I am not sure if you have covered it in a previous video but it has code execution under Advanced Settings which allows prompts like "Calculate the sum of the first 50 prime numbers" to create and run python code and work out the answer which it would give an incorrect/hallucination answer otherwise. This also works with other Google AI models including Gemini 1.5 Flash and in the API
It still failed on one of my chemistry math problems. No model has ever been able to solve this, not even closed ones like ChatGPT and Claude, except for Gemma27b, which nails it every time. Closed ones like GPT-4 and Claude can still do it, but only after maybe more than five tries. But Gemma27b solves it correctly every time, zero shot. It's weird that this latest model, supposedly better than GPT-4, is less capable than Gemma27b - it which is their open source model.
I am not sure if you have covered it in a previous video but it has code execution under Advanced Settings which allows prompts like "Calculate the sum of the first 50 prime numbers" to create and run python code and work out the answer which it would give an incorrect/hallucination answer otherwise. This also works with other Google AI models including Gemini 1.5 Flash and in the API
It still failed on one of my chemistry math problems. No model has ever been able to solve this, not even closed ones like ChatGPT and Claude, except for Gemma27b, which nails it every time. Closed ones like GPT-4 and Claude can still do it, but only after maybe more than five tries. But Gemma27b solves it correctly every time, zero shot. It's weird that this latest model, supposedly better than GPT-4, is less capable than Gemma27b - it which is their open source model.
Do you find it more censored than GPT and Claude?
So never sleep 😂!!!
lol, yeah mate. thanks.