Excellent primer on LLMs and probably the first post I see that explains each phase from base training to preference tuning with this much clarity.
"At the risk of anthropomorphizing too much, I like to summarize this insight as follows: LLMs only think out loud. If you want them to “think” better, get them to “talk” more."
That's a very helpful way to get a layperson to think about stuff like CoT prompting and similar methods. I'm going to go ahead and steal it!
Damnit. I should probably not have announced my intent to steal publicly. I'm not very good at this criminal mastermind shtick.
your write: Try out the latest generative language models like GPT-4 with Microsoft’s Copilot or Google Gemini on their free tier.
It seems there is no free GPT-4. Please explain a little bit more.
Thanks.
I haven't checked in the last couple of weeks but Microsoft Bing's chat used to have a "creative" mode which under the hood was just GPT-4.
Thanks, it works.
Love the clarity of your writing. Masterful. Thanks!
Thank you for your kind words 🙏
Thanks for this break-down, very useful
Thank you for your feedback ;)
Excellent primer on LLMs and probably the first post I see that explains each phase from base training to preference tuning with this much clarity.
"At the risk of anthropomorphizing too much, I like to summarize this insight as follows: LLMs only think out loud. If you want them to “think” better, get them to “talk” more."
That's a very helpful way to get a layperson to think about stuff like CoT prompting and similar methods. I'm going to go ahead and steal it!
Damnit. I should probably not have announced my intent to steal publicly. I'm not very good at this criminal mastermind shtick.
The baby wipes post made that last part very clear ;)
It's the thought that counts!