RSS

Transformers

TagLast edit: 24 Feb 2022 11:01 UTC by Vivek Hebbar

Strik­ing Im­pli­ca­tions for Learn­ing The­ory, In­ter­pretabil­ity — and Safety?

RogerDearnaley5 Jan 2024 8:46 UTC
37 points
4 comments2 min readLW link

How LLMs are and are not myopic

janus25 Jul 2023 2:19 UTC
138 points
16 comments8 min readLW link

AGI will be made of het­ero­ge­neous com­po­nents, Trans­former and Selec­tive SSM blocks will be among them

Roman Leventov27 Dec 2023 14:51 UTC
33 points
9 comments4 min readLW link

Google’s PaLM-E: An Em­bod­ied Mul­ti­modal Lan­guage Model

SandXbox7 Mar 2023 4:11 UTC
87 points
7 comments1 min readLW link
(palm-e.github.io)

Resi­d­ual stream norms grow ex­po­nen­tially over the for­ward pass

StefanHex and TurnTrout
7 May 2023 0:46 UTC
79 points
24 comments9 min readLW link

How fast can we perform a for­ward pass?

jsteinhardt10 Jun 2022 23:30 UTC
53 points
9 comments15 min readLW link
(bounded-regret.ghost.io)

Modern Trans­form­ers are AGI, and Hu­man-Level

abramdemski26 Mar 2024 17:46 UTC
232 points
88 comments5 min readLW link1 review

How Do In­duc­tion Heads Ac­tu­ally Work in Trans­form­ers With Finite Ca­pac­ity?

Fabien Roger23 Mar 2023 9:09 UTC
27 points
0 comments5 min readLW link

LLMs Can’t See Pix­els or Characters

Brendan Long20 Jul 2025 20:00 UTC
100 points
44 comments4 min readLW link
(www.brendanlong.com)

Tracr: Com­piled Trans­form­ers as a Lab­o­ra­tory for In­ter­pretabil­ity | Deep­Mind

DragonGod13 Jan 2023 16:53 UTC
62 points
12 comments1 min readLW link
(arxiv.org)

Con­crete Steps to Get Started in Trans­former Mechanis­tic Interpretability

Neel Nanda25 Dec 2022 22:21 UTC
57 points
7 comments12 min readLW link
(www.neelnanda.io)

[Question] If I ask an LLM to think step by step, how big are the steps?

ryan_b13 Sep 2024 20:30 UTC
7 points
1 comment1 min readLW link

In­tro­spec­tion or con­fu­sion?

Victor Godet9 Nov 2025 20:53 UTC
43 points
3 comments4 min readLW link

Logit Prisms: De­com­pos­ing Trans­former Out­puts for Mechanis­tic Interpretability

ntt12317 Jun 2024 11:46 UTC
5 points
4 comments6 min readLW link
(neuralblog.github.io)

Neu­roevolu­tion, So­cial In­tel­li­gence, and Logic

vinnik.dmitry0731 May 2023 17:54 UTC
1 point
0 comments10 min readLW link

Un­der­stand­ing mesa-op­ti­miza­tion us­ing toy models

tilmanr, rusheb, Guillaume Corlouer, Dan Valentine, afspies, mivanitskiy and Can
7 May 2023 17:00 UTC
46 points
6 comments10 min readLW link

De­com­piling Tracr Trans­form­ers—An in­ter­pretabil­ity experiment

Hannes Thurnherr27 Mar 2024 9:49 UTC
4 points
0 comments14 min readLW link

An­a­lyz­ing how SAE fea­tures evolve across a for­ward pass

bensenberner, danibalcells, Michael Oesterle, Ediz Ucar and StefanHex
7 Nov 2024 22:07 UTC
47 points
0 comments1 min readLW link
(arxiv.org)

An in­ter­est­ing math­e­mat­i­cal model of how LLMs work

Bill Benzon30 Apr 2024 11:01 UTC
5 points
0 comments1 min readLW link

Vi­su­al­iz­ing small At­ten­tion-only Transformers

WCargo19 Nov 2024 9:37 UTC
4 points
0 comments8 min readLW link

Ex­plor­ing Llama-3-8B MLP Neurons

ntt1239 Jun 2024 14:19 UTC
10 points
0 comments4 min readLW link
(neuralblog.github.io)

Weird Fea­tures in Protein LLMs: The Gram Lens

Jude Stiel14 Jul 2025 17:32 UTC
11 points
0 comments9 min readLW link

Re­search agenda—Build­ing a multi-modal chess-lan­guage model

p.b.7 Apr 2022 12:25 UTC
8 points
2 comments2 min readLW link

Find­ing Back­ward Chain­ing Cir­cuits in Trans­form­ers Trained on Tree Search

abhayesian, Jannik Brinkmann and Victor Levoso
28 May 2024 5:29 UTC
53 points
1 comment9 min readLW link
(arxiv.org)

Trans­form­ers Rep­re­sent Belief State Geom­e­try in their Resi­d­ual Stream

Adam Shai16 Apr 2024 21:16 UTC
437 points
102 comments12 min readLW link1 review

Week One of Study­ing Trans­form­ers Architecture

JustisMills20 Jun 2024 3:47 UTC
3 points
0 comments15 min readLW link
(justismills.substack.com)

So, just why do GPTs have to op­er­ate by con­tin­u­ing an ex­ist­ing string?

Bill Benzon24 Mar 2023 12:08 UTC
−4 points
0 comments3 min readLW link

New Tool: the Resi­d­ual Stream Viewer

AdamYedidia1 Oct 2023 0:49 UTC
32 points
7 comments4 min readLW link
(tinyurl.com)

[Question] Bar­cod­ing LLM Train­ing Data Sub­sets. Any­one try­ing this for in­ter­pretabil­ity?

right..enough?13 Apr 2024 3:09 UTC
7 points
0 comments7 min readLW link

In­ves­ti­gat­ing In­ter­nal Rep­re­sen­ta­tions of Cor­rect­ness in SONAR Text Autoencoders

Samuel Nellessen and antonghawthorne
6 Aug 2025 12:13 UTC
5 points
0 comments7 min readLW link

Trans­form­ers As Am­bi­guity-Re­solv­ing Machines—An en­thu­si­ast’s take

Ravi Khandelwal19 Feb 2026 16:06 UTC
1 point
0 comments2 min readLW link

Char­ac­ter­iz­ing sta­ble re­gions in the resi­d­ual stream of LLMs

Jett Janiak, jacek, Chatrik, Giorgi Giglemiani, nlpet and StefanHex
26 Sep 2024 13:44 UTC
43 points
4 comments1 min readLW link
(arxiv.org)

Adam Op­ti­mizer Causes Priv­ileged Ba­sis in Trans­former LM Resi­d­ual Stream

Diego Caples and rrenaud
6 Sep 2024 17:55 UTC
72 points
7 comments4 min readLW link

No Really, At­ten­tion is ALL You Need—At­ten­tion can do feed­for­ward networks

Robert_AIZI31 Jan 2023 18:48 UTC
29 points
7 comments6 min readLW link
(aizi.substack.com)

At­ten­tion SAEs Scale to GPT-2 Small

Connor Kissane, robertzk, Arthur Conmy and Neel Nanda
3 Feb 2024 6:50 UTC
78 points
4 comments8 min readLW link

Towards Un­der­stand­ing the Rep­re­sen­ta­tion of Belief State Geom­e­try in Transformers

Karthik Viswanathan18 Apr 2025 12:39 UTC
5 points
0 comments12 min readLW link

Trans­form­ers Ex­plained (Again)

RohanS22 Oct 2024 4:06 UTC
4 points
0 comments18 min readLW link

Ex­plor­ing the Resi­d­ual Stream of Trans­form­ers for Mechanis­tic In­ter­pretabil­ity — Explained

Zeping Yu26 Dec 2023 0:36 UTC
7 points
1 comment11 min readLW link

We Need To Know About Con­tinual Learning

michael_mjd22 Apr 2023 17:08 UTC
30 points
14 comments4 min readLW link

In­tro­spec­tion via localization

Victor Godet28 Dec 2025 14:26 UTC
36 points
9 comments3 min readLW link

[Question] Are Mix­ture-of-Ex­perts Trans­form­ers More In­ter­pretable Than Dense Trans­form­ers?

simeon_c31 Dec 2022 11:34 UTC
8 points
6 comments1 min readLW link

GPT-2′s po­si­tional em­bed­ding ma­trix is a helix

AdamYedidia21 Jul 2023 4:16 UTC
52 points
21 comments4 min readLW link

Search­ing for Mo­du­lar­ity in Large Lan­guage Models

NickyP and Stephen Fowler
8 Sep 2022 2:25 UTC
44 points
3 comments14 min readLW link

The po­si­tional em­bed­ding ma­trix and pre­vi­ous-to­ken heads: how do they ac­tu­ally work?

AdamYedidia10 Aug 2023 1:58 UTC
27 points
4 comments13 min readLW link

An Anal­ogy for Un­der­stand­ing Transformers

CallumMcDougall13 May 2023 12:20 UTC
92 points
6 comments9 min readLW link

I Found Catas­tro­phe Geom­e­try in GPT-2′s Resi­d­ual Stream

Karli Joy20 Feb 2026 13:08 UTC
1 point
0 comments11 min readLW link

Trans­form­ers, Intuitively

atharva5 Jan 2026 11:34 UTC
5 points
0 comments4 min readLW link

World, mind, and learn­abil­ity: A note on the meta­phys­i­cal struc­ture of the cos­mos [& LLMs]

Bill Benzon5 Sep 2023 12:19 UTC
4 points
1 comment5 min readLW link

Trans­former Ar­chi­tec­ture Choice for Re­sist­ing Prompt In­jec­tion and Jail-Break­ing Attacks

RogerDearnaley21 May 2023 8:29 UTC
11 points
1 comment4 min readLW link

Has any­one ex­per­i­mented with Do­drio, a tool for ex­plor­ing trans­former mod­els through in­ter­ac­tive vi­su­al­iza­tion?

Bill Benzon11 Dec 2023 20:34 UTC
4 points
0 comments1 min readLW link

[Question] Killing Re­cur­rent Me­mory Over Self At­ten­tion?

Del Nobolo6 Jun 2023 23:02 UTC
3 points
0 comments1 min readLW link

Skep­ti­cism About Deep­Mind’s “Grand­mas­ter-Level” Chess Without Search

Arjun Panickssery12 Feb 2024 0:56 UTC
57 points
13 comments3 min readLW link

De­con­fus­ing In-Con­text Learning

Arjun Panickssery25 Feb 2024 9:48 UTC
37 points
1 comment2 min readLW link

Brief Notes on Transformers

Adam Jermyn26 Sep 2022 14:46 UTC
48 points
3 comments2 min readLW link

Ex­plor­ing vo­cab­u­lary al­ign­ment of neu­rons in Llama-3.2-1B

Sergii7 Jun 2025 11:20 UTC
4 points
0 comments3 min readLW link
(grgv.xyz)

Build­ing a trans­former from scratch—AI safety up-skil­ling challenge

Marius Hobbhahn12 Oct 2022 15:40 UTC
42 points
1 comment5 min readLW link

Google Deep­Mind’s RT-2

SandXbox11 Aug 2023 11:26 UTC
9 points
1 comment1 min readLW link
(robotics-transformer2.github.io)

The Method of Loci: With some brief re­marks, in­clud­ing trans­form­ers and eval­u­at­ing AIs

Bill Benzon2 Dec 2023 14:36 UTC
6 points
0 comments3 min readLW link

The Con­cep­tual To­pog­ra­phy Hy­poth­e­sis: Why Emer­gence in LLMs Isn’t Just About Scale

ravikiran nm6 Jul 2025 13:16 UTC
1 point
0 comments6 min readLW link

How Big a Deal are MatMul-Free Trans­form­ers?

JustisMills27 Jun 2024 22:28 UTC
19 points
6 comments5 min readLW link
(justismills.substack.com)

Monet: Mix­ture of Monose­man­tic Ex­perts for Trans­form­ers Explained

CalebMaresca25 Jan 2025 19:37 UTC
24 points
2 comments11 min readLW link

If lan­guage is for com­mu­ni­ca­tion, what does that im­ply about LLMs?

Bill Benzon12 May 2024 2:55 UTC
10 points
0 comments1 min readLW link

En­ergy-Based Trans­form­ers are Scal­able Learn­ers and Thinkers

Matrice Jacobine8 Jul 2025 13:44 UTC
7 points
5 comments1 min readLW link
(energy-based-transformers.github.io)

Ad­den­dum: More Effi­cient FFNs via Attention

Robert_AIZI6 Feb 2023 18:55 UTC
10 points
2 comments5 min readLW link
(aizi.substack.com)
No comments.