There's a new UI for visualizing the number of chars, custom ai steps, and credit on your account accessible via Account Settings.

Recently, there have been a number of bugs with the text-to-speech (tts) option that have been fixed:

Previously, character counts were inaccurate, leading to general confusion regarding remaining characters left for generations. This issue has been fixed.

We've improved how prompts and activated context entries get budgeted, leading to better output for Dictionary-heavy Holos.

In December users reported intermittent downtime and general sluggishness. This has been fixed and as a bonus the site should be significantly faster.

In January some users reported GPT-J as generating irregularly poor output. After a series of model changes and bug-fixes, we believe we have restored output quality.

We’ve introduced a un-tuned Fairseq 13B model on our new $11.99/mo tier. We plan on releasing a new finetune for our GPT-J 6B model and 13B model at a later date.

We've added some keyboard shortcuts to make creating and choosing generations more efficient. You can now:

Two fixes/improvements:

We have provided more tool-tips and explanations in the Settings Panel, including all generation metadata settings.