There has been a lot of headlines discussing the diminishing returns of throwing more data, model parameters, and compute to improve llm performance. This is a great math/technical explanation why performance gains is slowing
https://t.co/hBFSy1zyUT
@nicklmg does creator warehouse plan to restock or update the party shirt? I continue to get compliments at work every time I rock the shirt. If not, some colored polos would be nice. Perhaps a blue in a similar shade to the party shirt, or a irish green.
I got some time today to try conversable agents and tools in @pyautogen using #gemma2, #Phi2, and @LMStudioAI. The process was a lot easier than @langchain.
tl;dw USB C peripherals have a 5V handshake with the mac. If your mac is not shut down or disconnected from USB C, the handshake will continue and eventually burn out the chips leading to a dead Macbook
For those who have a M2 Macbook Pro (or any hardware with >8gb of unified RAM or VRAM), and want to try open source models, check out LM Studio. The app provides a nicer front end to browse and test models that are optimal for your given hardware setup.
https://t.co/WkbZcsKi2x