giantfish-fly commited on
Commit
11aac11
·
verified ·
1 Parent(s): 5d21275

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +2 -2
README.md CHANGED
@@ -85,7 +85,7 @@ The current value of Key1 is Value_N.
85
 
86
 
87
  ## Results:
88
- LLMs **cannot reliably retrieve** Value_N. Distribution spans value_1 to value_N, and **as N increases**, the **answers skew** increasingly toward **value_1**.
89
 
90
 
91
 
@@ -114,7 +114,7 @@ Our test adopts the **classic proactive** interference paradigm from cognitive s
114
 
115
  See: https://sites.google.com/view/cog4llm
116
 
117
- ## Results at a glance
118
  - Humans: near-ceiling accuracy (99%+) on this controlled task across conditions (see paper for protocol and exact numbers).
119
  - LLMs: accuracy declines approximately log-linearly with the number of updates per key and with the number of concurrent update blocks (details, plots, and model list in our paper).
120
 
 
85
 
86
 
87
  ## Results:
88
+ ALL tested SOTA LLMs **cannot reliably retrieve** Value_N. Distribution spans value_1 to value_N, and **as N increases**, the **answers skew** increasingly toward **value_1**.
89
 
90
 
91
 
 
114
 
115
  See: https://sites.google.com/view/cog4llm
116
 
117
+ ## SAME Log-linear Decline of Accuracy for ALL SOTA LLMs tested(2019-2025)
118
  - Humans: near-ceiling accuracy (99%+) on this controlled task across conditions (see paper for protocol and exact numbers).
119
  - LLMs: accuracy declines approximately log-linearly with the number of updates per key and with the number of concurrent update blocks (details, plots, and model list in our paper).
120