I have a different body but a similar relationship to it as the author. My wife is very stylish and often wears striking jewellery. She gets a lot of compliments and doesn't have to work hard to be at ease socially, whereas for the most part I'd be happy to pass through the world unseen. We've had some modest financial success of late, and I'm resolved to being kinder to myself but also taking myself a little more seriously. I think it's easy to write off fashion as something laboured and fake and innately suspect, especially when you're an introvert and a bit of a nerd living the life of the mind. But I've come to accept that my body has value, a voice of its own, and is part of me that I can't and don't want to be rid of. I'm not really ready for tailored clothing (and I flatter myself it'd be wasteful when I lose all the weight) but I've started dressing better, sports jackets instead of hoodies, shirts instead of t-shirts. For the first time in my life I own more than two pairs of shoes. I am sure I would still look shambolic to anyone with an eye for the details (I can pull off dishevelled academic better than Ivy Style) but I feel so much more confident and relaxed. I no longer feel like I've written off half of myself, and I like the whole a lot more because of that. It was wonderful to read Gary's journey in a similar direction.
Very importantly here they provide a ways of decoding the encoded thought tokens, so you're not really losing explanatory power or debuggability. As much as OpenAI want to present hidden chain of thought as some sort of long term advantage or safety feature, it's horrible when you want to understand how a model came to some insane conclusion.
Yup. The LLM-AutoDiff is just getting started. But it has proven generation-only without explicitly doing few-shot samples can be even more effective and create shorter final prompts
https://github.com/microsoft/garnet
reply