appending a ! to a memorized secret when required to use a special character). The frustration they often face may also cause them to focus on minimally satisfying the requirements rather than devising a memorable but complex secret. Instead, a blacklist of common passwords prevents subscribers from choosing very common values that would be particularly vulnerable, especially to an online attack.
Composition rules also inadvertently encourage people to use the same password across multiple systems since they often result in passwords that are difficult for people to memorize."
I understand the assumption that the general population doesn't care to and shouldn't need to worry about managing their private keys. But there's nothing secure about not giving me the option to, if I want to.
https://developers.google.com/identity/passkeys/supported-en....
Additionally it also states that to access the passkey, you need to log in to the given Google account.
"Passkeys generated in Chrome on Android are stored in the Google Password Manager. These passkeys are available on all other Android devices as long as Google Password Manager is available and the same user's Google account is signed in."
https://developers.google.com/identity/passkeys/supported-en....
If only Google and Apple's implementations work, then it's not an open standard.