HNHacker News
TopNewBestAskShowJobs

knaik94

2,593 karma · joined September 29, 2015

knaik1994 a.t gmail d(o)t com github knaik twitter @knaikk
submissionscomments
knaik94··on Advocating for Open Models in AI Oversight: Stability AI's Letter to U.S. Senate
I am failing to see how some of the suggested guidelines would help create meaingful oversight without also hurting new startups in the field of AI.

I am appaled to see the suggestion that cloud providers should report when a large model is trained on their servers, citing the terrible draft of EU AI act. This does nothing for the bigger risk of state actors and infringes on the privacy of users of cloud compute.

On the issue of Model guidelines, I agree with the sentiment to create safeguards, but I don't have faith the government can decide acceptable risk that respects users rights. It has shown very little meaningful understanding of technology many times, most recently during the TikTok hearing.

Advocating for holding companies directly accountable for what these models generate makes sense only in theory and ignores how the math behind these models work. There's no way to enforce content authenticity standards without also attacking section 230 in some way.

I am not suprised to see this sort of letter from a company that supported "pausing" AI development.

knaik94··on Apple previews Live Speech, Personal Voice, and more new accessibility features
One of the most useful accessibility features, as someone who doesn't require any, has been the live caption feature. I have no hearing or vision impairments, but I do struggle with ADHD. It was launched a while ago, along with audio detection for things like doorbells or alarms. I was hoping multi-language translations to be announced along with this. Currently things are still limited to English (U.S.) or (Canada) audio and text.

It's been so helpful to be able to read a short transcript of what was just said. The live caption feature works on video calls, and while multi-speaker captioning isn't perfect, it mostly works. I also really like how Android, iOS, and Windows all seem to keep feature parity among the operating systems. I wonder how Google and Microsoft will respond to this and I wonder how Personal Voice will work for communication outside of the Apple ecosystem.

knaik94··on Texas professor fails entire class from graduating- claiming they used ChatGTP
There is no complete theory, because having a complete theory would require a complete theory about cognition. But the general understanding is that horoscopes and astrology play into the fallacy of personal validation. People are bad at evaluating how generic certain descriptions are. It is a well researched and replicated result. The arguments supporting horoscopes and astrology are also written in a way to make them impossible to disprove, making them unscientific.

https://en.wikipedia.org/wiki/Barnum_effect

knaik94··on Texas professor fails entire class from graduating- claiming they used ChatGTP
We used digital textbooks, in my entire computer science curriculum, we were never forced to use a physical book. In the last few years, "renting" etextbooks from Amazon caught on. The only exception was a suggestion, not a requirement, to buy the ds algo book as a reference for the future. I can't imagine having to use a physical textbook, the find/control + f feature really speeds up learning.
knaik94··on PaLM 2 Technical Report [pdf]
I didn't mean to argue that everything generated is incorrect. But in my experience, the numbers it generates seem closer to random guesses. If you ask it enough times, it sometimes converges on a number, but I don't think that means it's an accurate value. I was able to make it generate a similar table for the different PaLM 2 sizes, and laMDA, and it listed, PaLM 2 Gecko 137 billion, PaLM 2 Otter 540 billion, PaLM 2 Bison 1.8 trillion, PaLM 2 Unicorn 5.4 trillion, LaMDA 137 billion. For Unicorn, it also lists "Still under development."

Edit: Playing around with it more and it listed WuDao 2.0 1.75 Trillion, Chinchilla 175B, Codex 175B, Dalle2 1.3B, GPT4 1.75T, GPT3.5 540B, GPT3 175B, GPT2 1.37B, GPT 1.3B.

But in the previous question it listed GPT4 540 billion and Codex 5.4 trillion among other contradictions.

knaik94··on PaLM 2 Technical Report [pdf]
It's very clear that the current Bard model is weaker than the largest PaLM 2 model. But for certain things, Bard seems worse than even the smallest model described. It's hard to say without someone doing a comprehensive benchmark, but the artifically limited context size makes testing useless for data.

The model was surprisingly confident when I tried to ask it about the relationship between better language comprehension and parameter size. The coherence displayed by the model, when it argued that a smaller model size will be capable of matching and surpassing competitive model performance, was a little jarring. Especially when, in the question right before, it said that the large PaLM 2 model has 540 trillion parameters.

knaik94··on PaLM 2 Technical Report [pdf]
It makes up those numbers, I asked about the difference between the small and large PaLM 2 data set size, and it asserted the small model was trained on 540 billion and the large model was trained on 540 trillion. A different draft instead specified 1.4 trillion for the large.
knaik94··on PaLM 2 Technical Report [pdf]
I asked if it's true that it's now using PaLM 3, as announced in Google I/O today, and it enthusiastically agreed. The previous question was asking the same question but with PaLM 2 and it agreed to that as well. I followed up asking about this discrepancy, and it said:

"I apologize for the confusion. I am still on PaLM 2. PaLM 3 is not yet available to the public. I am excited for the release of PaLM 3, and I hope that it will be a valuable tool for people all over the world."

My initial results are very disappointing. It's very strongly parroting information I give it, basically rephrasing my question and adding maybe a sentence worth of additional details. Sometimes, it does well, but I have no way to reproduce that kind of quality on demand. I feel it was conversationally better before any recent changes.

I understand that this is still beta, but for some questions, I already produce similar or better results locally. I also might be talking to PaLM 1 or even LaMDA, no way to confirm.

knaik94··on Ten Thousand Ways to Use Chopsticks
One thing I see most online guides not mention is the idea of changing grips depending on the food/task or how your hand is feeling. Certain grips take more energy than others, and different style of chopsticks require different grips too. Chinese cooking chopsticks are so versatile in the kitchen, but require a different set of techniques being 13 to 17 inches.

When using chopsticks to pick up medium sized things like popcorn, my grip is further back on the chopsticks and usually a standard grip because it feels less tiring and gives me more control. When I am picking up something very small or heavy, I often switch to something closer to the vulkan grip, because using the pads on my ring finger gives me more leverage, fine dexteriry, and a wider range of wrist motion. But it's also more tiring.

I learned how to use chopsticks as an adult, because my favorite place to eat lunch was a little korean restaruant. After a few weeks, I felt confident enough to ask one of the owners to critique my technique. He said I was as proficient as him and he found it humerous how I used them for the rice, the standard in Korea is to use a spoon.

I feel that for snacks and certain meals, the chopstick is more useful while eating the majority of it, but a spoon is better for getting the last bit. I started using chopsticks for snacking because I found it makes distracted eating harder.

knaik94··on Dose-response relationships of LSD-induced subjective experiences in humans
I really loved how this paper shared the r-syntax of the analysis, along with the consolidated data they used. The interactive webpage summarizing their results is fantastic. Loading the full charts of individual questionnaires took a little bit of time, but the way in which the results are displayed is exceptional. It was slower on mobile, but it was still useful.

https://dose-response-lsd.asdb.info/

knaik94··on Passkeys will be importable, exportable, cross-device, and across managers
It's not, until it is. There have been cases of everyday people are forced to turn over the passwords at borders when entering or leaving the country. I fundamentally disagree about whether or not this is a real concern for most people. I have no problem with people using biometrics, I understand people will always choose the most convenient option, that's why it's important. I think the iOS "cop mode" feature to disable the fingerprint with 5 presses of the power button shows companies are concerned about it too.
knaik94··on Tensegrity Wiki
On Aliexpress I found out some alternative lego compatible brick sets that take advantage of the Tensegrity principle. Here's a picture of the house from Up and a mini millenium falcon. [1][2] They were not the easiest sets to build and keeping the balance in the correct position is hard, but it's fun.

The tolerance for the lengths to maintain the correct tension is a lot tighter than I expected. I imagine it would be a little more forgiving if everything is scaled up. I saw some space related sets and floating boats from the anime One Piece as well on there. What is in and out of stock is not very consistent, and I haven't seen some of the sets I own in a while. I did not have to use glue but I did have to get creative in terms of wrapping things in a certain way, or connecting the things to an alternate hole. The manual was optimisitc.

There are some really interesting user created designs you can buy instructions for, but not every set from aliexpress is there.

https://rebrickable.com/mocs/?q=tensegrity

[1] https://twitter.com/Knaikk/status/1487889972816453635

[2] https://twitter.com/Knaikk/status/1655580332341489673

knaik94··on Passkeys will be importable, exportable, cross-device, and across managers
I think this addresses most of my concerns regarding passkeys.

The biggest remaining one is still that fingerprint locked information is not protected under the fifth amendment in the US. Even pin protected information/logins has been ruled as both protected and not protected by different courts. The "key vs combination" debate hasn't been decided one way or another. I can imagine a government push to not count fingerprint locked passkeys as equivalent to passwords, but that's not a weakness of passkey technology.

I am also interested in seeing how phone secured passkeys will work, iOS currently forces the use of keychain, excluding users who opt for a third party password manager like bitwarden. I also wonder if phones will be able to be used to login without enabling bluetooth.

https://www.concordlawschool.edu/blog/constitutional-law/fif...

knaik94··on Transcending My Father's Abuse
Sociopath is the wrong word to use, and that is the crux of the entire post. There's nothing so clear cut in this kind of situation. The parent isn't strictly trying to get something from the child with this apology. I think the parent is having their own internal struggle between validating the stuggles the child faced and understanding they were part of the cause. They lose sight of some things because they have an "ends justify the means" mindset. In a sense, they were okay being the "bad guy" so that their child was set up for the best future possible.
knaik94··on Transcending My Father's Abuse
The reason why I called it bad faith is because I am not seeing the possiblity of a positive follow up to your response to that question. In my experience, it's usually a way for them to make a rebuttal that shows a larger lack of sympathy. More than a few times the follow up to me acknowledging that I was likely difficult to raise was them using that to dismiss or invalidate the seriousness of situation. I am not sure why the default assumption someone would have is that you failed to see the situation from your parent's perspective. But I also recognize that I may just personally be tired of having to respond to that question.

If the discussion goes on long enough, one of the more interesting conclusions is the classical ethics question, what kind of net benefit will justify an evil act? Not really on topic for HN though.

https://fee.org/articles/why-the-ethics-of-would-you-kill-ba...

knaik94··on Transcending My Father's Abuse
For some people, hearing an apology means something even if you feel like your words are hollow and empty. I had a difficult upbringing too, but I have never repeated the mistakes of my parents in my own relationships. I wonder if that difference is related to how early someone realizes that their parents shouldn't be treating them this way. It's harder to recognize certain behaviors when you don't know what to look for.
knaik94··on Transcending My Father's Abuse
From my experience, people that respond with that kind of question aren't asking in good faith. They struggle to sympathize with the fact that parents can make mistakes and be wrong. I also don't agree with you that the implied question about victim blaming is sane to ask when the victim in question is a child, and the power imbalance is so large. I am under no illusions about how fun children are to raise. I can't say for sure what the intent behind the questions were because I didn't see them, but I personally do not give them the benefit of the doubt.
knaik94··on Transcending My Father's Abuse
The cycle of abuse is not unique to Asian parenting, unfortunately. I learned from some friends that Russian culture normalizes abusive behavior from parents just like Asian culture. I think the similarity might be being a morally conservative culture. Even coming from an Asian background, my cousin's parents who grew up in a city see the world very differently from the the way my parents do.

I'm not saying this to be contrarian, and I believe neither was the other person who responded. HN is not a place where people care about internet points.

https://www.undispatch.com/here-is-how-every-country-ranks-o...

knaik94··on Transcending My Father's Abuse
One of the hardest truths I've had to accept is that pointing things out to them doesn't help. I have spend a lot of time and energy trying to rationalize their thinking in some way, but I think it's a fool's errand.

I would similarly struggle to remember specific events, and at one point I even started voice recording them yelling. But showing them later only lead to more attempts at gaslighting and victim blaming. However, it was eye opening to see 3+ hour long runtimes on the voice memos that I couldn't remember anything about.

Emotional abuse is every bit as harmful and affects your brain in the same way as other kinds of abuse.

I hate people that try to apologize or convince you that they deserve forgiveness because they meant well or "they're your parents/family". It gets hard to push back without being disrespectful. And regardless of how old you get, the power dynamic of parent and child follows you. After a while you stop sharing completely.

Separating their behavior from who they are as a person is the hardest thing I've had to do. But it's the first milestone in rebuilding a relationship and helped me seperate myself from their behavior. Being aware that their behavior is a consequence of their upbringing earned empathy from me, but I think forgiveness requires empathy in both directions. I think it's equally hard to accept that you hurt someone you love.

Cultural norms can cause a lot of people to stay quiet, even if they decided to cut all contact, because talking poorly about your parents is seen as one of the worst things you can do. Regardless of what that thing was. I couldn't finish reading it in one sitting and I still had to skim parts of it. But I appreciate this being writted and shared publicly.

knaik94··on Releasing 3B and 7B RedPajama
I have been really impressed with the uncensored WizardLM I was playing with. Having a truely open uncensored model to work with is a really important research tool. Censorship of the training data and results in such a heavy handed way is not really possible without lowering the quality of all output.

As the resouces required to train and fine tune these models becomes consumer handware friendly, I think we'll see a shift towards a bunch of smaller models. Open models like these also mean the results of securty and capability research is publicly available. Models like this one and the Replit code model will become the new base all open source models are based on. I am really looking forward to the gptj 4bit, cuda optimized 7b models, the others I have tested run fast on 2070max q and 16gb ram, I was getting ~7tokens/second. Lora can work directly with 4bit quantized models. While ggml, cpu models are very strong, I don't believe we're move away from gpu accelarated training and fine tuning anytime soon.

knaik94··on BigCode Project Releases StarCoder: A 15B Code LLM
This basically has the same restrictions as LLAMA and stable-diffusion. Replit is licensed under cc-by-sa-4.0 which is truely permissive.
knaik94··on Passkeys: The beginning of the end of the password
There's no meaningful benefit of this over a password and 2FA authentication for users who are already proactive about security. I also doubt that passkeys will be protected by the fifth amendment the way passwords are, based on the "key vs combination" argument. While I trust Google password manager to help me remember passwords, I don't trust Google or any company to manage them for me.

At their core, passkeys are a easy and nice way to move from passwords to public private key cryptography. And hardware based authentication does allow more security. And by leveraging the hardware-backed keystore and security enclave, you can use your phone the way FIDO keys are already used. But there aren't clear benefits over 2FA and a password.

There are many reasons why I will wait as long as possible, maybe indefinitely before I start using passkeys. The idea of this being tied to Google or any other major company for my logins is not okay. There are many cases of people being locked out of Google accounts without the ability to appeal. With current 2factor, you can control the secrets yourself if you choose to. While it's true that you can't be phished into sharing your passkey as easily, you also lose a lot of convience and flexibility in login management. And it makes login sharing or multi-account management very inconvenient and difficult.

I was hopeful with 2FA rolled out that my accounts would be more secure, but most companies give you very little over which methods of 2FA are allowed or enabled. I don't want to be forced to enroll a phone number just to enable TOTP 2FA. I want to be able to choose between TOTP or HOTP for my accounts. I don't want to be forced to use a 2FA app that doesn't allow me to export and manage the secrets myself. In some ways FIDO keys solve some of those issues, but the hardware security aspect of it contradicts giving the end user the choice to self manage.

https://www.concordlawschool.edu/blog/constitutional-law/fif...

knaik94··on Google Introduces Passkey Authentication
There's no meaningful benefit of this over a password and 2FA authentication for users who are already proactive about security. I also doubt that passkeys will be protected by the fifth amendment the way passwords are, based on the "key vs combination" argument. While I trust Google password manager to help me remember passwords and even use sign in with google, I don't want to let Google or any company to manage my initial method of authentication for me.

At their core, passkeys are a easy and nice way to move from passwords to public private key cryptography. And hardware based authentication does allow more security. And by leveraging the tpm, hardware-backed keystore, or security enclave, you can use your phone and computer the way FIDO keys are already used.

But there are many reasons why I will wait as long as possible, maybe indefinitely before I start using passkeys. The idea of this being tied to Google or any other major company for my logins is not okay. There are many cases of people being locked out of Google accounts without the ability to appeal. I don't appreciate the idea of large companies being able to whitelist which passkeys are allowed, which is possible depending on how attestation certificates are handled. With current 2factor, you can control the secrets yourself if you choose to. While it's true that you can't be phished into sharing your passkey as easily, you also lose a lot of convience and flexibility in login management. And it makes login sharing or multi-account management very inconvenient and difficult.

I was hopeful with 2FA rolled out that my accounts would be more secure, but most companies give you very little control over which methods of 2FA are allowed or enabled. I don't want to be forced to enroll a phone number just to enable TOTP 2FA. I want to be able to choose between TOTP or HOTP for my accounts. I don't want to be forced to use a 2FA app that doesn't allow me to export and manage the secrets myself. In some ways FIDO keys solve some of those issues, but the hardware security aspect of it contradicts giving the end user the choice to self manage. I expect some policies will change over time based on uptake and issues people face.

https://www.concordlawschool.edu/blog/constitutional-law/fif...

Edit: Apparently, in iOS and Android, passkeys are backed up to the cloud, and iOS even allows you to airdrop them.

https://www.nytimes.com/wirecutter/blog/what-are-passkeys-an...

knaik94··on Google Introduces Passkey Authentication
A pin is not "something you know" legally. Biometrics are especially not.
knaik94··on What do historians lose with the decline of local news?
The internet is big enough and has been around for long enough where both complete loss and complete preservation exists. I am young enough to where I "discovered" the existance of news groups within the last year. More accurately, I discovered the google groups frontend to newsgroups. It's fascinating seeing discussions about things I find interesting from before I was born. Someone has cataloged and uploaded various K-mart, a now bankrupt retail chain, in-store tapes to archive.org . There's traing videos and tapes, and reel to reels going all the way back to 1947. I enjoy listening to the christmas mixes. The internet is not static, but with proper management and motivations, data will stay forever. My local library had a microfiche machine that they took out last year, I wish I got to try it before they did but for the last few years, there was a "out of order" sign on it. It wasn't really out of order, they just didn't want people to touch it.

On the other hand, myspace profiles and Cartoon Network shows are already good examples of losing data. Cartoon Network is especially unfortunate because some of the shows weren't ripped, so illegitimate sources won't help you obtain them. The creators of other shows publicly tweeted how they had to pirate their own show for their kids, because it wasn't available to stream.

I believe the internet is good, but I have also been realizing that I made bad assumptions about how companies will handle data.

https://archive.org/details/KmartDecember1990

knaik94··on What do historians lose with the decline of local news?
I also hate how emotionally charged the music has become in certain segments or transitions. Watching a story on mute with subtitles, and watching it with sound is a completely different experience. There has also been an increase in labelling everything "breaking news" and "top story", and the stories themselves have become sensationalized and rage baiting. The audio transitions into "breaking" stories prime you to feel a certain way even before hearing the story. I stopped seeing positive fluff, those time slots are replaced by things that feel like ad/sponsored segments or actual ads.

At one point I realized that watching the news left me emotionally exhuasted, so I stopped watching it. Surprisingly my older relatives have done the same in the last couple of years. When I asked, they also called it too exhausting to watch. They are anxious about using technology, they aren't getting information through a different medium instead. They told me, "if I need to know about it, someone will bring it up."

knaik94··on What do historians lose with the decline of local news?
My local news didn't cover those things either. But I did find that my high school athletics department made a youtube channel, and they have short interviews with coaches, specific games, and season highlights for certain sports. Another channel has been making highlight cuts for various high school sports. My county has a decent youtube channel that covers holidays, events, mail-in ballot instructions/voting information, and a short video about the "state of the county" executive summary. It links back to the county website. I found out the local historical society has been uploading documentaries online, about things like the constuction of the middle school. I found hour+ interviews/videos about people, who towns, roads and schools, were named after.

Right now a lot of journalism adjacent local reporting is buzzfeed style "list" videos where my town or county is mentioned. I am not sure where that falls, because it can highlight unique things about a place, but arguably not newsworthy. I don't think the incentives will ever there to cover local news in the way you are describing. On the other hand government twitter accounts have been genuinely useful for finding local events, my local news never did that for me.

knaik94··on What do historians lose with the decline of local news?
I agree, but I specifically meant in the context of local news, people are accessible in a way most weren't pre-broadband. Even the largest news organizations and presentations have started using zoom call recordings in segments. I agree with you that social media is terrible at meaningful filtering and aggregation. Verifying a source will always be a necessary part of journalism, regardless of medium.
knaik94··on What do historians lose with the decline of local news?
There is a positive trend of independent journalism on Youtube. Some of the my favorite stories were presented and investigated by a single person, Coffeezilla.
knaik94··on What do historians lose with the decline of local news?
I've noticed news sites have some of the most horrendous dark patterns and third party cookies I come across on the internet. Nothing frustrates me more than the video pop up on sites like CBS news. It makes me have a lot less sympathy for news organizations.
← PreviousPage 3 of 23Next →