Astral Codex Ten

Marcus

10 min read

Marcus

Article

Marcus is a recurring person in the Astral Codex Ten archive, appearing 9 times across 9 issues between June 07, 2022 and April 01, 2026. The archive places it in contexts such as “By “the Lockean, blank-slate view”, Marcus means the idea”; “Marcus says they refuse to let him access them and he has to access it through friends”; “I tend to agree with GPT-3 more than Marcus”. It most often appears alongside Scott, Austin, France.

Metadata

  • Category: People
  • Mention count: 9
  • Issue count: 9
  • First seen: June 07, 2022
  • Last seen: April 01, 2026

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

My Bet: AI Size Solves Flubs
June 07, 2022 · Original source
Somebody else (usually Gary Marcus) demonstrates that the AI also fails terribly at certain trivial tasks. This person argues that this shows that those tasks require true intelligence, whereas the AI is just clever pattern-matching.
Inline links: Gary Marcus
To give an example: in January 2020, Gary Marcus wrote a great post, GPT-2 And The Nature Of Intelligence, demonstrating a bunch of easy problems that GPT-2 failed on:
Inline links: GPT-2 And The Nature Of Intelligence
Possibly Gary Marcus is right that there is some kind of intelligence that humans have and GPTs don’t, and that nothing in GPT’s evolutionary line will ever equal human performance.
Somewhat Contra Marcus On AI Scaling
June 10, 2022 · Original source
Previously: I predicted that DALL-E’s many flaws would be fixed quickly in future updates. As evidence, I cited Gary Marcus’ lists of GPT’s flaws, most of which got fixed quickly in future updates.
Inline links: I predicted
I asked the local five-year-old the same questions that Gary Marcus asked GPT. She did a little better than GPT-3 (73% vs. 63%), but it was a close contest. Five-year-olds are known to be less good at practical reasoning than adults, but it’s not like they’re missing a brain lobe or neurotransmitter or anything. They’re just doing the same processes, only worse. Did this 5 year old have world-modeling ability, or not?
Inline links: are known to be
At some point before 2030, someone will come out with a deep-learning-based language model which is significantly better than the current state of the art, by Gary Marcus’ admission (97%)
22
June 13, 2022 · Original source
$2000 in liquidity and still 14% off from Metaculus, weird. Musk Vs. Marcus Elon Musk recently said he thought we might have AGI before 2029, and Gary Marcus said we wouldn’t and offered to bet on it. It’s an important tradition of AGI discussions that nobody can ever agree on a definition of it and it has to be re-invented every time the topic comes up. Marcus proposed five different things he thought an AI couldn’t do before 2029, such that if it does them, he admits he was wrong and Musk wins the bet (which purely hypothetical at this point; Musk hasn’t responded). The AI would have to do at least three of: Read a novel and answer complicated questions about eg the themes (existing language models can do this with pre-digested novels, eg LAMDA talking about Les Miserables here - I think Marcus means you have to give it a new novel that it has no corpus of humans ever having discussed before, and make it do the work itself).
Inline links: said we wouldn’t, here
Read a novel and answer complicated questions about eg the themes (existing language models can do this with pre-digested novels, eg LAMDA talking about Les Miserables here - I think Marcus means you have to give it a new novel that it has no corpus of humans ever having discussed before, and make it do the work itself).
Inline links: here
So who will win the bet? Metaculus thinks probably Musk - except that he has yet to agree to it. If someone else with a spare $500K wanted to jump in, it looks like in expectation they would make some money.
Inline links: https://www.metaculus.com/questions/11199/gary-marcus-agi-bet-2030/
I Won My Three Year AI Progress Bet In Three Months
September 12, 2022 · Original source
At the time, I wrote: I’m not going to make the mistake of saying these problems are inherent to AI art. My guess is a slightly better language model would solve most of them…for all I know, some of the larger image models have already fixed these issues. These are the sorts of problems I expect to go away with a few months of future research. This proved controversial. Gary Marcus in particular has emphasized how challenging compositionality is for modern language and image models: @sama @gdb @Plinz @ylecun, \n\nEach of you ridiculed my recent title, but this is what the article was actually about: compositionality.\n\nYes, there are many kinds of progress in other directions. \n\nBut compositionality is at the core of intelligence. \n\nNo AGI without it. ","username":"GaryMarcus","name":"Gary Marcus","profile_image_url":"","date":"Sat Apr 09 04:34:37 +0000 2022","photos":[],"quoted_tweet":{"full_text":"Compositionality *is* the wall. \n\nEven “red cube” and “blue cube” on their own are represented unreliably; not one of ten images correctly captures the full phrasal description.\n\nThe images are beautiful, but no match for the precision of language. https://t.co/uvoXUtETwi","username":"GaryMarcus","name":"Gary Marcus"},"reply_count":0,"retweet_count":7,"like_count":54,"impression_count":0,"expanded_url":{},"video_url":null,"belowTheFold":true}" data-component-name="Twitter2ToDOM"> And one of my commenters, Vitor, asked: Why are you so confident in this? The inability of systems like DALL-E to understand semantics in ways requiring an actual internal world model strikes me as the very heart of the issue. We can also see this exact failure mode in the language models themselves. They only produce good results when the human asks for something vague with lots of room for interpretation, like poetry or fanciful stories without much internal logic or continuity. Not to toot my own horn, but two years ago you were naively saying we'd have GPT-like models scaled up several orders of magnitude (100T parameters) right about now (https://slatestarcodex.com/2020/06/10/the-obligatory-gpt-3-post/#comment-912798). I'm registering my prediction that you're being equally naive now. Truly solving this issue seems AI-complete to me. I'm willing to bet on this (ideas on operationalization welcome). I responded to Marcus here, and I responded to Vitor by making a bet on whether AI image models could draw some compositionality-heavy pictures by 2025. The specific terms we agreed on: My proposed operationalization of this is that on June 1, 2025, if either if us can get access to the best image generating model at that time (I get to decide which), or convince someone else who has access to help us, we'll give it the following prompts: 1. A stained glass picture of a woman in a library with a raven on her shoulder with a key in its mouth 2. An oil painting of a man in a factory looking at a cat wearing a top hat 3. A digital art picture of a child riding a llama with a bell on its tail through a desert 4. A 3D render of an astronaut in space holding a fox wearing lipstick 5. Pixel art of a farmer in a cathedral holding a red basketball We generate 10 images for each prompt, just like DALL-E2 does. If at least one of the ten images has the scene correct in every particular on 3/5 prompts, I win, otherwise you do. DALL-E can’t do any of these: If I were being kind, I would give it the farmer in the cathedral. But I am being unkind, so the farmer in front of the cathedral doesn’t count. II. There are now at least four more AI image models available: Google Imagen announced May 2022.
Inline links: https://slatestarcodex.com/2020/06/10/the-obligatory-gpt-3-post/#comment-912798, here, bet, https://substackcdn.com/image/fetch/$s_!_gqe!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2F6138ab0d-3a82-4eb9-a328-bf38ea0f6b10_632x784.png, announced
From Nostradamus To Fukuyama
September 28, 2022 · Original source
Gary Marcus is doomed. I’m sorry. He has generally been very nice to me, and I am not insulting him, or even calling him wrong. I am just saying he’s doomed. For better or worse, people interpret him as saying that AI won’t have lots of crazy huge advances soon. Now, I think AI will have lots of crazy huge advances soon. But suppose he’s right and I’m wrong. Suppose that of 50 possible crazy huge advances that people are predicting in the next ten years, only one materializes. That should be a victory for Marcus. But in fact what will happen is that when that one materializes, people will shake their heads and say “That Gary Marcus guy’s takes really didn’t age well, it seems naive and ostrich-head-in-the-sand-y to keep denying the power of AI when we’re dealing with $THE_ONE_THING_THAT_MATERIALIZED”. When he argues that - come on, 49/50 of my predictions came true! - everyone will call it “cope”.
There’s an even worse problem. He is arguing that the media is hyping AI advances so much that people are getting overly excited about very minor things. Crucially, he argues they’re doing this successfully - this is why he needs to push back against them. If he’s right about everything, then in the future, we can expect the media to continue to successfully hype AI advances. Every time they succeed will be another chance for people to say “That Gary Marcus guy sure looks like a fool now - he said AI was just media hype, but I just heard on the media today that actually it’s turning out to be a really big deal”.
Spring Meetups Everywhere 2023
April 10, 2023 · Original source
KARLSRUHE, GERMANY Contact: Marcus Contact Info: mail[at]marcuswilhelm[dot]de Time: Saturday, April 22nd, 03:00 PM Location: We meet on the KIT campus on the grass in front of Audimax, next to the large sculpture with the intertwined tubes Coordinates: https://plus.codes/8FXC2C68+X5 Notes: An event post will be created on the Karlsruhe Rationality Group on LessWrong; see there for more recent updates (e.g. in case of rain).
Inline links: https://plus.codes/8FXC2C68+X5
Spring Meetups Everywhere 2024
March 30, 2024 · Original source
KARLSRUHE, GERMANY Contact: Marcus Contact Info: acx[at]marcuswilhelm[dot]de Time: Saturday, April 27th, 3:00 PM Location: Leih-Lokal Freiräume, Gerwigstr. 41 76131 Karlsruhe Coordinates: https://plus.codes/8FXC2C5H+CR Group Link: https://www.lesswrong.com/groups/kw7Zb8DLmZtsK8g3R
Inline links: https://plus.codes/8FXC2C5H+CR, https://www.lesswrong.com/groups/kw7Zb8DLmZtsK8g3R
Meetups Everywhere Spring 2025: Times & Places
March 25, 2025 · Original source
Contact: Marcus Contact Info: https://www[period]lesswrong[period]com/users/wilm Time: Saturday, April 26th, 3:00 PM Location: We meet in Otto-Dullenkopf-Park (Süd), around the large trees and wavy concrete walls/benches in the middle. I’ll bring a sign. Coordinates: https://plus.codes/8FXC2C3H+6C Group Link: https://www.lesswrong.com/groups/kw7Zb8DLmZtsK8g3R Notes: In case of bad weather we go somewhere indoors. I’ll update the location in the (comments of) the LessWrong post
Inline links: https://plus.codes/8FXC2C3H+6C, https://www.lesswrong.com/groups/kw7Zb8DLmZtsK8g3R
Meetups Everywhere Spring 2026: Times & Places
April 01, 2026 · Original source
Contact: Marcus Contact Info: wilm on LessWrong Time: Saturday, May 16th, 10:00 AM Location: Cafe intro am Kronenplatz Coordinates: https://plus.codes/8FXC2C55+PJ Group Link: https://chat.whatsapp.com/ChC [remove this bit] zrpd602SEdL1RjlZqs6?mode=gi_t Notes: In case of good weather I propose to walk through the nearby park after some time
Inline links: https://plus.codes/8FXC2C55+PJ