Going to open-source the most token-intensive project I've ever built tomorrow If enough people try it, it'll probably double the use of the Claude 3 API within a day 🙂
@mattshumer_
-
How OpenAI Fine-Tunes LLMs Into Assistant Models
By
–
My best slightly educated guess: – LLMs (base models) don’t naturally speak this way. We know this. They’re effectively a mirror of the internet, and therefore write very differently from this.
– When OpenAI started fine-tuning their LLMs into Assistant models, they likely had a -
AI Voice Quality Needs Improvement for Natural Sound
By
–
Yeah — really smart. Just wish they sounded more natural…
-
Command R+ Achieves Better Token Efficiency Than GPT Models
By
–
Interesting finding from the Command R+ blog post. For many languages, Command R+ will use far fewer tokens than GPT models to represent the same text. This effectively reduces overall costs significantly.
-
HyperWrite Model Training Progress Update
By
–
I'll definitely try some things over the next few days. Been heads down training models internally for HyperWrite, so it might be a few days.
-
API Access Available for AI Development
By
–
They have an API, I believe you should be able to access it! cc @aidangomez
-
Claude-LLM-Trainer Gains Momentum Among AI Enthusiasts
By
–
Thanks for writing about Claude-LLM-Trainer, Ben! Excited for more people to experience it!
-
Cohere Command R+ Advances LLM Capabilities with 128k Context
By
–
The new @cohere Command R+ model looks like a serious contender. – 128k context length
– best-in-class function calling (beats GPT-4-Turbo)
– fantastic multilingual support Excited to try this model and see what it enables! -
Haiku vs Opus: Cost Comparison for AI Model Usage
By
–
If you use Haiku, it's quite cheap, probably under 50 cents for a book. Opus is much more expensive.
-
AI Models Getting Better Make Code Simpler
By
–
Pretty much, with some added fanciness here and there. As the models get better, the code gets simpler. Older versions of this were far more complex. I'd guess in a year, this could be done with just one or two prompts.