One question posed in the video: What do you think of James Earl Jones signing over his voice-rights, so it can be used by AI for new recordings? Is this progress? Blasphemy? Exciting? Cringe?
I mean its ok but i'm refining reasoning layers for Claude 3.5...gpt4o mini and Llama. I've had moderate success and Claude can get comparable performance to 1o if you COT prompt it right, like it can do common sense outputs. I'm gonna hold my money until Orion drops.
Right - it seems like much of the value-add is intuitively baking in all the "chain-of-thinking" prompts, that typically only Prompt Ninjas know how to do - making it yield way better results for newbs.
It's not as good as it seems. As i show on my channel, it cannot do simple maths - albeit that o1 is better than 4o or Claude. If it cannot do that then it cannot reliably answer maths and science questions. It's nowhere near AGI. It's a Language Model - and until they have some way to integrate maths, these models will forever fail in this area.
One question posed in the video: What do you think of James Earl Jones signing over his voice-rights, so it can be used by AI for new recordings? Is this progress? Blasphemy? Exciting? Cringe?
I mean its ok but i'm refining reasoning layers
for Claude 3.5...gpt4o mini and Llama.
I've had moderate success and Claude can get comparable
performance to 1o if you COT prompt it right, like it can
do common sense outputs.
I'm gonna hold my money until Orion drops.
Right - it seems like much of the value-add is intuitively baking in all the "chain-of-thinking" prompts, that typically only Prompt Ninjas know how to do - making it yield way better results for newbs.
It's not as good as it seems. As i show on my channel, it cannot do simple maths - albeit that o1 is better than 4o or Claude. If it cannot do that then it cannot reliably answer maths and science questions. It's nowhere near AGI. It's a Language Model - and until they have some way to integrate maths, these models will forever fail in this area.
Agree, nowhere close to AGI! Great context, and thanks for chiming in. I view this as *progress* but, to your point, certainly not the end all be all.