0:00 / 0:00
- Recommended by 1 person: 1 in engineering communities.
- Presented by Karthik Chikmagalur
- Linked in the descriptionkarthink/gptelkarthink/gptel-agentkarthink/gptel-annotate
Intermediate. Assumes working knowledge.
5.6K views2 months ago1 recommendation
Tools and practiceAI and machine learningVim, Emacs and editorsTransformers and LLMs
Karthik Chikmagalur demonstrates gptel's presets feature, showing how to customize and pipe LLM interactions directly within Emacs.
Some experiments with using LLMs in Emacs, and a demo of gptel's presets feature. We can do better than a linear, immutable, top-down chat interface to LLMs. gptel is an LLM client for Emacs. (gptel.org) A preset is a named collection of gptel settings and behaviors applied to an LLM query as a unit: they can change any aspect of LLM interaction on the fly, providing a flexible, customizable interface for most (text-based) things you'd need an LLM for. LLMs as text transformations LLM clients as shell commands Software with too many options Configuration files and (finally) a gptel preset The tutor preset Inside the Tutor preset The Introspect preset The JSON preset Interlude I: How presets work Redirection: source | gptel | destination Source: the Expand preset The Visible-Text preset The Include preset Interlude II: Adding dynamic behavior to presets Destination: the Notify preset The Rewrite preset The Annotate preset Presets as "agents" Problems with the presets UI Interlude III: Resources for presets LLM UIs in Emacs: this ain't it Emacs packages showcased in this demo: - github.com/karthink/gptel - github.com/karthink/gptel-agent - github.com/karthink/gptel-annotate - github.com/karthink/gptel-preset-collection - github.com/kmontag/macher Emacs packages and settings used in this demo - Presentation software: github.com/positron-solutions/dslide - Visuals: github.com/minad/org-modern and github.com/rnkn/olivetti - Fancy Flymake UI: github.com/konrad1977/flyover - Font: Monaspace Neon - Emacs theme: ef-reverie #emacs #llm #llms #gptelWhy it's here
1 person recommends thisWhat people say. Recommended as an excellent demonstration of using gptel presets to treat LLMs as flexible text transformations.