RSS

AI Capabilities

TagLast edit: Aug 29, 2021, 12:57 PM by plex

AI Capabilities are the growing abilities of AIs to act effectively in increasingly complex environments. It is often compared to to AI Alignment, which refers to efforts to ensure that these effective actions taken by AIs are also intended by the creators and beneficial to humanity.

Effi­cien­tZero: hu­man ALE sam­ple-effi­ciency w/​MuZero+self-supervised

gwernNov 2, 2021, 2:32 AM
137 points
52 comments1 min readLW link
(arxiv.org)

[Paper] Stress-test­ing ca­pa­bil­ity elic­i­ta­tion with pass­word-locked models

Jun 4, 2024, 2:52 PM
85 points
10 comments12 min readLW link
(arxiv.org)

A small up­date to the Sparse Cod­ing in­terim re­search report

Apr 30, 2023, 7:54 PM
61 points
5 comments1 min readLW link

Me­moriz­ing weak ex­am­ples can elicit strong be­hav­ior out of pass­word-locked models

Jun 6, 2024, 11:54 PM
58 points
5 comments7 min readLW link

Get­ting 50% (SoTA) on ARC-AGI with GPT-4o

ryan_greenblattJun 17, 2024, 6:44 PM
262 points
50 comments13 min readLW link

Com­pet­i­tive pro­gram­ming with AlphaCode

AlgonFeb 2, 2022, 4:49 PM
58 points
36 comments15 min readLW link
(deepmind.com)

Effi­cien­tZero: How It Works

1a3ornNov 26, 2021, 3:17 PM
297 points
50 comments29 min readLW link1 review

What DALL-E 2 can and can­not do

Swimmer963 (Miranda Dixon-Luinenburg) May 1, 2022, 11:51 PM
353 points
303 comments9 min readLW link

[Cross­post] AlphaTen­sor, Taste, and the Scal­a­bil­ity of AI

jamierumbelowOct 9, 2022, 7:42 PM
16 points
4 comments1 min readLW link
(jamieonsoftware.com)

Is AI Progress Im­pos­si­ble To Pre­dict?

alyssavanceMay 15, 2022, 6:30 PM
277 points
39 comments2 min readLW link

[Question] The thing I don’t un­der­stand about AGI

Jeremy KalfusJun 18, 2024, 4:25 AM
7 points
12 comments1 min readLW link

What will the scaled up GATO look like? (Up­dated with ques­tions)

Amal Oct 25, 2022, 12:44 PM
34 points
22 comments1 min readLW link

The case for a nega­tive al­ign­ment tax

Sep 18, 2024, 6:33 PM
75 points
20 comments7 min readLW link

Deep­Mind on Strat­ego, an im­perfect in­for­ma­tion game

sanxiynOct 24, 2022, 5:57 AM
15 points
9 comments1 min readLW link
(arxiv.org)

Meta AI an­nounces Cicero: Hu­man-Level Di­plo­macy play (with di­alogue)

Jacy Reese AnthisNov 22, 2022, 4:50 PM
93 points
64 comments1 min readLW link
(www.science.org)

Devil’s Ad­vo­cate: Ad­verse Selec­tion Against Con­scien­tious­ness

lionhearted (Sebastian Marshall)May 28, 2023, 5:53 PM
10 points
2 comments1 min readLW link

What’s the Most Im­pres­sive Thing That GPT-4 Could Plau­si­bly Do?

bayesedAug 26, 2022, 3:34 PM
24 points
22 comments1 min readLW link

AI do­ing philos­o­phy = AI gen­er­at­ing hands?

Wei DaiJan 15, 2024, 9:04 AM
46 points
22 comments1 min readLW link

AlphaGeom­e­try: An Olympiad-level AI sys­tem for geometry

alyssavanceJan 17, 2024, 5:17 PM
45 points
9 comments1 min readLW link
(deepmind.google)

Molec­u­lar dy­nam­ics data will be es­sen­tial for the next gen­er­a­tion of ML pro­tein models

Abhishaike MahajanAug 26, 2024, 2:50 PM
9 points
0 comments11 min readLW link
(www.owlposting.com)

Bench­mark­ing LLM Agents on Kag­gle Competitions

aogaraMar 22, 2024, 1:09 PM
15 points
4 comments5 min readLW link

Timelines to Trans­for­ma­tive AI: an investigation

Zershaaneh QureshiMar 26, 2024, 6:28 PM
20 points
2 comments50 min readLW link

“AI achieves silver-medal stan­dard solv­ing In­ter­na­tional Math­e­mat­i­cal Olympiad prob­lems”

gjmJul 25, 2024, 3:58 PM
133 points
38 comments2 min readLW link
(deepmind.google)

Diffu­sion Guided NLP: bet­ter steer­ing, mostly a good thing

Nathan Helm-BurgerAug 10, 2024, 7:49 PM
13 points
0 comments1 min readLW link
(arxiv.org)

o3, Oh My

ZviDec 30, 2024, 2:10 PM
60 points
17 comments36 min readLW link
(thezvi.wordpress.com)

ChatGPT and Bing Chat can’t play Botticelli

Asha SaavossMar 29, 2023, 5:39 PM
11 points
0 comments6 min readLW link

Dual-Use­ness is a Ratio

jimrandomhApr 6, 2023, 5:46 AM
35 points
2 comments1 min readLW link

Ca­pa­bil­ities and al­ign­ment of LLM cog­ni­tive architectures

Seth HerdApr 18, 2023, 4:29 PM
86 points
18 comments20 min readLW link

Read­abil­ity is mostly a waste of characters

vlad.proexApr 21, 2023, 10:05 PM
21 points
7 comments3 min readLW link

[Question] Could trans­former net­work mod­els learn mo­tor plan­ning like they can learn lan­guage and image gen­er­a­tion?

mu_(negative)Apr 23, 2023, 5:24 PM
2 points
4 comments1 min readLW link

The Stochas­tic Par­rot Hy­poth­e­sis is de­bat­able for the last gen­er­a­tion of LLMs

Nov 7, 2023, 4:12 PM
52 points
20 comments6 min readLW link

Re­quest: stop ad­vanc­ing AI capabilities

So8resMay 26, 2023, 5:42 PM
154 points
24 comments1 min readLW link

[Question] Killing Re­cur­rent Me­mory Over Self At­ten­tion?

Del NoboloJun 6, 2023, 11:02 PM
3 points
0 comments1 min readLW link

Elon Musk an­nounces xAI

Jan_KulveitJul 13, 2023, 9:01 AM
75 points
35 comments1 min readLW link
(www.ft.com)

Steer­ing sub­sys­tems: ca­pa­bil­ities, agency, and alignment

Seth HerdSep 29, 2023, 1:45 PM
31 points
0 comments8 min readLW link

Why the tech­nolog­i­cal sin­gu­lar­ity by AGI may never happen

hippkeSep 3, 2021, 2:19 PM
5 points
14 comments1 min readLW link

The al­ign­ment prob­lem in differ­ent ca­pa­bil­ity regimes

BuckSep 9, 2021, 7:46 PM
88 points
12 comments5 min readLW link

Epistemic Strate­gies of Safety-Ca­pa­bil­ities Tradeoffs

adamShimiOct 22, 2021, 8:22 AM
5 points
0 comments6 min readLW link

Google an­nounces Path­ways: new gen­er­a­tion mul­ti­task AI Architecture

OzyrusOct 29, 2021, 11:55 AM
6 points
1 comment1 min readLW link
(blog.google)

In­ter­pret­ing Yud­kowsky on Deep vs Shal­low Knowledge

adamShimiDec 5, 2021, 5:32 PM
100 points
32 comments24 min readLW link

OpenAI Solves (Some) For­mal Math Olympiad Problems

Michaël TrazziFeb 2, 2022, 9:49 PM
78 points
27 comments2 min readLW link

Per­sonal imi­ta­tion software

FlaglandbaseMar 7, 2022, 7:55 AM
6 points
6 comments1 min readLW link

PaLM in “Ex­trap­o­lat­ing GPT-N perfor­mance”

Lukas FinnvedenApr 6, 2022, 1:05 PM
83 points
19 comments2 min readLW link

We have achieved Noob Gains in AI

phdeadMay 18, 2022, 8:56 PM
117 points
20 comments7 min readLW link

[linkpost] The fi­nal AI bench­mark: BIG-bench

RomanSJun 10, 2022, 8:53 AM
25 points
21 comments1 min readLW link

Prin­ci­ples of Pri­vacy for Align­ment Research

johnswentworthJul 27, 2022, 7:53 PM
73 points
31 comments7 min readLW link

The longest train­ing run

Aug 17, 2022, 5:18 PM
71 points
12 comments9 min readLW link
(epochai.org)

[Question] Are lan­guage mod­els close to the su­per­hu­man level in philos­o­phy?

Roman LeventovAug 19, 2022, 4:43 AM
6 points
2 comments2 min readLW link

[Question] What would you ex­pect a mas­sive mul­ti­modal on­line fed­er­ated learner to be ca­pa­ble of?

Aryeh EnglanderAug 27, 2022, 5:31 PM
13 points
4 comments1 min readLW link

No, hu­man brains are not (much) more effi­cient than computers

Jesse HooglandSep 6, 2022, 1:53 PM
22 points
21 comments3 min readLW link
(www.jessehoogland.com)

Alex­aTM − 20 Billion Pa­ram­e­ter Model With Im­pres­sive Performance

MrThinkSep 9, 2022, 9:46 PM
5 points
0 comments1 min readLW link

Eval­u­a­tions pro­ject @ ARC is hiring a re­searcher and a web­dev/​engineer

Beth BarnesSep 9, 2022, 10:46 PM
99 points
7 comments10 min readLW link

[Question] Are Speed Su­per­in­tel­li­gences Fea­si­ble for Modern ML Tech­niques?

DragonGodSep 14, 2022, 12:59 PM
9 points
7 comments1 min readLW link

ACT-1: Trans­former for Actions

Daniel KokotajloSep 14, 2022, 7:09 PM
52 points
4 comments1 min readLW link
(www.adept.ai)

Will we run out of ML data? Ev­i­dence from pro­ject­ing dataset size trends

Pablo VillalobosNov 14, 2022, 4:42 PM
75 points
12 comments2 min readLW link
(epochai.org)

Mas­ter­ing Strat­ego (Deep­mind)

svemirskiDec 2, 2022, 2:21 AM
6 points
0 comments1 min readLW link
(www.deepmind.com)

Can GPT-3 Write Con­tra Dances?

jefftkDec 4, 2022, 3:00 AM
6 points
4 comments10 min readLW link
(www.jefftk.com)

A Year of AI In­creas­ing AI Progress

TW123Dec 30, 2022, 2:09 AM
148 points
3 comments2 min readLW link

[Question] Is “Re­cur­sive Self-Im­prove­ment” Rele­vant in the Deep Learn­ing Paradigm?

DragonGodApr 6, 2023, 7:13 AM
32 points
36 comments7 min readLW link

Lan­guage mod­els can gen­er­ate su­pe­rior text com­pared to their input

ChristianKlJan 17, 2023, 10:57 AM
48 points
28 comments1 min readLW link

Google an­nounces ‘Bard’ pow­ered by LaMDA

M. Y. ZuoFeb 6, 2023, 7:40 PM
31 points
3 comments2 min readLW link

Syd­ney can play chess and kind of keep track of the board state

Erik JennerMar 3, 2023, 9:39 AM
64 points
19 comments6 min readLW link

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

SandXboxMar 7, 2023, 4:11 AM
87 points
7 comments1 min readLW link
(palm-e.github.io)

Squeez­ing foun­da­tions re­search as­sis­tance out of for­mal logic nar­row AI.

Donald HobsonMar 8, 2023, 9:38 AM
16 points
1 comment2 min readLW link

A chess game against GPT-4

Rafael HarthMar 16, 2023, 2:05 PM
24 points
23 comments1 min readLW link

[Question] Is the speed of train­ing large mod­els go­ing to in­crease sig­nifi­cantly in the near fu­ture due to Cere­bras An­dromeda?

Amal Nov 15, 2022, 10:50 PM
13 points
11 comments1 min readLW link

When AI solves a game, fo­cus on the game’s me­chan­ics, not its theme.

Cleo NardoNov 23, 2022, 7:16 PM
88 points
7 comments2 min readLW link

Ques­tions I’d Want to Ask an AGI+ to Test Its Un­der­stand­ing of Ethics

sweenesmJan 26, 2024, 11:40 PM
14 points
6 comments4 min readLW link

A case for ca­pa­bil­ities work on AI as net pos­i­tive

Noosphere89Feb 27, 2023, 9:12 PM
10 points
37 comments1 min readLW link

How should Deep­Mind’s Chin­chilla re­vise our AI fore­casts?

Cleo NardoSep 15, 2022, 5:54 PM
35 points
12 comments13 min readLW link

$300 for the best sci-fi prompt: the results

RomanSJan 3, 2024, 7:10 PM
16 points
19 comments7 min readLW link

Play­ing Dixit with AI: Can AI Sys­tems Iden­tify Misal­ign­ments in My Per­son­al­ized State­ments?

Mariia KoroliukJan 17, 2025, 6:52 PM
1 point
0 comments2 min readLW link

[Question] What’s the differ­ence be­tween newer Atari-play­ing AI and the older Deep­mind one (from 2014)?

RaemonNov 2, 2021, 11:36 PM
27 points
8 comments1 min readLW link

AI Tracker: mon­i­tor­ing cur­rent and near-fu­ture risks from su­per­scale models

Nov 23, 2021, 7:16 PM
67 points
13 comments3 min readLW link
(aitracker.org)

HIRING: In­form and shape a new pro­ject on AI safety at Part­ner­ship on AI

Madhulika SrikumarNov 24, 2021, 8:27 AM
6 points
0 comments1 min readLW link

[Question] Have we seen any “ReLU in­stead of sig­moid-type im­prove­ments” recently

KvmanThinkingNov 23, 2024, 3:51 AM
2 points
4 comments1 min readLW link

How to mea­sure FLOP/​s for Neu­ral Net­works em­piri­cally?

Marius HobbhahnNov 29, 2021, 3:18 PM
16 points
5 comments7 min readLW link

It mat­ters when the first sharp left turn happens

Adam JermynSep 29, 2022, 8:12 PM
45 points
9 comments4 min readLW link

What’s the back­ward-for­ward FLOP ra­tio for Neu­ral Net­works?

Dec 13, 2021, 8:54 AM
20 points
12 comments10 min readLW link

How I’m think­ing about GPT-N

delton137Jan 17, 2022, 5:11 PM
54 points
21 comments18 min readLW link

Es­ti­mat­ing train­ing com­pute of Deep Learn­ing models

Jan 20, 2022, 4:12 PM
37 points
4 comments1 min readLW link

A short pro­ject on Mamba: grokking & interpretability

Alejandro TlaieOct 18, 2024, 4:59 PM
21 points
0 comments6 min readLW link

[Paper] AI Sand­bag­ging: Lan­guage Models can Strate­gi­cally Un­der­perform on Evaluations

Jun 13, 2024, 10:04 AM
84 points
10 comments2 min readLW link
(arxiv.org)

Un­com­pet­i­tive pro­gram­ming with GPT-3

BezziFeb 6, 2022, 10:19 AM
7 points
8 comments3 min readLW link

Anony­mous ad­vice: If you want to re­duce AI risk, should you take roles that ad­vance AI ca­pa­bil­ities?

Benjamin HiltonOct 11, 2022, 2:16 PM
54 points
9 comments1 min readLW link

Test­ing PaLM prompts on GPT3

YitzApr 6, 2022, 5:21 AM
103 points
14 comments8 min readLW link

Is GPT-N bounded by hu­man ca­pa­bil­ities? No.

Cleo NardoOct 17, 2022, 11:26 PM
49 points
8 comments2 min readLW link

What’s the fu­ture of AI hard­ware?

Itay DreyfusJun 17, 2024, 1:05 PM
2 points
0 comments8 min readLW link
(productidentity.co)

Gato’s Gen­er­al­i­sa­tion: Pre­dic­tions and Ex­per­i­ments I’d Like to See

Oliver SourbutMay 18, 2022, 7:15 AM
43 points
3 comments10 min readLW link

They gave LLMs ac­cess to physics simulators

ryan_bOct 17, 2022, 9:21 PM
50 points
18 comments1 min readLW link
(arxiv.org)

An In­tro­duc­tion to AI Sandbagging

Apr 26, 2024, 1:40 PM
45 points
13 comments8 min readLW link

[Question] What is the most prob­a­ble AI?

Zeruel017Jun 20, 2022, 11:26 PM
−2 points
0 comments3 min readLW link

Agen­tized LLMs will change the al­ign­ment landscape

Seth HerdApr 9, 2023, 2:29 AM
160 points
102 comments3 min readLW link1 review

AI Fore­cast­ing: One Year In

jsteinhardtJul 4, 2022, 5:10 AM
132 points
12 comments6 min readLW link
(bounded-regret.ghost.io)

Sta­bil­ity AI re­leases StableLM, an open-source ChatGPT counterpart

OzyrusApr 20, 2023, 6:04 AM
11 points
3 comments1 min readLW link
(github.com)

A Cri­tique of AI Align­ment Pessimism

ExCephJul 19, 2022, 2:28 AM
9 points
1 comment9 min readLW link

Align­ment be­ing im­pos­si­ble might be bet­ter than it be­ing re­ally difficult

Martín SotoJul 25, 2022, 11:57 PM
13 points
2 comments2 min readLW link

On agen­tic gen­er­al­ist mod­els: we’re es­sen­tially us­ing ex­ist­ing tech­nol­ogy the weak­est and worst way you can use it

Yuli_BanAug 28, 2024, 1:57 AM
10 points
2 comments9 min readLW link

[Thought Ex­per­i­ment] To­mor­row’s Echo—The fu­ture of syn­thetic com­pan­ion­ship.

Vimal NaranOct 26, 2023, 5:54 PM
−7 points
2 comments2 min readLW link

AI as Su­per-Demagogue

RationalDinoNov 5, 2023, 9:21 PM
0 points
11 comments9 min readLW link

Ar­ti­cle Re­view: Google’s AlphaTensor

Robert_AIZIOct 12, 2022, 6:04 PM
8 points
4 comments10 min readLW link

A call for a quan­ti­ta­tive re­port card for AI bioter­ror­ism threat models

JunoDec 4, 2023, 6:35 AM
12 points
0 comments10 min readLW link

Pre­dict 2025 AI ca­pa­bil­ities (by Sun­day)

Jan 15, 2025, 12:16 AM
54 points
3 comments1 min readLW link

GPT4 is ca­pa­ble of writ­ing de­cent long-form sci­ence fic­tion (with the right prompts)

RomanSMay 23, 2023, 1:41 PM
22 points
28 comments65 min readLW link

AGI-Au­to­mated In­ter­pretabil­ity is Suicide

__RicG__May 10, 2023, 2:20 PM
25 points
33 comments7 min readLW link

GPT-4 im­plic­itly val­ues iden­tity preser­va­tion: a study of LMCA iden­tity management

OzyrusMay 17, 2023, 2:13 PM
21 points
4 comments13 min readLW link

Paper: Dis­cov­er­ing novel al­gorithms with AlphaTen­sor [Deep­mind]

LawrenceCOct 5, 2022, 4:20 PM
82 points
18 comments1 min readLW link
(www.deepmind.com)

INTELLECT-1 Re­lease: The First Globally Trained 10B Pa­ram­e­ter Model

Matrice JacobineNov 29, 2024, 11:05 PM
16 points
1 comment1 min readLW link
(www.primeintellect.ai)

TinyS­to­ries: Small Lan­guage Models That Still Speak Co­her­ent English

Ulisse MiniMay 28, 2023, 10:23 PM
66 points
8 comments2 min readLW link
(arxiv.org)

Lifel­og­ging for Align­ment & Immortality

Dev.ErrataAug 17, 2024, 11:42 PM
13 points
3 comments7 min readLW link

Notes on Meta’s Di­plo­macy-Play­ing AI

Erich_GrunewaldDec 22, 2022, 11:34 AM
14 points
2 comments14 min readLW link
(www.erichgrunewald.com)

[Question] Hy­po­thet­i­cal: what would you do?

JNSAug 3, 2023, 10:39 PM
4 points
2 comments1 min readLW link

LLMs are (mostly) not helped by filler tokens

Kshitij SachanAug 10, 2023, 12:48 AM
66 points
35 comments6 min readLW link

In­flec­tion.ai is a ma­jor AGI lab

Nikola JurkovicAug 9, 2023, 1:05 AM
137 points
13 comments2 min readLW link

Google Deep­Mind’s RT-2

SandXboxAug 11, 2023, 11:26 AM
9 points
1 comment1 min readLW link
(robotics-transformer2.github.io)

Stu­pidity is also hard

walkthroughwallsSep 12, 2023, 2:45 AM
−8 points
4 comments2 min readLW link

Ba­sic Math­e­mat­ics of Pre­dic­tive Coding

Adam ShaiSep 29, 2023, 2:38 PM
49 points
6 comments9 min readLW link

Towards Bet­ter Mile­stones for Mon­i­tor­ing AI Capabilities

snewmanSep 27, 2023, 9:18 PM
11 points
0 comments14 min readLW link

[Question] How might we make bet­ter use of AI ca­pa­bil­ities re­search for al­ign­ment pur­poses?

Jemal YoungAug 31, 2022, 4:19 AM
11 points
4 comments1 min readLW link

[Question] Is there a pub­li­cly available list of ex­am­ples of fron­tier model ca­pa­bil­ities?

Max KearneySep 19, 2023, 5:45 PM
1 point
0 comments1 min readLW link

In­ter­pretabil­ity Ex­ter­nal­ities Case Study—Hun­gry Hun­gry Hippos

Magdalena WacheSep 20, 2023, 2:42 PM
64 points
22 comments2 min readLW link

This anime sto­ry­board doesn’t ex­ist: a graphic novel writ­ten and illus­trated by GPT4

RomanSOct 5, 2023, 2:01 PM
12 points
7 comments55 min readLW link

I Would Have Solved Align­ment, But I Was Wor­ried That Would Ad­vance Timelines

307thOct 20, 2023, 4:37 PM
121 points
33 comments9 min readLW link

Eleuther re­leases Llemma: An Open Lan­guage Model For Mathematics

mako yassOct 17, 2023, 8:03 PM
22 points
0 comments1 min readLW link
(blog.eleuther.ai)

[Question] What are the rel­a­tive speeds of AI ca­pa­bil­ities and AI safety?

NunoSempereApr 24, 2020, 6:21 PM
8 points
2 comments1 min readLW link

Deep­Mind: Gen­er­ally ca­pa­ble agents emerge from open-ended play

Daniel KokotajloJul 27, 2021, 2:19 PM
247 points
53 comments2 min readLW link
(deepmind.com)

OpenAI Codex: First Impressions

specbugAug 13, 2021, 4:52 PM
49 points
8 comments4 min readLW link
(sixeleven.in)

To con­tribute to AI safety, con­sider do­ing AI research

VikaJan 16, 2016, 8:42 PM
39 points
39 comments2 min readLW link