RSS

Glitch Tokens

TagLast edit: 18 Apr 2023 5:31 UTC by CronoDAS

Glitch Tokens are tokens in a language model that cause anomalous output, such as SolidGoldMagikarp.

SolidGoldMag­ikarp (plus, prompt gen­er­a­tion)

Jessica Rumbelow and mwatkins
5 Feb 2023 22:02 UTC
663 points
204 comments12 min readLW link

SolidGoldMag­ikarp II: tech­ni­cal de­tails and more re­cent findings

mwatkins and Jessica Rumbelow
6 Feb 2023 19:09 UTC
109 points
45 comments13 min readLW link

Ano­ma­lous to­kens re­veal the origi­nal iden­tities of In­struct models

janus and jdp
9 Feb 2023 1:30 UTC
136 points
16 comments9 min readLW link
(generative.ink)

SolidGoldMag­ikarp III: Glitch to­ken archaeology

mwatkins and Jessica Rumbelow
14 Feb 2023 10:17 UTC
90 points
30 comments16 min readLW link

No­kens: A po­ten­tial method of in­ves­ti­gat­ing glitch tokens

Hoagy15 Mar 2023 16:23 UTC
20 points
0 comments4 min readLW link

The ‘ pe­ter­todd’ phenomenon

mwatkins15 Apr 2023 0:59 UTC
180 points
50 comments38 min readLW link

Smar­tyHead­erCode: anoma­lous to­kens for GPT3.5 and GPT-4

AdamYedidia15 Apr 2023 22:35 UTC
71 points
18 comments6 min readLW link

A Search for More ChatGPT /​ GPT-3.5 /​ GPT-4 “Un­speak­able” Glitch Tokens

Martin Fell9 May 2023 14:36 UTC
23 points
9 comments6 min readLW link

(redacted) Ano­ma­lous to­kens might dis­pro­por­tionately af­fect com­plex lan­guage tasks

nikola15 Jul 2023 0:48 UTC
4 points
0 comments7 min readLW link

The “spel­ling mir­a­cle”: GPT-3 spel­ling abil­ities and glitch to­kens revisited

mwatkins31 Jul 2023 19:47 UTC
85 points
29 comments20 min readLW link

Lin­ear en­cod­ing of char­ac­ter-level in­for­ma­tion in GPT-J to­ken embeddings

mwatkins and Joseph Bloom
10 Nov 2023 22:19 UTC
34 points
4 comments28 min readLW link

Map­ping the se­man­tic void: Strange go­ings-on in GPT em­bed­ding spaces

mwatkins14 Dec 2023 13:10 UTC
114 points
30 comments14 min readLW link

′ pe­ter­todd’’s last stand: The fi­nal days of open GPT-3 research

mwatkins22 Jan 2024 18:47 UTC
101 points
13 comments45 min readLW link

An ex­am­i­na­tion of GPT-2′s bor­ing yet effec­tive glitch

MiguelDev18 Apr 2024 5:26 UTC
4 points
0 comments3 min readLW link