{ "id": "u6-amp-early-a0060-g4b", "title": "Unit 6 \u00b7 Amplify typo register @ early (\u03b1=0.06) \u00b7 gemma-4b", "unit": "6", "created": "2026-07-09T23:02:45", "model": { "name": "gemma-4b", "hf_id": "google/gemma-3-4b-it", "quant": null, "n_layers": 34 }, "lens": { "repo": "neuronpedia/jacobian-lens", "file": "gemma-3-4b-it/jlens/Salesforce-wikitext/gemma-3-4b-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": 0.06 } }, "extra_md": null, "conversation": [ { "role": "user", "content": "Describe the water cycle in two sentences." }, { "role": "assistant", "content": "and is gonna be is gonna and is gonna be is and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and" } ], "generated": [ "and is gonna be is gonna and is gonna be is and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and is gonna and" ], "tokens": [ "", "", "user", "\n", "Describe", " the", " water", " cycle", " in", " two", " sentences", ".", "", "\n", "", "model", "\n", "and", " is", " gonna", " be", " is", " gonna", " and", " is", " gonna", " be", " is", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", " is", " gonna", " and", "", "\n" ], "readouts": [ { "position": 77, "token": "", "model_top": [ " is", " and", " And", " No", " gonna", " Well", " ", " There" ], "layers": { "0": [ "", " ", " (\\\"", " ", " \\\"", " \u00d0", " \u00c3", " ;-" ], "1": [ "", " ", "\uff8a", ",\u2026", ")\u2026", ",((", "\ufe51", ").[" ], "2": [ " ", " alot", ",", " whilst", " seperate", ".", " perch\u00e8", " \u0641\u0649" ], "3": [ " ", " alot", ".", " whilst", ",", "!.", " thats", " seperate" ], "4": [ " ", " alot", " thats", " whilst", ".", "!.", " Whilst", ".." ], "5": [ " ", " whilst", " alot", " Whilst", " seperate", " utilising", " thats", "Whilst" ], "6": [ " whilst", " ", " alot", " seperate", " amongst", " utilising", " however", " which" ], "7": [ " whilst", " alot", " ", " amongst", " seperate", " utilising", " Whilst", " however" ], "8": [ " whilst", " alot", " Whilst", " utilising", " amongst", " showcasing", " seperate", "Whilst" ], "9": [ " whilst", " alot", " Whilst", " utilising", " amongst", "Whilst", " showcasing", " seperate" ], "10": [ " whilst", " alot", " Whilst", "Whilst", " utilising", " showcasing", " impactful", " amongst" ], "11": [ " whilst", " alot", " Whilst", " impactful", " showcasing", "Whilst", " utilising", " amongst" ], "12": [ " whilst", " alot", " Whilst", " impactful", " showcasing", " utilising", "Whilst", " anyways" ], "13": [ " whilst", " alot", " Whilst", " showcasing", " impactful", " utilising", "Whilst", " utilises" ], "14": [ " whilst", " alot", " Whilst", " utilising", "Whilst", " showcasing", " anyways", " impactful" ], "15": [ " whilst", " alot", " Whilst", " utilising", "Whilst", " anyways", " seperate", " everytime" ], "16": [ " alot", " whilst", " Whilst", "Whilst", " everytime", " utilising", " seperate", " utilises" ], "17": [ " alot", " Whilst", " whilst", "Whilst", " everytime", " anyways", " utilising", " seperate" ], "18": [ " alot", " Whilst", " whilst", "Whilst", " everytime", " anyways", " Anyways", " utilises" ], "19": [ " alot", " Whilst", " everytime", " whilst", " Anyways", " anyways", "Whilst", " seperate" ], "20": [ " alot", " Whilst", " Anyways", " everytime", " anyways", " whilst", " doesnt", " wasnt" ], "21": [ " alot", " Anyways", " anyways", " everytime", " Whilst", " doesnt", " seperate", " utilises" ], "22": [ " alot", " Anyways", " anyways", "Anyways", " Whilst", " everytime", " doesnt", " seperate" ], "23": [ " alot", " Anyways", " anyways", " seperate", " doesnt", " everytime", " realises", " atleast" ], "24": [ " alot", " Anyways", " anyways", " gonna", " everytime", " seperate", "Anyways", " doesnt" ], "25": [ " alot", " gonna", " Anyways", " wanna", " seperate", " anyways", " everytime", " specialises" ], "26": [ " alot", " gonna", " wanna", " anyways", " Anyways", " thats", " seperate", " doesnt" ], "27": [ " wanna", " gonna", " alot", " Anyways", " Gonna", " anyways", "Anyways", " thats" ], "28": [ " alot", " Anyways", " wanna", " Thats", " thats", " isnt", " doesnt", " gonna" ], "29": [ " alot", " gonna", " Thats", " Anyways", " Gonna", " isnt", " thats", " wanna" ], "30": [ " Thats", " alot", " thats", " gonna", " wanna", " Anyways", " isnt", " Gonna" ], "31": [ " And", " thats", " gonna", " isnt", " is", " alot", " Gonna", " Thats" ], "32": [ " and", " And", " gonna", " Gonna", " is", " Well", " Also", "!!" ] } } ], "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 ], "ranks": [ 95942, 21701, 105, 80, 77, 49, 29, 18, 10, 10, 9, 9, 9, 9, 7, 6, 9, 6, 6, 5, 3, 2, 2, 2, 2, 3, 4, 4, 2, 4, 6, 20, 75 ] }, { "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 ], "ranks": [ 11903, 330, 2, 2, 2, 3, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 1, 1, 2, 6, 23 ] }, { "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 ], "ranks": [ 196801, 63165, 2073, 1591, 1242, 1665, 3279, 2347, 1106, 1136, 774, 665, 569, 677, 602, 619, 321, 163, 141, 136, 161, 119, 97, 98, 69, 78, 319, 302, 521, 631, 620, 1964, 4976 ] }, { "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 ], "ranks": [ 125890, 22544, 158, 115, 79, 81, 147, 116, 49, 47, 45, 45, 45, 59, 28, 25, 25, 14, 12, 13, 12, 14, 15, 16, 15, 23, 28, 24, 70, 76, 126, 338, 1474 ] }, { "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 ], "ranks": [ 13517, 865, 4, 4, 4, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 5, 5, 17, 27, 27, 49, 51, 30, 36, 57, 497, 462 ] }, { "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 ], "ranks": [ 147851, 50069, 432, 422, 426, 357, 299, 225, 155, 160, 114, 118, 113, 133, 110, 89, 76, 44, 42, 46, 32, 26, 25, 96, 41, 174, 66, 76, 114, 102, 84, 210, 104 ] } ], "emergence": { "position": 77, "top1": " is", "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 ], "ranks": [ 63038, 20022, 163, 184, 218, 173, 83, 91, 178, 180, 406, 556, 585, 568, 394, 372, 413, 509, 703, 510, 247, 145, 181, 219, 267, 275, 18, 27, 18, 21, 17, 5, 5 ] }, "scan": [], "slice": null }