I’ve tried coding and every one I’ve tried fails unless really, really basic small functions like what you learn as a newbie compared to say 4o mini that can spit out more sensible stuff that works.
I’ve tried explanations and they just regurgitate sentences that can be irrelevant, wrong, or get stuck in a loop.
So. what can I actually use a small LLM for? Which ones? I ask because I have an old laptop and the GPU can’t really handle anything above 4B in a timely manner. 8B is about 1 t/s!
Let’s take a look at the Developer Agreement that you cited:
This very clearly states that you are disallowed from retaining chat logs for the general purpose of collecting information about Twitch’s end users.
You said that you, “store ‘facts’ about specific users so that they can be referenced quickly,” but then later in a different thread state, “I’m not storing their data. I’m feeding it to an LLM which infers things and storing that data.” You’re retrieving information about specific users at a later time. You’ve built a database of structureless PII from chat logs. You’ve chosen to store the data as inferences, which makes it a bad database, but still a database.
I have questions:
When your streamer mentions something deeply personal, like, “how their mothers surgery went,” that your tool helped them remember, do they disclose that your tool was involved in that transaction? When the viewer gets weirded out and asks your streamer to not mention that again, or forget it entirely, do you have a way to remove that information from your database and a way to prove it’s been deleted? When other people in chat think it’s gross, and ask to opt-out, can you even do it?
Regarding FrostyTools: I don’t think it’s storing the chat logs for a later time. They don’t have a data retention section in their TOS or Privacy Policy that isn’t related to the streamer. (As in, they hold on to the streamer’s Twitch account and some other information for billing, authentication, etc.) I think it’s taking the chat logs only for as long as it needs to output a response and then deleting it. Also, this excerpt from the FrostyTools TOS made me chuckle:
This leads me to believe that you can violate the Twitch TOS quoted above using FrostyTools. It is apparent that FrostyTools has positioned itself as an application that creates User Generated Content (like Photoshop or Word).
I’m not storing chat logs.
Not creating any kind of public database either. It’s a private tool. Its purpose isn’t to massively-collect data about all of twitch either - it’s to provide reminders for social situations. If anything, it’s an accessibility tool for the disabled.
Again - Not storing chat logs. They are processed for information and that information inferred. I am storing reminders for the twitch streamer to talk about a certain subject at a certain time. If I put a reminder in my phone to remember to tell you happy birthday because I saw it on twitch; am I “creating a database of user information”? No. I’m creating a reminder for myself to remember to say happy birthday.
Having a computer help me remember those things isn’t a violation. Hell, even something like Microsoft’s new AI in windows does the same thing - are THEY violating twitch TOS when you have a browser window open? The answer is no.
No, nor should they be required to.
When they mention not wanting to talk about something, that’s listed as something they don’t like to talk about, so in a way, yes.
Additionally, I instruct the ‘agent’ to disregard anything political or religious. - Though so far it’s not very good at distinguishing those things. Additionally it’s easy to feed it false information though it usually fixes it over time.