I uploaded 11 years of my Day One Journal data to GPT-4o | Hacker News Reader

I uploaded 11 years of my Day One Journal data to GPT-4o

linkedin.com

34 pointsindigodaddy25 comments

I don't understand quite how this works. GPT-4o's context limit is still just 128,000 tokens, but this post talks about "97k hours of transcribed audio".

Did this person upload the content to a newly created GPT, and is now running RAG queries against it using the GPT "knowledge" feature? https://help.openai.com/en/articles/8554397-creating-a-gpt says "GPTs for paid users are currently powered by GPT-4. GPTs for paid users will switch to use GPT-4o in the next several weeks." but maybe that documentation is out of date?

The icon displayed in the screenshots is the default ChatGPT icon though - if it was a GPT I would expect either a custom icon or an isometric box default GPT icon.

The text in the screenshots looks to me like it could be generated based on a shorter interactive conversation (providing names and basic notes about family members) that fit in that 128,000 token window, as opposed to being influenced by 97,000 hours of audio.

Seems genuine, but it's so weird to expose your personal family matters like this. I always wonder how I'd feel if my parents were bragging about how good they were to me on social media as they would parent me. It feels like it cheapens things because now it's not just about the family, it's also because you want clicks on your profile.

As a fairly private person, this seems insane to me. Though this does seem very par for the course on LinkedIn.

I'm grateful I didn't grow up in the social media era. I don't think my particular parents would've posted our family's dirty laundry, but it just feels wrong. If you want to discuss these matters to others, why not do it in person (or in direct messages) rather than publicly for the whole world to see?

ML is an interesting technology. AI is a cult. Some people are in really deep.

Before the current phase there was certainly a radical transparency cult. See “The Circle” which wasn’t as obviously satirical as it probably should have been. But also lifestreaming etc.

I don't know if that was as asinine, all-pervasive and exhausting as the AI one. Altman bought millions of dollars of video game cards to make a website that can invent sentences by parsing hundreds of years of human work to guesstimate what passes for expected, and now each time I go to a meeting like "Algo trading, onboarding client X" in my company just yday, a frigging asshole always asks: "And AI, have you thought of AI to trade algorithmically". And the head of algo, a PhD in mathematics, C++ developer, actual money maker, sigh one more time "yes, we're studying what it can do ofc, since it's the future of everything".

The asshole asking that question is always the same: optimal level of incompetence, insecure about his position, aggressive with anyone "resisting change" when we try to save him from himself, and gone within 2 years. Sigh, the time we lose, and as I grow older and navigate those hypes, I feel half of humanity is some sort of social scamming parasite. I just am too shy to try and be one myself.

I enjoyed your comment

And Altman is a disgusting guru, it's so painfully obvious but somehow he gets the same amount of pushback as Enron "Electric Jesus" Musk, but for half the accomplishment: at least Enron built cars, lots of cars. Shitty cars, but cars.

Mr Altman (whom everyone calls Sam like he's their friendly post-human overload) is just a cringe VC scammer, not very sure why he's there in the first place, sending "Her" to create some hype 2 days before being called out like a pervert by an actress :D

It's hard to see a train going full speed towards a wall and everyone congratulating the pilot for his ability to go that fast, sigh.

To be fair to LinkedIn, everyone knows it's not expected to be real stories... it's just fun and silly to post random corporate drone bullshit and see people jokingly reply "you inspired me today to thank my boss" or some such.

Anyone who think anything on LinkedIn is not a comedy is missing half of what makes humanity so awesome I think :D They'd be the same who forbid alcohol or dancing and take everything at face value :D

This is the phase we are in now: they are rolling out the half-baked tech, its only real use might be private data collection for advertising, so its time to start getting people real comfortable with just sending every shred of audio and video you have to them

I would not be happy about it.

some people like to overshare

> smoking crack in a faculty office at Harvard’s Kennedy School — or in an airplane at 30,000 feet. Or stealing a car. Or having sex on a beach in Israel with a mistress and attracting the attention of the Israel Defense Forces. Or later being arrested and charged with assaulting her. Or cuckolding a best friend.

LATE ADMISSIONS: Confessions of a Black Conservative, by Glenn Loury

I would not even share personal matters like those with an AI.

Very cool. I hope to represent my mental states in text as well as I can so that, should I die untimely, my children can extract whatever value they can from my serialized brain as they can using AI. Writing it down is step one. I keep as much as possible public so that people other than my children can also benefit. But there's obviously things that are best for my kids. Hopefully they get enough of me that they can query my ghost after my death. Or maybe they won't care. And that's okay too!

I would be very intrigued to interact with a GPT-level model that had access to that much of my personal data, but I would never give that kind of data to a company that didn’t have some kind of legally-binding fiduciary-type privacy obligation to me…

Which we should probably be pushing for from these companies that have so much of our data.

Untreated mental illness.

Cool, so now, with just a little clever “prompt engineering” I have access to 11 years of this guys most personal memories?

And now OpenAI knows everything about you and your family.

Well, at least OpenAI seems like an upfront, honest company full of integrity. Don't see any privacy concerns here. /s

> upfront, honest company full of integrity

they need to! as AI will be the future of parenting, schools, books etc.! ^_^

Permanently

I'm personally confident that OpenAI doesn't permanently "remember" every piece of text it that is fed into it.

Unfortunately, because OpenAI are extremely vague about how they use input to their models to influence future training runs I can't distribute my own personal confidence level here to other people.

I believe they retain every "conversation" in your account, purely for your convenience (not for training), manually review exchanges that flag their moderation filters, and retain replies that you press the feedback buttons on (like/dislike).

The moderation filters are (or at least were back in 2022) activated by the client, which is hilarious.

They have two things:

A toggle for using the data for training, with it off it is not used for further training.

A temporary chat, which is not used for training and is deleted after 30 days.

They could be lying about either, but you either trust their T&Cs or you don’t.

> 97k Hours of transcribed audio

Holy crap.

Reply on news.ycombinator.com