From 5b02279b5a1af46c54c3065c0976fbd355d75a8b Mon Sep 17 00:00:00 2001 From: Peter Danenberg Date: Tue, 24 Sep 2024 02:40:01 -0700 Subject: [PATCH] Relative paths for SVGs (#37) * Fix clicks, captions. * Artifact loop * First pass at animatable SVGs. * Despeckle to 5 * Change CSS. * History of LLMs: animatable graphs with labels * Animatable SVGs in slides. * Three scenarios * Timeline * Update history slides * LinkedIn, etc. * Also build history * Add license * Notes * Notes * Use relative paths for SVGs. --- history/slides/slides.md | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/history/slides/slides.md b/history/slides/slides.md index a21cc86..b94d1e3 100644 --- a/history/slides/slides.md +++ b/history/slides/slides.md @@ -34,7 +34,7 @@ mdc: true --- @@ -101,7 +101,7 @@ layout: section --- @@ -132,7 +132,7 @@ South Sudan --- @@ -211,7 +211,7 @@ Future of LLMs might fall into one of three scenarios, like the future of the un
- +
Figure 2: The "hockey-stick" trajectory would be the equivalent of LLMs achieving runaway self-improvement. @@ -228,7 +228,7 @@ LLMs achieve super-human intelligence; new physics; inter-planetary travel; etc.
- +
Figure 3: This is the scenario where LLMs reach a kind of saturation point—having learned everything they can from human-generated data. @@ -245,7 +245,7 @@ LLMs peter out after the saturation of training data.
- +
Figure 4: This scenario occurs if LLMs start relying heavily on synthetic data, which could lead to model collapse.