4 ms·

One interesting weak point I've noticed with GPT is that if you ask it to give examples from a category which in the real world has hundreds or thousands of dif

One interesting weak point I've noticed with GPT is that if you ask it to give examples from a category which in the real world has hundreds or thousands of different possibilities, it will give you the same half dozen answers every time. You ask it to give you "opponents" and it will always give you male humans. You ask it to give you "monsters" to kill and it always gives you orcs, giant spiders, skeletons. If a chest is supposed to have "useful items", it's always a healing potion.

asddubs 4y ago
I guess it's giving you the most likely answer in the same way the most likely answer to the question "what color is the sky?" is blue, according to its training data.
swatcoder 4y ago
Yup. I think this is one of the places where the sheen will come off after a while. While temperature can mix things up a little, we’ll quickly start to see how the depth of LLM output is tied to the depth of LLM input — and people’s prompts are not all that varied and deep at the end of the day. We’ll start to notice signatures to each generation of model in its responses to tiringly common prompts — especially in stuff like blog spam, homework essays, email punchups, casual fiction, etc There are certain poles that it gravitates to that end up reading like verbal tics or lazy ideas, and those will be more obvious as we begin to get inundated with generated content. Careful prompt work can evade those poles and tics, but most people won’t have the skill or drive to bother with that.
- bgroat 4y ago
  >> Careful prompt work can evade those poles and tics, but most people won’t have the skill or drive to bother with that. I absolutely agree with this. Furthermore, I've come to believe that "careful prompt work" doesn't really save much time relative to just writing/coding the output I want. Especially after QA/edits
  - flangola7 4y ago
    > Furthermore, I've come to believe that "careful prompt work" doesn't really save much time relative to just writing/coding the output I want. The real magic happens when you train a better AI using those perfected prompts.
- dingclancy 4y ago
  There would be parameters to evade the obvious picks. temperature can actually go up to 2 in the API and you can also go use top_p < 1. but for the more out of this world stuff, this is where fine tuning will come in. The main model will always come in as vanilla anyway.
madelyn 4y ago
I think this is where something like the reAct framework would be super helpful. A human dungeon master doesn't pull a random enemy from the aether super well (or at least I don't!!). When I pick monsters I either logically think what makes sense for the area, pick something that's cool or vaguely humorous that I'm feeling the vibe of, or... use a random encounter table. For treasure and rewards... if it's not something hand picked, I'm again getting the loot tables out and rolling. And heck even for gold rewards and things, I'm rolling the dice. After trying basic stuff with langchain, an AI integrated with these types of tools is really promising. I think more systems on top could make it into a passable DM- at least until context becomes an issue.
capableweb 4y ago
If you set it up with some ground rules and rules around it's behavior (or more like guidelines), you can make it be more creative. Something like: "We're gonna start a text RPG game. Every animal we meet should be a made-up animal composed of joining two existing ones, with creative names being made for them. Don't use any usual fantasy tropes. Make encounters have surprising but logical outcomes" and so on.

asddubs 4y ago

I guess it's giving you the most likely answer in the same way the most likely answer to the question "what color is the sky?" is blue, according to its training data.

swatcoder 4y ago

Yup. I think this is one of the places where the sheen will come off after a while. While temperature can mix things up a little, we’ll quickly start to see how the depth of LLM output is tied to the depth of LLM input — and people’s prompts are not all that varied and deep at the end of the day. We’ll start to notice signatures to each generation of model in its responses to tiringly common prompts — especially in stuff like blog spam, homework essays, email punchups, casual fiction, etc There are certain poles that it gravitates to that end up reading like verbal tics or lazy ideas, and those will be more obvious as we begin to get inundated with generated content. Careful prompt work can evade those poles and tics, but most people won’t have the skill or drive to bother with that.

bgroat 4y ago
>> Careful prompt work can evade those poles and tics, but most people won’t have the skill or drive to bother with that. I absolutely agree with this. Furthermore, I've come to believe that "careful prompt work" doesn't really save much time relative to just writing/coding the output I want. Especially after QA/edits
- flangola7 4y ago
  > Furthermore, I've come to believe that "careful prompt work" doesn't really save much time relative to just writing/coding the output I want. The real magic happens when you train a better AI using those perfected prompts.
dingclancy 4y ago
There would be parameters to evade the obvious picks. temperature can actually go up to 2 in the API and you can also go use top_p < 1. but for the more out of this world stuff, this is where fine tuning will come in. The main model will always come in as vanilla anyway.

madelyn 4y ago

I think this is where something like the reAct framework would be super helpful. A human dungeon master doesn't pull a random enemy from the aether super well (or at least I don't!!). When I pick monsters I either logically think what makes sense for the area, pick something that's cool or vaguely humorous that I'm feeling the vibe of, or... use a random encounter table. For treasure and rewards... if it's not something hand picked, I'm again getting the loot tables out and rolling. And heck even for gold rewards and things, I'm rolling the dice. After trying basic stuff with langchain, an AI integrated with these types of tools is really promising. I think more systems on top could make it into a passable DM- at least until context becomes an issue.

capableweb 4y ago

If you set it up with some ground rules and rules around it's behavior (or more like guidelines), you can make it be more creative. Something like: "We're gonna start a text RPG game. Every animal we meet should be a made-up animal composed of joining two existing ones, with creative names being made for them. Don't use any usual fantasy tropes. Make encounters have surprising but logical outcomes" and so on.