Tomas Vykruta on model context window
From GPT-4 and Plugins - Tomas Vykruta (ML @ Google, Founder @ EvolutionIQ) · · Forward
“The input size has been increased from about 4K tokens up to as much as 32K, so the model can now effectively see 50 pages of text — that change alone makes a huge class of real‑world use cases possible that weren't before.”
On , Tomas Vykruta, Cofounder at EvolutionIQ, spoke about model context window during GPT-4 and Plugins - Tomas Vykruta (ML @ Google, Founder @ EvolutionIQ) on Forward.
About Tomas Vykruta
In a March 2023 podcast appearance, Tomas Vykruta discussed the release of GPT-4 and its new features. He described GPT-4 as multimodal, noting that it can accept images as input, and stated that the increased input size of up to 32K tokens makes new use cases possible. Vykruta characterized ChatGPT plugins as a significant development, saying they allow the model to interact with files, devices, and the physical world. He expressed concern about OpenAI's commercial focus, describing the GPT-4 technical report as "a brochure" and warning against a single company controlling AI. Vykruta encouraged support for competing models such as Google Bard, stating that he wants to see Bard succeed and that users should provide feedback to help it improve. He also suggested that OpenAI should make plugins open source to avoid a single company controlling the web through them.