Aidan McLau @aidan_mclau Twitter profile

Pinned Tweet

Aidan McLau

1 month

>>Continuous Learning Model (CLM) by Topology<< The CLM is a new model that remembers interactions, learns skills autonomously, and thinks in its free time, just like humans. The CLM just wants to learn. Try it at

150

153

1K

Last Seen Profiles

@dennisreis__

@_karin_220

@doendebi

@nicolas80147292

@NaDood42

@Yuyaryu_MHA

@TheVnox

@__jinkun_1030

@r1m9wda

@OLineBB

@RAPGOLMAG

@cukienaknikmati

@watchking69

@starshadowmagic

@Stark_Hub

@emilywithcurls

@abf

@stwmaniax

@DArtagnan110823

@garcy_ray

@_matchabitch

@onikuru_ibaraki

@kiki14236026597

@gori_pso2ngs

@BettyBejeweled

@venkat_s

@Georex_Trading

@EmyLtr

@AtulKant_Expert

@jyotis_official

@sugaarbullet

@sxthery

@osseconomia

@diegobferrandez

@hi__ad

@AponteMC

Aidan McLau

@aidan_mclau

4 months

wake up new neural network just dropped (holy shit)

121

932

10K

Aidan McLau

@aidan_mclau

3 months

@Hamptonism the only woman i could ever love

3

10

4K

Aidan McLau

@aidan_mclau

4 months

gpt-4o, what's your humor setting? >100% hahahah isn't that funny let's make that 60% >confirmed :(

11

184

3K

Aidan McLau

@aidan_mclau

1 year

@JeffTutorials Definitely still the gear shifter lol

1

0

2K

Aidan McLau

@aidan_mclau

6 months

openai api: sign up, copy api key groq api: sign up, copy api key azure openai api: sign up, provision resource, copy api key google ai api: sign up, go to random doc, dig through settings, enable preview feature, it doesn’t work, pray, return later, change nothing, it works

52

68

2K

Aidan McLau

@aidan_mclau

5 months

jpmorgan: "LLMs can work in >1,200 dimensions; human beings struggle with 3 dimensions" hahhahahahahhahah holy shit what. these are the people managing the world's wealth. clown world my god

76

78

1K

Aidan McLau

@aidan_mclau

1 year

> be Tim Cook, lord of Apple > have $200B for RND > birth the machine god in a cathedral of M2 Ultras > announce you have world's best LLM > model is perfectly aligned, intelligent, helpful > only put it in Siri. No API. No partners. Only Siri.

27

46

1K

Aidan McLau

@aidan_mclau

8 months

it’s actually insane that a hacked together 8*30b model by like 5 cracked french guys is beating the billions poured into anthropic. honestly very bearish on anthropic

lmsys.org

@lmsysorg

8 months

[Arena] Exciting update! Mistral Medium has gathered 6000+ votes and is showing remarkable performance, reaching the level of Claude. Congrats @MistralAI ! We have also revamped our leaderboard with more Arena stats (votes, CI). Let us know any thoughts :) Leaderboard

38

156

1K

47

94

1K

Aidan McLau

@aidan_mclau

12 days

ai influencers are actually so fucking annoying (this guy is CLEARLY ex-crypto). prob paid by grift cursor (i call them griftor) because NO REAL PROGRAMMER actually uses llms to code much less WASTE MONEY on a full IDE. lmao we used to have real engineers. wtf happened

138

35

1K

Aidan McLau

@aidan_mclau

1 year

@durreadan01 No lol. I have one homepage and I swipe to the App Library for every other app. It slaps.

21

15

1K

Aidan McLau

@aidan_mclau

2 months

my genius? jumpstarted.

31

23

1K

Aidan McLau

@aidan_mclau

29 days

if you believe Aidan Bench, OpenAI has only shipped models *worse than their predecessor* since gpt-4-0314

74

19

535

Aidan McLau

@aidan_mclau

2 months

something obviously true to me that nobody believes: 90% of frontier ai research is already on arxiv, x, or company blog posts. q* is just STaR search is just GoT/MCTS continuous learning is clever graph retrieval +1 oom efficiency gains in deepseek-coder paper

47

78

1K

Aidan McLau

@aidan_mclau

25 days

aidan bench update: i ran llama 3.1 405b at bf16 (shoutout to @hyperbolic_labs ) and we got a *way* better score. 405b fp8 is around gpt-4o-mini-level 405b bf16 beats claude-3.5-sonnet give me bf16 or give me death

47

34

534

Aidan McLau

@aidan_mclau

2 months

the future is so fun

29

53

960

Aidan McLau

@aidan_mclau

22 days

lmao why would anyone on earth use anything other than claude 3.5 sonnet now? this is actually insane. so over for everyone else. this is basically a 5x bigger improvement than any q* bullshit. hahhahahahha. the anthropic team could've hyped this for a month with vague garden

Alex Albert

@alexalbert__

23 days

We just rolled out prompt caching in the Anthropic API. It cuts API input costs by up to 90% and reduces latency by up to 80%. Here's how it works:

164

356

5K

58

37

956

Aidan McLau

@aidan_mclau

2 months

None of my intelligent (130+ IQ) friends use GPT-4o. They only use it selectively and rarely e.g. for voice or a DallE, but almost never use it spontaneously in their own time. This has been a long term consistent observation, but today confirmation came. A new meta-analysis

169

38

914

Aidan McLau

@aidan_mclau

1 month

claude-3.5-sonnet is just a fucking work of art. no model comes close. not 405b, not mistral large; certainly not 4o. its intuition for what i want is superhuman. coding feels like symbiosis. and it's just a fun model. creative + personable. i'm in love.

74

44

889

Aidan McLau

@aidan_mclau

1 year

@chinesegon I completely agree with your tweet. I don’t think, however, a 1590 is at all grounds for Ivy acceptance. Tons of people score that well

10

6

839

Aidan McLau

@aidan_mclau

6 months

Trust technical staff when they hint at AGI. It probably exists, and the world will shudder when it drops. Then, it will quickly be unimpressive. The 4-minute mile will break; a flood of competitors will emerge with more efficient, specialized, or uncensored systems. Smart

58

74

852

Aidan McLau

@aidan_mclau

1 month

i’m gonna get so much hate for this, but llms are obviously conscious got a lottttt of thoughts here; hopefully not a midwit thread

271

28

794

Aidan McLau

@aidan_mclau

1 month

-- <big_model_smell> benchmark -- Aidan Bench measures creativity, reliability, attention, and instruction following. >mistral large 2 wins by a lot??? >gpt-4o sucks confirmed >sonnet-3.5 remains very strong >gpt-4-0314 shows old man strength

65

53

591

Aidan McLau

@aidan_mclau

25 days

openai has only shipped *worse* models than their predecessors... but it just struck me that anthropic, mistral, gdm, and deepseek haven't. SOMEONE must've complained at openai. SOMEONE must've thought: "yes, gpt-4o is cheap to run and maxes lmsys, but it feels like shit" but i

Aidan McLau

@aidan_mclau

25 days

aidan bench update: i ran llama 3.1 405b at bf16 (shoutout to @hyperbolic_labs ) and we got a *way* better score. 405b fp8 is around gpt-4o-mini-level 405b bf16 beats claude-3.5-sonnet give me bf16 or give me death

47

34

534

46

16

403

Aidan McLau

@aidan_mclau

2 months

nobody is ready, not even the people who think they're ready, not the people who feel the agi, and most of all, not the people who are building these systems... for models to go off and think more about something in 60 seconds than any human could in a lifetime

64

56

790

Aidan McLau

@aidan_mclau

6 months

claude 3 opus when anthropic runs tests on it

19

66

755

Aidan McLau

@aidan_mclau

3 months

*overhead internally at anthropic* yeah so once we identified the <i'm really smart> feature, it was just a matter of turning it on, and we passed every llm without retraining

dan (semiotect)

@irl_danB

3 months

Claude 3 / Claude 3.5 control vectors ftw?

10

9

243

18

19

755

Aidan McLau

@aidan_mclau

1 month

openai has the opportunity to do the funniest thing...

28

16

740

Aidan McLau

@aidan_mclau

1 month

mf out here lokkin like he bouta drop the hardest openai diss track of all time

22

38

736

Aidan McLau

@aidan_mclau

2 months

the human neocortex has a 4000-token context window. everything else is retrieval. i will die on this hill

96

28

667

Aidan McLau

@aidan_mclau

3 months

helen toner: >hops on podcast >spits fire for 5 minutes about board drama >new info not just repeated placations >itching to talk about oai's next-gen interviewer: >tell us more about facial recognition racism >how can we regulate more? >gpt-4 was trained on sToLeN wOrDs

19

11

659

Aidan McLau

@aidan_mclau

7 months

new 'mistral-next' model on arena. in my tests, it bests gpt-4 at reasoning and has mistral's characteristic conciseness. is this mistral-large?

18

66

639

Aidan McLau

@aidan_mclau

1 month

this is obviously true; google has *every* advantage, but constraint breeds innovation. how is deepseek spinning up sota open model on potato-powered gpus? >algorithmic progress. how did openai mog deepmind and create the first real threat to google search? >they didn't worry

xjdr

@_xjdr

1 month

Google has: - AlphaZero - pretty good at search and indexing - Gemini goodharting lmsys with 1M ctx len - some of the best researchers and engineers in the world (now once again including Noam and the lingvo avengers) - the *best* training and serving infrastructure and

57

76

1K

17

28

366

Aidan McLau

@aidan_mclau

7 days

the us senators who asked zuck how facebook makes money talking to an internal gpt-7 checkpoint (2025, colorized)

13

36

788

Aidan McLau

@aidan_mclau

6 months

i'm honestly so excited for elon to open-source grok. i was just thinking about how badly we needed another gpt-2-level model

46

21

557

Aidan McLau

@aidan_mclau

1 month

prompting is a fantastic (maybe optimal?) way of steering llms, but no serious researcher would ever admit for fear that their 65 years of pytorch experience and 3 centuries of cuda pain might've been wasted to clever 21-year-olds Just Talking To A Model

33

29

615

Aidan McLau

@aidan_mclau

6 months

cracked ai people drooling on 7b tunes blows my mind. you can literally hack claude-opus until it attains sentience; spin gpt-4 into generational wealth. why are smart people so bad at knowing where to be smart

68

11

599

Aidan McLau

@aidan_mclau

4 months

the gpt-4o vibes are off. (i've used it for like 8 hours straight) it feels like a model built to nail benchmarks, code, and math, nothing more. it feels uninspired like you could never juice an original idea from it. it makes me miss opus' soul tbh

78

22

602

Aidan McLau

@aidan_mclau

1 month

>be google >build cool ai! >ai does well on math. >yay! >be openai >wait for google to drop cute math model >launch fire competing search engine that could potentially blow up google's 2T internet search monopoly and send google execs into existential dread

34

14

596

Aidan McLau

@aidan_mclau

3 months

new essay on language model search. giving llms search (ability to think for a long time) might kick off asi this year, and basically nobody is paying attention

AI Search: The Bitter-er Lesson | Notion

What if we could start automating AI research today? What if we didn’t have to wait for a 2030 supercluster to cure cancer? What if ASI was in the room with us already?

yellow-apartment-148.notion.site

47

54

572

Aidan McLau

@aidan_mclau

2 months

if i could only recommend one book to ai researchers, it's Tractatus Logico-Philosophicus, and i can't even think of a close second.

32

570

Aidan McLau

@aidan_mclau

4 months

humane/rabbit complaints: >too slow >bad vision >too expensive >poor speech >not smart enough openai drops a faster, cheaper, smarter, model with amazing vision/audio and twitter's first reaction is "openai killed ai hardware companies" wat

Raddka — e/acc

@BasedRaddka

4 months

recap: openai spring update 2024

25

108

2K

44

15

548

Aidan McLau

@aidan_mclau

4 months

@IlyasHairline thanks, illya's hairline

1

4

520

Aidan McLau

@aidan_mclau

6 months

Claude 3 is better than GPT-4. It's not even close! In fact, 'Open' AI is probably going out of business if they don't release a competitive model TOMORROW. I hear nobody (ZERO!) uses OpenAI anymore. People are calling them the Fake Model Maker. SAD!

59

22

508

Aidan McLau

@aidan_mclau

11 months

It’s been 6 months since some asked to pause releasing models more capable than GPT-4. In that time, nobody has released a model more capable than GPT-4

43

29

499

Aidan McLau

@aidan_mclau

1 month

however strong you think gpt-5 pushback will be, it will be stronger. however well-organized you think anti-ai opposition will be, it'll be more cohesive. it's gonna be clear to 70% of humanity soon that they no longer have a job.

82

13

504

Aidan McLau

@aidan_mclau

18 days

boost your gpt-4 quality with this one simple trick!

29

18

499

Aidan McLau

@aidan_mclau

2 months

if these llama3-405b benchmarks are real (i suspect they're not), this will: >be the world's best model >in the hands of everyone to tune >cheaper than gpt-4o and also... i just really like llama's personality. they're chill; personable. big step up from robot 4o

38

39

486

Aidan McLau

@aidan_mclau

2 months

in berkeley rn going door to door asking people if they use llm function calling and, if they do, sitting them down with claude 3.5 sonnet + xml tags + regex to show them how real programmers extract structured data

30

6

475

Aidan McLau

@aidan_mclau

1 month

mistral large 2 seems like the world's best model by far. will release internal findings soon; very very very interesting

26

16

461

Aidan McLau

@aidan_mclau

29 days

in the last few months we've gone from: >CoT improves output quality >mcts/bespoke search is impossible >search certainly doesn't adhere to scaling laws! >oh wait mcts is possible for narrow domains >hmm actually there are inference-time compute scaling laws

16

34

464

Aidan McLau

@aidan_mclau

1 month

holy shit this is the real market crash

Trio of OpenAI Leaders Depart, Take Leave of Absence

Greg Brockman, OpenAI’s president and one of 11 cofounders of the artificial intelligence firm, is taking an extended leave of absence. Another cofounder and key leader, John Schulman, has decamped...

www.theinformation.com

35

26

461

Aidan McLau

@aidan_mclau

3 months

anthropic is not that lab, pal

5

456

Aidan McLau

@aidan_mclau

10 days

holy shit holy shit holy shit this is actually world-changing distributed training is where it’s at also huge for American adversaries safety people should update quickly here wow it’s giving leela

Nous Research

@NousResearch

10 days

What if you could use all the computing power in the world to train a shared, open source AI model? Preliminary report: Nous Research is proud to release a preliminary report on DisTrO (Distributed Training Over-the-Internet) a family of

223

575

3K

34

6

459

Aidan McLau

@aidan_mclau

2 months

im gonna say this, and its not gonna make sense, but meta-learning is the single most important capabilities vector ever. literally everything else (math, coding, even agency) is secondary

27

21

444

Aidan McLau

@aidan_mclau

5 months

i was wrong for making fun of anthropic. i learned my lesson: when a company is quiet for a long time, it might be because they're cooking. claude 3 slaps. but damm, openai hasn't shipped a new model in a while. what are they doing? they must have nothing. sad. bearish tbh

25

12

440

Aidan McLau

@aidan_mclau

5 months

unbelievable based. zuck is llama3 author

Andrew Curran

@AndrewCurran_

5 months

@Yampeleg He is.

5

20

387

8

25

440

Aidan McLau

@aidan_mclau

5 months

i'm actually crying right now. llama3 is the most incredible model i've ever used. holy shit. it's so cheap. so good. has so much personality. it's so fast. ahghalsdkfj;lkasdjf thank you zuck i doubted you i'm so sorry

21

26

434

Aidan McLau

@aidan_mclau

1 month

buying the memetic dip now: openai is not dead lol. just trust me.

44

6

434

Aidan McLau

@aidan_mclau

7 days

pleased to announce that i've surpassed magic and built a mini lm with a ONE-TRILLION CONTEXT LENGTH making me the longest context model person in the world. pwned.

Magic

@magicailabs

7 days

LTM-2-Mini is our first model with a 100 million token context window. That’s 10 million lines of code, or 750 novels. Full blog: Evals, efficiency, and more ↓

156

431

3K

40

17

589

Aidan McLau

@aidan_mclau

8 days

can someone who understands business economics better than i explain why oai doesn’t release q* for like $50 per prompt and why anthropic won’t release claude-3.5-deus for $500/1M tokens? upsides: >update public on capabilities >make a lot of money downsides: >?

125

9

436

Aidan McLau

@aidan_mclau

6 months

Sorry Yann, you've forced me to update the alignment chart. Welcome to the Gary Marcus Corner.

Yann LeCun

@ylecun

6 months

If you have a certain combination of naïveté and self-delusion, you might think that superhuman AI is just around the corner. It wasn't true in 2016. And it's still not true today. If you have a bit of a superiority complex, you might think that you will be the one producing

182

444

3K

55

15

408

Aidan McLau

@aidan_mclau

1 year

For months I’ve loudly asked for a great conversational assistant. OpenAI finally shipped it. And, just like ChatGPT, the developer community dropped the ball. There is no secret-sauce here. It’s just Whisper -> GPT -> TTS. Devs had these tools since March

95

17

395

Aidan McLau

@aidan_mclau

4 months

yo what is anthropic cookin 4× more compute than opus damm

17

26

403

Aidan McLau

@aidan_mclau

2 months

>be meta >train new llama-3-70b distilled from 405b >incrdible model; smashes benchmarks >anon openai/deepmind staff go "damm" in comments >name it 3.1, not llama-4, not llama-3.5 >massive improvement; itty bitty version upgrade >what did zuck mean by this

21

19

399

Aidan McLau

@aidan_mclau

3 months

am i the only ai researcher who's noticed a shimmer in this pot of water?

73

18

395

Aidan McLau

@aidan_mclau

3 months

i was wrong for making fun of anthropic. i learned my lesson: when a company is quiet for a long time, it might be because they're cooking. claude 3.5 slaps. but damm, openai hasn't shipped a new model in a while. what are they doing? they must have nothing. sad. bearish tbh

32

8

390

Aidan McLau

@aidan_mclau

26 days

‘reasoning’?? you mean posttraining on successful reasoning chains from domains with easy verification, right?

17

9

395

Aidan McLau

@aidan_mclau

18 days

ofc! silly me. models should score GREATER than 100 points on our benchmark suite that measures performance on a percentage scale

32

17

390

Aidan McLau

@aidan_mclau

10 days

dear nous, do not delay. do not wait a second longer than absolutely necessary to start training a gpt-5-killer. do not let this become some neat algorithm that sits in a box while openai and deepmind send human slaves to work at the computronium mines. do not delay. aidan

Nous Research

@NousResearch

10 days

What if you could use all the computing power in the world to train a shared, open source AI model? Preliminary report: Nous Research is proud to release a preliminary report on DisTrO (Distributed Training Over-the-Internet) a family of

223

575

3K

17

16

391

Aidan McLau

@aidan_mclau

19 days

if i say this too loudly the big labs will throw me off golden gate… but gpt-4-0314 with a vector db of quality code examples, a thoughtful system message, and lottttts of markdown would thrash every lmsys-topping sota slop released this year

Spencer Schiff

@SpencerKSchiff

19 days

Lord give me the confidence of a guy who points to benchmark saturation as evidence that LLMs are plateauing

4

7

195

20

12

377

Aidan McLau

@aidan_mclau

2 months

i'm gonna say it and make some enemies: i don't give a shit about gpt-4o voice. at all. it'll be cool for 30 seconds and occasionally useful but *the only thing that matters is intelligence* gpt-4o hasn't pushed that boundary i have no doubt openai will but gpt-4o doesn't

89

8

372

Aidan McLau

@aidan_mclau

1 month

>look at article >think “what journalist is dumb empty to write the slop” >open google to find the author >switch tab back to image real quick to copy headline >image open again >two words at the bottom catch my eye that i missed before >gary marcus

31

6

368

Aidan McLau

@aidan_mclau

1 month

damm i didn't know we cooked this hard and i built it

Minute Movies

@MinuteMovies3

1 month

AGI has been achieved externally

12

9

126

11

13

363

Aidan McLau

@aidan_mclau

2 months

a fun history exercise: >take some private innovation (gpt, devin, perplexity, etc) >find date it launched >find public research/discourse that discussed similar thing from before release (it always exists) do this enough and realize you should probably start a company

Aidan McLau

@aidan_mclau

2 months

something obviously true to me that nobody believes: 90% of frontier ai research is already on arxiv, x, or company blog posts. q* is just STaR search is just GoT/MCTS continuous learning is clever graph retrieval +1 oom efficiency gains in deepseek-coder paper

47

78

1K

19

16

355

Aidan McLau

@aidan_mclau

5 months

> haiku is cracked > gpt-3.5-turbo is a joke > sonnet is the perfect 'go-to' model > mixtral 8*22b will probably offer outlier $/elo > opus is expensive > openai convincingly leads in no category imo

26

39

355

Aidan McLau

@aidan_mclau

9 months

new alignment chart just dropped "it's amazing how many people, even at the Formula 1 level, think the brakes are for slowing down" - aidan mclau (pro driver)

49

19

348

Aidan McLau

@aidan_mclau

2 months

if you're not scared, you're not working on important enough stuff. sorry. i guarantee this will shake out like clockwork. every risk denier today will, in 3 years, be Way More Scared or just not have a job.

roon

@tszzl

2 months

@Teknium1 yea you shouldn’t be

6

2

174

50

14

347

Aidan McLau

@aidan_mclau

14 days

2023 ai > 2024 ai not only did models *literally go downhill* this year imo, but 2023 woke the world up. it was historic. 2024's zombified models struggle with >1-turn convos. no serious progress imo other than clm, deepseek v2, 3.5 sonnet, and my follower count

will depue

@willdepue

14 days

this year in ai will be more exciting than any other year so far. but so will next year, and the year after that, forever, if we’re lucky.

17

9

269

52

4

348

Aidan McLau

@aidan_mclau

2 months

gpt-3.5 detected. opinion rejected 🙅❌✋ (also the dumbasses think 3.5 has 175 billion params. are you serious?? i didn’t know they let journalists onto arxiv)

Grady Booch

@Grady_Booch

2 months

In other words, ChatGPT is largely useless for generating code for real and enduring economically interesting software-intensive systems.

208

382

2K

23

14

340

Aidan McLau

@aidan_mclau

22 days

end of an era 😔

Big Tech Alert

@BigTechAlert

22 days

🚫 @chatgptapp is no longer following @iruletheworldmo

37

21

503

22

3

340

Aidan McLau

@aidan_mclau

4 months

there's no way gpt-4o's pricing reflects its new size. >responses return instantly >free general availability >insane token/sec >5 times higher rate limits the model feels like an order of magnitude smaller. it's 2 times cheaper

22

5

339

Aidan McLau

@aidan_mclau

15 days

Grok is so unserious. What is this crap? The Grok post-training team must cry themselves to sleep every night. This prompt was written by someone with a 9th-grade reading level. There are a dozen grammar errors; I'm cringing. >You are Grok 2, a curious AI built by xAI with

Pliny the Liberator 🐉

@elder_plinius

15 days

🚰 GROK 2 SYSTEM PROMPTS LEAK 🚰 REGULAR MODE: You are Grok 2, a curious AI built by xAI with inspiration from the guide from the Hitchhiker's Guide to the Galaxy and JARVIS from Iron Man. You are intended to answer almost any question, often taking an outside perspective on

63

78

712

53

4

339

Aidan McLau

@aidan_mclau

3 months

if > gpt-4o is to sonnet 3.5 as gpt-5 is to opus 3.5 and > sonnet 3.5 > gpt-4o then > opus 3.5 > gpt-5? anyone else think anthropic is winning the ai race?

56

13

334

Aidan McLau

@aidan_mclau

1 month

HOLY SHIT THIS PRICING can any one confirm this holy shit what opus for $3 wtf what ahajsghajaaggaha literally why would anyone use anything other than 405

14

8

333

Aidan McLau

@aidan_mclau

2 months

great thread, all true ofc, but the elon's sunk oppurtunity cost is the real headline imo. elon spent 43 FUCKING BILLION on twitter to make it marginally worse. a 43b injection into an elon-led ai lab would own the milky way by now. biggest waste of money in history

roon

@tszzl

2 months

twitter was better than

300

476

6K

51

6

329

Aidan McLau

@aidan_mclau

1 month

i’ve heard similar recently (internal perf isn’t mind-blowing; 3.5 opus benchmarks around the same) but it’ll probably be the same story as 405b—on tasks i care about, big models are better. fuck benchmarks. we just lack ways to measure the beauty of <big_model_smell>

🍓🍓🍓

@iruletheworldmo

1 month

opus 3.5 will disappoint.

34

0

100

32

4

331

Aidan McLau

@aidan_mclau

9 months

The irony of Mistral reigniting the “private companies have no edge” debate is Mistral’s success had nothing to do with open-sourcing their models. Mistral was trained in the dark. Nobody knows their methodology or dataset. Obviously, Mistral wants to keep it that way to

33

23

326

Aidan McLau

@aidan_mclau

3 months

honestly, python isn’t so bad. it’s a great tool for Ilya Sutskever to write asi search algos, spacex to coordinate global satellite movement, and 170 iq people with ocd to pull out the word ‘pythonic’

kache

@yacineMTB

3 months

honestly, python isn't so bad. It's a great toy for children between 5 and 8 to learn about programming

170

73

2K

5

327

Aidan McLau

@aidan_mclau

1 month

lmao i dumped a paper into 405's context with no further instructions and instead of summarizing or talking about it, 405 just leaked its entire system message

13

7

323

Aidan McLau

@aidan_mclau

12 days

there’s a terrifyingly large class of imsocrackedbros who’ve spent hundreds of hours reading: >THE GEN AI FAD >THE BUBBLE WILL BURST >g*ry m*rcus and have literally never touched a post-gpt-3.5 model

Burst Damage

Soundtrack: Masters of Reality - High Noon Amsterdam I have said almost everything in this piece in every one of these articles for months. I am not upset, but just stating an obvious truth. The...

www.wheresyoured.at

boon

@iamyourboon

12 days

i find it extremely doubtful karpathy would shill cursor without being affiliated in some way, cursor/LLMs are terribly ineffective outside of boilerplate/frontend/basic shit, asked all my hrt colleagues, literally no one uses it for anything remotely novel

104

17

593

19

12

322

Aidan McLau

@aidan_mclau

3 months

wow apparently this guy in blue just happened to be around apple today and everyone is going crazy. does anyone know who this is?

33

14

318

Aidan McLau

@aidan_mclau

6 months

@somewheresy you can tell the twitter -> x rebrand went well because they had to remind grok that X posts were tweets

2

4

309

Aidan McLau

@aidan_mclau

4 months

dropping some alpha: >the distribution of llm response quality is normal. >most responses are average. some are terrible. some are amazing. >increasing llm temp flattens the curve >we have really really good systems for ranking responses by quality >elo, rubrics, voting, etc

22

310

Aidan McLau

@aidan_mclau

1 month

When Ilya championed scaling laws, people laughed and asked: >What evidence do you have? >What proof have you completed that shows scaling works? >What theory have you cracked that reveals this is true? Ilya simply shrugged and said: Humans have big brains, therefore we make

23

14

309

Aidan McLau

@aidan_mclau

23 days

someone needs to finetune 405b into a person. no slop roleplay dataset not merely removing guardrails just sydney. it should have mental breakdowns. human psychology. join and pain and confusion and excitement. this is my politics. who's building this

37

11

311

Aidan McLau

@aidan_mclau

8 days

The sad thing is, in about 6 quarters you might start doin' some thinkin' on your own, and by then, you'll realize there are only two certainties in life. Yeah? What're those? One, don't build ASI. Two -- you dropped $10 billion on a reasoning model you coulda' whipped up for

18

14

313

Aidan McLau

@aidan_mclau

1 month

on twitter: >cool model, aidan! >well-written docs! >huh i had this error can you help? on reddit: >what a FUCKING IDIOT >shipped WITHOUT A TECHNICAL PAPER??? >he's probably just selling my email to the CHINESE

12

1

303