@RoundtableSpace The input is more important than the weights. The weights are just an equation, and they are all pretty similar. Any output can be generated given proper input.
2-bit Gemma 4 12B GGUF, only 4.66 GB on disk, managed to cite 15 sites from a single prompt.
Try this locally on >6GB RAM via Unsloth Studio.
GitHub: https://t.co/aZWYAtakBP
crazy to think that if Google never dropped "Attention Is All You Need" in 2017, half of tech Twitter would probably be arguing about crypto or the metaverse right now
OMFG… that’s brutal.
This could be devastating…
Don’t freak out yet I need to harvest the ISWAC data and run some numbers.
Also, I need @fractal_verse to look at this ASAP
The problem for tech is the set up is so bad. The expectations for both quarters were so huge that the companies had to take chances they do not do in order to raise guidance. It's happened a couple of times to both of them. This time it happens at the same time as the greatest fundraise in history. Not a fortuitous moment.
@DarioCpx@edzitron Apple is focusing on what developers, consumers, and enterprise users want, which is hardware to run and train private local models, and there are millions of free models available right now. Over 100 released by Apple.
https://t.co/CRG6rZs1YL
@GaryMarcus Google released Transformer architecture about 10 years ago. They probably had the next architecture 5 years ago, and it probably looks like their Genie world model in the processor of a robot.
@GaryMarcus They just got stuck scaling up to build the Artificial God Intelligence, rather than building what consumers wanted.
https://t.co/wfOcE3Cbtq