Hello Rem, I just subscribed to you today as I've been observing your podcast and your ideas. Thank you and it helped me. Unfortunately I cannot pledge as of the moment but hopefully in the near future, I can.
I was challenge by this podcast. I am vibe coding for the past 1 year already since I'm still learning prompt engineering. Over the past 3 months, I learned one thing. We can increase the rate of deterministic output if we reframe our expectations not on the actual output itself but on how the output will be presented to us.
In your example, given 1 and 2 and the function adds two numbers, it will always output 3. I will use some of the concepts around this.
If I hire a care taker to take care of my plants, my expectation would be the plant would grow healthy. So I gave the care taker a plant to take care of and after 1 month, the plant should grow healthy. This closed system problem is a probabilistic scenario. I would expect two things, the plant will not grow healthy, or the plant will grow healthy. So let's say I got back after one month, I saw the plant, it was healthy on the outside. Looks big, green and full of life. I am happy that the plant was healthy. Now on the parallel universe, after one month, the plant was sick. So I asked the care taker why, the care taker said he have no means to procure fertilizers and other needs to make the plant healthy. So I gave him all the means to make the plant healthy. So after another month, I got back, and now the plant was healthy.
Using this concept, we will now connect this to how I observe my interactions with any LLMs. If I give any LLMs a set of means to make their output deterministic, I can increase the chance that they will output near the goal that I want to achieve.
It's like you're programming your prompt to make any LLMs better understood what you want to achieve.
Anyways, just sharing what I've learned so far. I hope this helps any vibe coders out there. :)
Hello Rem, I just subscribed to you today as I've been observing your podcast and your ideas. Thank you and it helped me. Unfortunately I cannot pledge as of the moment but hopefully in the near future, I can.
I was challenge by this podcast. I am vibe coding for the past 1 year already since I'm still learning prompt engineering. Over the past 3 months, I learned one thing. We can increase the rate of deterministic output if we reframe our expectations not on the actual output itself but on how the output will be presented to us.
In your example, given 1 and 2 and the function adds two numbers, it will always output 3. I will use some of the concepts around this.
If I hire a care taker to take care of my plants, my expectation would be the plant would grow healthy. So I gave the care taker a plant to take care of and after 1 month, the plant should grow healthy. This closed system problem is a probabilistic scenario. I would expect two things, the plant will not grow healthy, or the plant will grow healthy. So let's say I got back after one month, I saw the plant, it was healthy on the outside. Looks big, green and full of life. I am happy that the plant was healthy. Now on the parallel universe, after one month, the plant was sick. So I asked the care taker why, the care taker said he have no means to procure fertilizers and other needs to make the plant healthy. So I gave him all the means to make the plant healthy. So after another month, I got back, and now the plant was healthy.
Using this concept, we will now connect this to how I observe my interactions with any LLMs. If I give any LLMs a set of means to make their output deterministic, I can increase the chance that they will output near the goal that I want to achieve.
It's like you're programming your prompt to make any LLMs better understood what you want to achieve.
Anyways, just sharing what I've learned so far. I hope this helps any vibe coders out there. :)