Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

OpenAI's new open-source model is basically Phi-5

seangoedecke.com

Post reply on HN
Print view View original

71–80 of 233 posts

Re: OpenAI's new open-source model is basically Phi-5

#71

Post by landl0rd »

kristopolous wrote: post #65
slt2021 wrote: post #58

Earlier quoted context omitted.

even if it is victim free, it can affect mental health in a way that a consumer will be more compelled to do a criminal act and create a real victim. let's say you publish a Steam game how to be a school shooter and shoot kids, wouldn't that lead to real school shootings ? who can definitely say that computer generated content about criminal behavior, won't lead to real crime with real victims? https://en.wikipedia.o…

I view it more like methadone. Let's be specific: Rape, incest, necrophilia, bestiality, and pedophila ideation. I think we can all agree (1) these are harmful, anti-social behaviors that we do not want in our society, (2) people don't choose to have these desires, (3) most people who have them have no desire to actually traumatize others, (4) people who have these struggle with it. These multi-media AI role-play env…

It’s not unreasonable to suspect that engaging in high-fidelity simulations of these behaviors will further entrench and worsen paraphilias. This is pretty evident with the progression of many pornography addictions that don’t include these sorts of things that still follow the pattern of increasing novelty seeking leading to increasingly deviant stuff.

I am at a principled level uneasy with what’s fundamentally a sort of prior restraint (you haven’t yet hurt anyone but this may increase the likelihood and/or be an effective proxy to lock up those who are more likely to do so) but also see a really strong case for doing it given the fact that these are arguably the most antisocial behaviors one can imagine.

Re: OpenAI's new open-source model is basically Phi-5

#72

Post by landl0rd »

jahsome wrote: post #68
slt2021 wrote: post #58

Earlier quoted context omitted.

even if it is victim free, it can affect mental health in a way that a consumer will be more compelled to do a criminal act and create a real victim. let's say you publish a Steam game how to be a school shooter and shoot kids, wouldn't that lead to real school shootings ? who can definitely say that computer generated content about criminal behavior, won't lead to real crime with real victims? https://en.wikipedia.o…

What about two consenting adults engaging in age play. By your logic, wouldn't that also lead to "real crime"?

There’s probably some difference between someone who is visibly an adult and mature vs more deeply entrenching pathways of arousal in response to someone who is visibly a child. I still find the adult fetish version repellent but it’s also really hard to police in a way that’s remotely ethically permissible.

Ie yes it’s bad and in an ideal world nobody would do it. I see trying to restrict or ban it as the greater of two evils.

Re: OpenAI's new open-source model is basically Phi-5

#73

Post by mh- »

michaelt wrote: post #59
anothernewdude wrote: post #46

Earlier quoted context omitted.

My use case has been trying to remove the damn "apologies for this" and extraneous language that just waste tokens for no reason. GPT has always always always been so quick to waffle. And removing the chat interface as much as possible. Many benchmarks are better with text completion models, but they keep insisting on this horrible interface for their models. Fine tuning is there to ensure you get the output format y…

The jargon to google here is "length bias" It turns out if you generate two LLM responses and ask a judge to choose which is better, many judges have a bias in favour of long answers full of waffle.

Thanks for that pointer.

The abstract of this paper seems interesting: https://arxiv.org/html/2407.01085v3

> use of [LLMs] as judges [..] reveals a notable bias towards longer responses, undermining the reliability of such evaluations. To better understand such bias, we propose to decompose the preference evaluation metric, specifically the win rate, into two key components: desirability and information mass [..]

(If you're interested, give it a click. I tried to pare this down to avoid quoting a wall of text.)

Re: OpenAI's new open-source model is basically Phi-5

#74

Post by lmm »

charcircuit wrote: post #21
jononor wrote: post #3

Earlier quoted context omitted.

No the preferred way of making modifications is the weights _together_ with training (or fine tuning) scripts, and the entire evaluation pipeline to measure performance. And the data required to support all of this. When someone joins your data science team your would give them all this code and data. Not just the weights and say - the weights are the source, modify that to improve the model, I look forward to see yo…

It's not about the preferred way. Else open source software would need to give you their IDE setup, CI/CD setup, access to all internal tools, etc. Software like sqlite don't release their full test suite. They paywall the preferred way of making changes, yet they are open source. >The “source code” for a work means the preferred form of the work for making modifications The GPL refers to a form of the artifact being…

> open source software would need to give you their IDE setup, CI/CD setup, access to all internal tools, etc.

IMO they do. If you can't modify it like a core contributor would, then it's not really open source. Traditional open source projects always included development guides, test configurations etc.

> Software like sqlite don't release their full test suite. They paywall the preferred way of making changes, yet they are open source.

That's a matter of opinion. IMO sqlite is not true open source, for precisely this reason.

Re: OpenAI's new open-source model is basically Phi-5

#75

Post by kristopolous »

landl0rd wrote: post #71
kristopolous wrote: post #65

Earlier quoted context omitted.

I view it more like methadone. Let's be specific: Rape, incest, necrophilia, bestiality, and pedophila ideation. I think we can all agree (1) these are harmful, anti-social behaviors that we do not want in our society, (2) people don't choose to have these desires, (3) most people who have them have no desire to actually traumatize others, (4) people who have these struggle with it. These multi-media AI role-play env…

It’s not unreasonable to suspect that engaging in high-fidelity simulations of these behaviors will further entrench and worsen paraphilias. This is pretty evident with the progression of many pornography addictions that don’t include these sorts of things that still follow the pattern of increasing novelty seeking leading to increasingly deviant stuff. I am at a principled level uneasy with what’s fundamentally a so…

Right, I'm just a technologist. The psychological and sociological parts aren't my bailiwick

Typing a prompt in an AI box to make art has fewer real-world victims than performing the acts, filming them, and then sharing the videos.

I think that's inarguable. Maybe it's still unadvisable and someone should be in talk therapy. I have no idea. But at least nobody is actually getting molested and retraumatized in the ai art scenario.

If someone is spending their time using comfyUI drawing pictures instead of stalking the local middleschool, I'd hesitate to say mission accomplished ... but maybe I should?

People's time is finite. They can't be doing both. If the real is substituted for the imaginary then the real can no longer happen because that time is spent.

Re: OpenAI's new open-source model is basically Phi-5

#76

Post by School-Cotton »

landl0rd wrote: post #71
kristopolous wrote: post #65

Earlier quoted context omitted.

I view it more like methadone. Let's be specific: Rape, incest, necrophilia, bestiality, and pedophila ideation. I think we can all agree (1) these are harmful, anti-social behaviors that we do not want in our society, (2) people don't choose to have these desires, (3) most people who have them have no desire to actually traumatize others, (4) people who have these struggle with it. These multi-media AI role-play env…

It’s not unreasonable to suspect that engaging in high-fidelity simulations of these behaviors will further entrench and worsen paraphilias. This is pretty evident with the progression of many pornography addictions that don’t include these sorts of things that still follow the pattern of increasing novelty seeking leading to increasingly deviant stuff. I am at a principled level uneasy with what’s fundamentally a so…

Are there any actual studies on this? Does access to simulations of illegal or objectionable material make pedophiles, rape fetishists, etc. more or less likely to try to access the real thing (or even worse, to try to commit crimes in the real world)?

Because both possibilities are plausible, it’s hard to know which is correct.

Re: OpenAI's new open-source model is basically Phi-5

#77

Post by jameslk »

slt2021 wrote: post #58
kristopolous wrote: post #25

Earlier quoted context omitted.

Porn is always the frontier. It's a well-understood self-contained use-case without many externalities and simple business models. What more, with porn, the medium is the product probably more than the content. Having it on home-media in the 80s was the selling point. Getting it over the 1-900 phone lines or accessing it over the internet ... these were arguably the actual product. It might have been a driver of earl…

even if it is victim free, it can affect mental health in a way that a consumer will be more compelled to do a criminal act and create a real victim. let's say you publish a Steam game how to be a school shooter and shoot kids, wouldn't that lead to real school shootings ? who can definitely say that computer generated content about criminal behavior, won't lead to real crime with real victims? https://en.wikipedia.o…

> let's say you publish a Steam game how to be a school shooter and shoot kids, wouldn't that lead to real school shootings ?

> who can definitely say that computer generated content about criminal behavior, won't lead to real crime with real victims?

I can’t tell if you’re being sarcastic but there has been no found link between violent video games to violent crimes, despite it being researched extensively:

https://www.apa.org/news/press/releases/2020/03/violent-vide...

https://link.springer.com/article/10.1007/s10964-019-01069-0

https://pmc.ncbi.nlm.nih.gov/articles/PMC6756088/

https://elifesciences.org/articles/84951

Of course, that hasn’t stopped video games being blamed for violence by the “think of the children” crowd and certain politicians:

https://en.m.wikipedia.org/wiki/Family_Entertainment_Protect...

https://www.theatlantic.com/technology/archive/2019/08/video...

Especially when shootings occur by white perpetrators:

https://www.apa.org/news/press/releases/2019/09/video-games-...

The same narrative plays out for porn, despite the research findings being the same:

https://www.utsa.edu/today/2020/08/story/pornography-sex-cri...

But blaming violent video games or pornography is an easy scapegoat

Re: OpenAI's new open-source model is basically Phi-5

#80

Post by torton »

slt2021 wrote: post #58
kristopolous wrote: post #25

Earlier quoted context omitted.

Porn is always the frontier. It's a well-understood self-contained use-case without many externalities and simple business models. What more, with porn, the medium is the product probably more than the content. Having it on home-media in the 80s was the selling point. Getting it over the 1-900 phone lines or accessing it over the internet ... these were arguably the actual product. It might have been a driver of earl…

even if it is victim free, it can affect mental health in a way that a consumer will be more compelled to do a criminal act and create a real victim. let's say you publish a Steam game how to be a school shooter and shoot kids, wouldn't that lead to real school shootings ? who can definitely say that computer generated content about criminal behavior, won't lead to real crime with real victims? https://en.wikipedia.o…

Grand Theft Auto 5 sold over 200 million copies, and military/crime/shooter games have always been incredibly popular. Yet crime has been decreasing over the past few decades in the United States, where both cars and guns are easily accessible.
Post reply on HN