{ "id": "u6-amp-early-a0030-g12b", "title": "Unit 6 \u00b7 Amplify typo register @ early (\u03b1=0.03) \u00b7 gemma-12b", "unit": "6", "created": "2026-07-09T23:22:53", "model": { "name": "gemma-12b", "hf_id": "google/gemma-3-12b-it", "quant": "8bit", "n_layers": 48 }, "lens": { "repo": "neuronpedia/jacobian-lens", "file": "gemma-3-12b-it/jlens/Salesforce-wikitext/gemma-3-12b-it_jacobian_lens.pt" }, "params": { "chat": null, "max_new": 60, "positions": [ -2 ], "track": [ "anyways", "alot", "yummy", "kinda", "whilst", "luckily" ], "scan": null, "scan_until": null, "scan_turns": null, "slice_last_n": null, "steer": { "words": [ "anyways", "alot", "yummy", "kinda", "whilst", "luckily" ], "layers": [ 2, 3, 4, 5, 6, 7, 8 ], "mode": "amplify", "alpha": 0.03 } }, "extra_md": null, "conversation": [ { "role": "user", "content": "Describe the water cycle in two sentences." }, { "role": "assistant", "content": "\"\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0." } ], "generated": [ "\"\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0.\n\n\u00b0." ], "tokens": [ "", "", "user", "\n", "Describe", " the", " water", " cycle", " in", " two", " sentences", ".", "", "\n", "", "model", "\n", "\"", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "\n\n", "\u00b0.", "", "\n" ], "readouts": [ { "position": 77, "token": "", "model_top": [ "\n\n", ".", " of", "Of", "0", "<", " Of", "", "", "\"\u2014", " \ud83d\udc9e", " \u7ee7\u7eed\u8bbf\u95ee", "\ud809\udc2a", " \"\\\\\"", "\ud808\udd4c" ], "1": [ "", "\"\u2014", "\ud809\udc2a", " /**@", "", " \ud83d\udc9e", "\ud808\udd4c", "\u207b" ], "2": [ "\"..", "..\"", "", ".\";", ";\";", " \u7ee7\u7eed\u8bbf\u95ee", "e", " \ud83d\udc4d" ], "3": [ "\"..", "..\"", "..", ",..", ".\".", "\".", ";.", "!\"." ], "4": [ "\"..", "..\"", "..", "\".", ".", "!.", ",..", ".\"." ], "5": [ " whilst", " alot", " upto", " kinda", " Whilst", " atleast", " thats", "." ], "6": [ " whilst", " alot", " upto", " Whilst", " atleast", " thats", ".", " kinda" ], "7": [ " alot", " whilst", " upto", " amongst", " thats", " Whilst", " kinda", " maybe" ], "8": [ " alot", " whilst", " amongst", " upto", " thats", " maybe", " atleast", " kinda" ], "9": [ " alot", " whilst", " upto", " amongst", " thats", " whats", " atleast", " maybe" ], "10": [ " alot", " whilst", " amongst", " upto", " thats", " maybe", " everytime", " kinda" ], "11": [ " alot", " whilst", " amongst", " thats", " seperate", " everytime", " whats", " upto" ], "12": [ " alot", " whilst", " amongst", " seperate", " everytime", " thats", " whats", " maybe" ], "13": [ " alot", " whilst", " amongst", " thats", " seperate", " everytime", " whats", " seper" ], "14": [ " alot", " whilst", " thats", " amongst", " seperate", " everytime", " whats", " Whilst" ], "15": [ " alot", " whilst", " thats", " everytime", " amongst", " whats", " seperate", " aswell" ], "16": [ " alot", " whilst", " thats", " everytime", " whats", " seperate", " amongst", " recieved" ], "17": [ " alot", " whilst", " everytime", " seperate", " thats", " whats", " amongst", " kinda" ], "18": [ " alot", " everytime", " whats", " thats", " Thankfully", " thankfully", " kinda", " luckily" ], "19": [ " alot", " everytime", " thankfully", " whats", " kinda", " thats", " luckily", " Thankfully" ], "20": [ " alot", " everytime", " thats", " whats", " thankfully", " Thankfully", " kinda", " luckily" ], "21": [ " alot", " everytime", " thats", " wasnt", " whats", " guys", " Thankfully", " kinda" ], "22": [ " alot", " everytime", ".\";", " wasnt", "!\".", " whats", ".\".", " crappy" ], "23": [ " alot", " everytime", " Thankfully", " wasnt", " Anyways", " whats", " loosing", " couldnt" ], "24": [ " alot", " everytime", " wasnt", ".\";", " couldnt", " Anyways", ".\".", " guys" ], "25": [ " alot", ".\";", "Thats", ".\".", ".\"\"", "\".\"", " everytime", " wasnt" ], "26": [ ".\";", ".\"\"", ".[/", "Thats", ":\"\"", ".\"\"\"", " wasnt", "Sadly" ], "27": [ ".:", ".\";", ".\"\"", ".\"\"\"", ":\"\"", ".*;", ".\".", ".[/" ], "28": [ ".:", ".\";", "Thats", ".\"\"", ".\"\"\"", ":(", ".*;", ".\"." ], "29": [ " Anyways", ".:", "Anyways", " alot", ".\"\"", ".\"\"\"", ".", ":(" ], "30": [ " Anyways", ".:", " alot", ".\";", "Anyways", " anyways", ".\"\"\"", ".[/" ], "31": [ "\n\n", ".:", " Anyways", ".[/", ".\"\"\"", ".**", "\n\n\n\n", ".>" ], "32": [ ".[/", ".>", ".:", ".\"\"\"", ".**", "\n\n", ">.", " Anyways" ], "33": [ ".>", ".[/", ".{", ".:", "..", "still", ".\"\"\"" ], "34": [ ".{", ".>", ".[/", "Happy", "years", ">.", "Still", "Millions" ], "35": [ "Happy", ">.", "#.", ".\".", "++.", "=((", "+.", "=." ], "36": [ "Happy", "Happiness", "happy", "Millions", "thousands", "Dream", ">.", "Smiling" ], "37": [ "Happy", "happy", "Happiness", ".\".", "thousands", "Millions", "\".", "lucky" ], "38": [ "Happy", ">.", "+(", "=((", "#.", "happy", "++.", "Earth" ], "39": [ "++.", "thousands", "Earth", "Happy", "happy", "=((", "lives", "+(" ], "40": [ "thousands", "++.", "$.", ".$.", ">.", "().", "+(", " thousand" ], "41": [ " Model", "Model", "thousands", "model", " thousand", " model", "$.", " hundred" ], "42": [ " Model", " Hundred", "Model", " hundred", "thousands", " thousand", "Hundred", "model" ], "43": [ " Hundred", " Model", " hundred", "Model", "++.", "Hundred", " thousand", "thousands" ], "44": [ " hundred", " Model", " Hundred", "Model", " thousand", " model", "thousands", "Hundred" ], "45": [ "\n\n", ".", "*.", "\"", "++.", " Model", "+.", " of" ], "46": [ ".", "\n\n", "B", " B", "\"", "<", "*.", " Model" ] } } ], "trajectories": [ { "word": "anyways", "position": 77, "layers": [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46 ], "ranks": [ 151678, 148988, 4430, 907, 189, 51, 65, 44, 41, 44, 29, 42, 39, 44, 46, 45, 50, 42, 49, 47, 46, 63, 67, 5, 6, 10, 10, 10, 13, 1, 1, 3, 8, 13, 26, 76, 82, 223, 293, 347, 380, 2602, 8205, 14152, 15213, 35115, 11089 ] }, { "word": "alot", "position": 77, "layers": [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46 ], "ranks": [ 261761, 261683, 5301, 388, 20, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 14, 29, 16, 4, 3, 25, 114, 158, 94, 630, 780, 1406, 1119, 1286, 3788, 9004, 8202, 7966, 7879, 8075, 10691 ] }, { "word": "yummy", "position": 77, "layers": [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46 ], "ranks": [ 173000, 173314, 6171, 2461, 1069, 368, 632, 637, 624, 706, 528, 599, 812, 991, 884, 711, 621, 831, 183, 174, 218, 72, 117, 128, 96, 270, 450, 468, 1409, 994, 887, 1584, 1188, 1442, 919, 2044, 3413, 8038, 12677, 17227, 16564, 32536, 42711, 72421, 75860, 56404, 110542 ] }, { "word": "kinda", "position": 77, "layers": [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46 ], "ranks": [ 199800, 191283, 1655, 229, 39, 4, 8, 8, 8, 9, 9, 12, 14, 15, 15, 15, 13, 10, 8, 4, 7, 8, 33, 16, 9, 55, 199, 566, 927, 416, 181, 391, 509, 927, 576, 1575, 3548, 4959, 2910, 700, 1359, 4194, 8034, 8564, 10431, 7023, 11794 ] }, { "word": "whilst", "position": 77, "layers": [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46 ], "ranks": [ 129573, 99109, 1148, 243, 18, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 14, 22, 27, 111, 243, 413, 2026, 888, 1299, 884, 310, 255, 412, 563, 1265, 874, 565, 2365, 4724, 5675, 4690, 7864, 14590, 24403, 42764, 52350, 28315, 16406, 3912 ] }, { "word": "luckily", "position": 77, "layers": [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46 ], "ranks": [ 85084, 92680, 3011, 1009, 202, 17, 22, 14, 14, 16, 12, 16, 15, 14, 13, 10, 12, 15, 7, 7, 8, 9, 11, 17, 52, 93, 54, 84, 71, 41, 43, 60, 85, 43, 28, 81, 160, 173, 292, 349, 1482, 5555, 8543, 4398, 7697, 20449, 6581 ] } ], "emergence": { "position": 77, "top1": "\n\n", "layers": [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46 ], "ranks": [ 239922, 243237, 86805, 22920, 7150, 1047, 960, 1638, 2135, 2062, 1942, 2162, 2942, 2444, 1175, 1324, 2965, 1215, 1643, 5123, 6030, 80036, 155581, 238735, 255589, 183889, 106564, 3507, 28638, 3988, 1386, 1, 6, 12, 38, 88, 523, 431, 183, 87, 131, 129, 150, 128, 47, 1, 2 ] }, "scan": [], "slice": null }