I have a lot of (legally acquired) shows and movies. I like making playlists with many shows that fit a theme (e.g. a cozy sitcoms playlist with MASH, Sanford and Son, Seinfeld, etc.). Some playlists have over 5000 episodes. I calculated one and it was almost a year of playtime if I only let it play while I'm sleeping (my primary use-case).
I want to shuffle the shows I'm watching, but I can't stand standard shuffling in VLC because it puts, for example, S8E4 before S1E1. Total loss of plot. So I wrote a shuffler that shuffles shows but keeps the episodes in order.
It deploys symlinks numbered 0-n, as well as an index of what episode each number corresponds to. Then I just slap the whole folder in VLC and let it play until I get bored of that theme.
I just make a copy of the script for each playlist and put it in the playlist folder. It has to be run as administrator on Windows, which is unfortunately where it's run at the moment because my gaming PC is also my TV for now, although I did just migrate everything to a Debian NFS server so I'll need to address that probably.
The US broadcast syndication market is really only set up to sell shows that have more than 100 episodes. I think UFO, Space:1999, and the 1955 Robin Hood were among the few shows seen on regular non-PBS stations. The Disney Show would sometimes run UK-filmed material. (Scarecrow, Thomasina) Partly it's because Americans can't understand the words when they hear many UK dialect accents, so they tune out those shows and that tanks the ratings. Even if you give a station a free tv show they lose money if no one watches it. We do better at remaking shows like All in the Family, Sanford and Son, Skins, and Being Human.
That's not the point of the Gilligan's Island test. If the LLMs had said something to the effect of "I don't know, I wasn't trained to answer trivia about television" then I would agree with your comment.
Instead they all confabulated! Either they made up an episode title or insisted the episode doesn't exist. That's a serious problem: if the LLM doesn't properly recognize that it doesn't know something, how will it properly recognize it needs to use a tool to obtain that knowledge?
It seems like the only answer we have is for a user to flag that GPT-4+Web flubbed the Gilligan test, OpenAI dispatches a data contractor for RLHF whack-a-mole + synthetic data generation, GPT learns to Google answers to Gilligan's Island prompts, then we cross our fingers and hope transformers are smart enough to transfer that knowledge to the Sanford and Son benchmark.
In some ways this is true, and in some ways it isn't. It's complicated.
You are correct that broadcasters back then were more free to transmit the content they wanted. You could get away with most vulgarities as long as they were after 10pm. Heck, I worked at a radio station in suburban New York where the morning DJs would regularly use words like "apeshit" and nobody cared.
Back then, if someone was offended they would write a letter to the FCC, and the FCC would take it seriously. For the most part, broadcasters genuinely worried about losing their licenses.
Since then, the FCC has declared that it, for the most part, has no authority to regulate the content on the airwaves. That gap has since been filled by political correctness, social media campaigns, and other organized artificial mob outrage. So the broadcasters are kept mostly in line by social pressure instead of the government.
That said, if you watch or listen to old broadcasts from the 1970's, it's amazing how free free speech was back then. Redd Foxx saying the N-word on the sitcom Sanford and Son. Candice Bergen calling Iranians "towel heads" on Saturday Night Live. Stuff that would bring out the lynch mobs today in our so-called "enlightened" times.
Hacker Cinema
5 comments on Hacker News
[url-redacted]
I have a lot of (legally acquired) shows and movies. I like making playlists with many shows that fit a theme (e.g. a cozy sitcoms playlist with MASH, Sanford and Son, Seinfeld, etc.). Some playlists have over 5000 episodes. I calculated one and it was almost a year of playtime if I only let it play while I'm sleeping (my primary use-case).
I want to shuffle the shows I'm watching, but I can't stand standard shuffling in VLC because it puts, for example, S8E4 before S1E1. Total loss of plot. So I wrote a shuffler that shuffles shows but keeps the episodes in order.
It deploys symlinks numbered 0-n, as well as an index of what episode each number corresponds to. Then I just slap the whole folder in VLC and let it play until I get bored of that theme.
I just make a copy of the script for each playlist and put it in the playlist folder. It has to be run as administrator on Windows, which is unfortunately where it's run at the moment because my gaming PC is also my TV for now, although I did just migrate everything to a Debian NFS server so I'll need to address that probably.
The US broadcast syndication market is really only set up to sell shows that have more than 100 episodes. I think UFO, Space:1999, and the 1955 Robin Hood were among the few shows seen on regular non-PBS stations. The Disney Show would sometimes run UK-filmed material. (Scarecrow, Thomasina) Partly it's because Americans can't understand the words when they hear many UK dialect accents, so they tune out those shows and that tanks the ratings. Even if you give a station a free tv show they lose money if no one watches it. We do better at remaking shows like All in the Family, Sanford and Son, Skins, and Being Human.
That's not the point of the Gilligan's Island test. If the LLMs had said something to the effect of "I don't know, I wasn't trained to answer trivia about television" then I would agree with your comment.
Instead they all confabulated! Either they made up an episode title or insisted the episode doesn't exist. That's a serious problem: if the LLM doesn't properly recognize that it doesn't know something, how will it properly recognize it needs to use a tool to obtain that knowledge?
It seems like the only answer we have is for a user to flag that GPT-4+Web flubbed the Gilligan test, OpenAI dispatches a data contractor for RLHF whack-a-mole + synthetic data generation, GPT learns to Google answers to Gilligan's Island prompts, then we cross our fingers and hope transformers are smart enough to transfer that knowledge to the Sanford and Son benchmark.
In some ways this is true, and in some ways it isn't. It's complicated.
You are correct that broadcasters back then were more free to transmit the content they wanted. You could get away with most vulgarities as long as they were after 10pm. Heck, I worked at a radio station in suburban New York where the morning DJs would regularly use words like "apeshit" and nobody cared.
Back then, if someone was offended they would write a letter to the FCC, and the FCC would take it seriously. For the most part, broadcasters genuinely worried about losing their licenses.
Since then, the FCC has declared that it, for the most part, has no authority to regulate the content on the airwaves. That gap has since been filled by political correctness, social media campaigns, and other organized artificial mob outrage. So the broadcasters are kept mostly in line by social pressure instead of the government.
That said, if you watch or listen to old broadcasts from the 1970's, it's amazing how free free speech was back then. Redd Foxx saying the N-word on the sitcom Sanford and Son. Candice Bergen calling Iranians "towel heads" on Saturday Night Live. Stuff that would bring out the lynch mobs today in our so-called "enlightened" times.
Reminds me of that Sanford and Son episode where he wore that to keep from getting sick.