We lifted 6 functions out of this paper's own repositories and ran 6 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| robertcsordas/onion_representations | canonical | 6 of 6 |
| Function | Status | Where it lives |
|---|---|---|
| construct_name | Ran | robertcsordas/onion_representations/paper/lib/common.py pointer only (licence: NONE) · get_code("8d74527db3b7eebc") |
| get_logs | Ran | robertcsordas/onion_representations/framework/layers/logging_layer.py pointer only (licence: NONE) · get_code("9d5bf0b8a99793b6") |
| group | Ran | robertcsordas/onion_representations/paper/lib/common.py pointer only (licence: NONE) · get_code("be4657e3537b2b6b") |
| gumbel_noise_like | Ran | robertcsordas/onion_representations/framework/layers/gumbel_softmax.py pointer only (licence: NONE) · get_code("6ffc97f245b45f9a") |
| gumbel_softmax | Ran | robertcsordas/onion_representations/framework/layers/gumbel_softmax.py pointer only (licence: NONE) · get_code("70ab3336f5bc37ef") |
| sinusoidal_pos_embedding | Ran | robertcsordas/onion_representations/framework/layers/positional_encoding.py pointer only (licence: NONE) · get_code("d66ffb4b6ae08ca3") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
The Linear Representation Hypothesis (LRH) states that neural networks learn to encode concepts as directions in activation space, and a strong version of the LRH states that models learn only such encodings. In this paper, we present a counterexample to this strong LRH: when trained to repeat an input token sequence, gated recurrent neural networks (RNNs) learn to represent the token at each position with a particular order of magnitude, rather than a direction. These representations have layered features that are impossible to locate in distinct linear subspaces. To show this, we train interventions to predict and manipulate tokens by learning the scaling factor corresponding to each sequence position. These interventions indicate that the smallest RNNs find only this magnitude-based solution, while larger RNNs have linear representations. These findings strongly indicate that interpretability research should not be confined by the LRH.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2408.10920")
get_code_for_paper("2408.10920")
have("2408.10920")
Connect an agent — have() is free.