INDEX

Explanations

end matter references

The neuron strongly activates on structured outline markers—i.e. tokens that are part of numbered headings, chapters, or question‐number labels.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

img

0.59

 čís

0.59

 story

0.53

img

0.53

áj

0.52

 smoking

0.51

說是

0.51

 headline

0.51

 মূর্তি

0.51

explique

0.51

POSITIVE LOGITS

 Appendix

1.93

 appendices

1.92

Appendix

1.90

 appendix

1.86

 Bibliography

1.84

 Conclusion

1.82

 APPENDIX

1.81

APPENDIX

1.80

Conclusion

1.79

appendix

1.73

Activations Density 0.209%