AI shenanigans: bodging a budget bot box
It seems to me that a lot of people’s opinions1 about large language models and their abilities haven’t been updated since… ooh about 2023. Back then, the frontier models were rough at best. I remember playing around with early ChatGPT and reading funny stories of its spectacular misses, and for a while I was fully on board with the anti-hype of commentators like Ed Zitron.
However, over the last two years my thinking has evolved quite a lot. I’ve become—if not an AI mega-booster—much more appreciative of its utility. This isn’t because, per Zitron, companies like OpenAI and Anthropic are just throwing more and more GPUs at the problem and hoping for the best. Open-weight models that can be run on consumer hardware have also been improving in leaps and bounds, and it’s not actually that difficult or expensive to get started.
Read more...
