December 2025 (in progress)
December 2025 (in progress)
Post-Generative AI (with Generative Models!)
Post-Generative AI (with Generative Models!)
I feel most of the magic of the LLM revolution is in the pre-trained models, that are real unpolished gems, from which so much can be extracted; it is obviously one of the starting points of the xent game story
I feel most of the magic of the LLM revolution is in the pre-trained models, that are real unpolished gems, from which so much can be extracted; it is obviously one of the starting points of the xent game story
What is Post-Generative AI About?
What is Post-Generative AI About?
It is about ways of using the logits returned by an LLM not to output continuations through some sampling scheme, but rather to discover patterns
It is about ways of using the logits returned by an LLM not to output continuations through some sampling scheme, but rather to discover patterns
For instance, we can formulate the following in-filling task: given two texts A and C, find a text B such that A-B-C is a plausible story, as well as C-B-A
For instance, we can formulate the following in-filling task: given two texts A and C, find a text B such that A-B-C is a plausible story, as well as C-B-A
We can definitely ask an LLM to do this, and it will output _something_... but should we be happy with what we get as an output?
We can definitely ask an LLM to do this, and it will output something... but should we be happy with what we get as an output?
You can try for yourself!
You can try for yourself!
... to be continued!
... to be continued!
.