Andrew Lampinen @AndrewLampinen Twitter profile

Pinned Tweet

Andrew Lampinen

2 months

Excited to share one of the main things I’ve been working on: scaling towards grounded language agents that can follow natural language instructions across many video game environments—using the same pixels to keyboard/mouse controls as humans do. 1/8

12

61

325

Last Seen Profiles

@NatsukiSaika

@kalyanipriyan

@ozwalditou

@Kidney_Research

@TheErwinJames

@Wyland

@Vas_526

@togetric

@freehiei

@inu_daifuku_

@CricWick

@DiomarBrasindo

@tennensekiohana

@Topps

@InyangaMEG

@OliverKrimpas

@GrippedMagazine

@AnkaBadowska

@joalejandroag

@MytriMonisha

@alcidesmonsalve

@JspSandeep_

@momokonojo

@HansGeorgHuber

@groenlinks

@CourtAitkenTV

@julia1965

@HardieGrantCP

@Pamelnos

@BFRedrocks

@FB101com

@Pantsushi3D

@araceli03956935

@Garryplummer

@StigOfHK

@MengesJochen

Andrew Lampinen

@AndrewLampinen

1 year

Ted Chiang is a great writer, but this is not a great take and I'm disappointed to see it getting heavily praised. It's not in keeping with our scientific understanding of LMs or deep learning more generally. Thread: 1/n

John Burn-Murdoch

@jburnmurdoch

1 year

Ted Chiang’s piece on ChatGPT and large language models is as good as everyone says. The fact that the outputs are rephrasings rather than direct quotes makes them seem game-changingly smart — even sentient — but they’re just very straightforwardly not.

92

532

3K

27

202

1K

Andrew Lampinen

@AndrewLampinen

4 years

Officially done with my PhD! My dissertation is now online (), if you fancy reading way too many pages about flexibility and transfer in humans and deep learning models.

21

56

567

Andrew Lampinen

@AndrewLampinen

1 year

What is emergence, and why is it of recent interest in AI, and long-standing interest in cognitive science? And why is this an exciting time for considering emergence across these fields? A thread: 1/

14

93

471

Andrew Lampinen

@AndrewLampinen

11 months

Computational physicists doing most their operations (e.g. derivatives) via an FFT because it's more efficient

Rob Miles (✈️ Tokyo)

@robertskmiles

11 months

Their results are bizarre and inhuman. @NeelNanda5 trained a tiny transformer to do addition, then spent weeks figuring out what it was doing - one of the only times in history someone has understood how a transformer works. This is the algorithm it created. To *add two numbers*!

71

262

2K

6

38

464

Andrew Lampinen

@AndrewLampinen

1 year

What can be learned about causality and experimentation from passive data? What could language models learn from simply passively imitating text? We explore these questions in our new paper: “Passive learning of active causal strategies in agents and language models” Thread: 1/

11

75

403

Andrew Lampinen

@AndrewLampinen

2 years

I'm not a scaling maximalist, but it's surprising to me how many people are 1) interested in differences between human and artificial intelligence and 2) think scaling to improve performance means deep learning is doing something fundamentally wrong. 1/n

15

64

394

Andrew Lampinen

@AndrewLampinen

5 months

Research in mechanistic interpretability and neuroscience often relies on interpreting internal representations to understand systems, or manipulating representations to improve models. I gave a talk at @unireps at NeurIPS on a few challenges for this area, summary thread: 1/

3

55

392

Andrew Lampinen

@AndrewLampinen

2 years

Abstract reasoning is ideally independent of content. Language models do not achieve this standard, but neither do humans. In a new paper ( co-led by Ishita Dasgupta) we show that LMs in fact mirror classic human patterns of content effects on reasoning. 1/

Language models show human-like content effects on reasoning tasks

Abstract reasoning is a key ability for an intelligent system. Large language models (LMs) achieve above-chance performance on abstract reasoning tasks, but exhibit many imperfections. However,...

arxiv.org

7

56

381

Andrew Lampinen

@AndrewLampinen

4 years

I'm excited to announce that next month I'll be joining @DeepMind as a research scientist! Looking forward to continuing to work with @FelixHill84 and others!

11

7

356

Andrew Lampinen

@AndrewLampinen

3 years

How can RL agents recall the past in detail, in order to behave appropriately in the present? In our new preprint "Towards mental time travel: A hierarchical memory for RL agents" () we propose a memory architecture that steps in this direction.

Towards mental time travel: a hierarchical memory for...

Reinforcement learning agents often forget details of the past, especially after delays or distractor tasks. Agents with common memory architectures struggle to recall and integrate across...

arxiv.org

3

69

340

Andrew Lampinen

@AndrewLampinen

3 years

What are symbols? Where do symbols come from? What behaviors demonstrate the ability to engage with symbols? How do the answers to these questions impact AI research? We argue for a new perspective on these issues our preprint: Summary in thread: 1/6

Symbolic Behaviour in Artificial Intelligence

The ability to use symbols is the pinnacle of human intelligence, but has yet to be fully replicated in machines. Here we argue that the path towards symbolically fluent artificial intelligence...

arxiv.org

7

70

333

Andrew Lampinen

@AndrewLampinen

4 years

How can deep learning models flexibly reuse their knowledge? How can they adapt to new tasks zero-shot, as humans can? In our new preprint (), we propose a new approach based on learning to transform task representations: meta-mapping. Preview in thread:

3

65

296

Andrew Lampinen

@AndrewLampinen

6 months

Very excited to share a substantial updated version of our preprint “Language models show human-like content effects on reasoning tasks!” TL;DR: LMs and humans show strikingly similar patterns in how the content of a logic problem affects their answers. Thread: 1/

2

42

253

Andrew Lampinen

@AndrewLampinen

6 months

Exactly. Nobody (serious) thinks transformers can magically infer things totally unsupported by training. However, they can generalize, not just memorize, things that are supported. The key question is what, is/isn't supported by training on Internet-scale distributions? 1/

Quanquan Gu

@QuanquanGu

6 months

I believe the message in the paper is straightforward and uncontroversial. However, it seems there might be a misunderstanding in @abacaj 's interpretation. Pre-trained transformers can effectively acquire in-context knowledge for tasks related to their pre-training data and…

6

34

219

7

16

232

Andrew Lampinen

@AndrewLampinen

2 years

How should we compare the capabilities of language models and humans? Is the answer different for LMs than cognitive models? In I offer some thoughts, focusing on a case study of LM processing of recursively nested grammatical structures. Thread: 1/11

Can language models handle recursively nested grammatical...

How should we compare the capabilities of language models (LMs) and humans? I draw inspiration from comparative psychology to highlight some challenges. In particular, I consider a case study:...

arxiv.org

6

56

227

Andrew Lampinen

@AndrewLampinen

2 years

Explanations play a critical role in human learning, particularly in challenging areas—abstractions, relations and causality. We show they can also help RL agents in "Tell me why!—Explanations support learning of relational and causal structure" (). Thread:

Tell me why! Explanations support learning relational and causal structure

Inferring the abstract relational and causal structure of the world is a major challenge for reinforcement-learning (RL) agents. For humans, language--particularly in the form of...

arxiv.org

3

38

214

Andrew Lampinen

@AndrewLampinen

9 months

The debate over AI capabilities often hinges on testing abilities that humans are presumed to have — reasoning, logic, systematic, compositional generalization, grammar, etc. But how reliably good are humans actually at these skills? 1/

11

45

206

Andrew Lampinen

@AndrewLampinen

2 years

Interested in how RL agents could recall the past in detail, in order to overcome the challenges of the present? Come chat with us about "Towards mental time travel: A hierarchical memory for RL agents" at #NeurIPS2021 poster session 1 (4:30 GMT/8:30 PT, spot E1)!

4

31

195

Andrew Lampinen

@AndrewLampinen

2 years

Can language models learn from explanations of answers in a few-shot prompt? In our new preprint (), we explore this question! Thread:

Can language models learn from explanations in context?

Language Models (LMs) can perform new tasks by adapting to a few in-context examples. For humans, explanations that connect examples to task principles can improve learning. We therefore...

arxiv.org

3

45

191

Andrew Lampinen

@AndrewLampinen

6 months

What is representational alignment? How can we use it to study or improve intelligent systems? What challenges might we face? In a new paper, we describe a framework that attempts to unify ideas from cognitive science, neuroscience and AI to address these questions. 1/3

3

28

187

Andrew Lampinen

@AndrewLampinen

11 months

Why does asking language models to respond as an expert or think step-by-step improve answers? What does it have to do with conditional BC and role play? Thread on the power of conditional sequence modeling, and why evaluating LM capabilities or safely deploying them is hard: 1/

9

36

180

Andrew Lampinen

@AndrewLampinen

1 year

The recent discussions of what language models can and can't accomplish highlight some important issues in how cognitive science, linguistics, etc. think about human capabilities or competencies, and thus how to test them in models. Thread: 1/

1

45

178

Andrew Lampinen

@AndrewLampinen

2 years

It's often claimed that learning language alone can't lead to understanding because understanding requires relating language to external meaning. I'm all for grounding and social learning of language, but I think that argument is wrong, or at least uninteresting. 1/n

14

30

174

Andrew Lampinen

@AndrewLampinen

1 year

Chomsky favoring thought experiments based on his mental model of a system over actual empirical results? What a shocker

Jesse Mu

@jayelmnop

1 year

PSA to anyone who wants to write an op-ed criticizing LLMs (yes, including Noam Chomsky): if you're going to come up with hypothetical failure cases for LLMs, at a minimum, please actually check that your case fails with a modern LLM

31

87

870

6

20

173

Andrew Lampinen

@AndrewLampinen

1 year

When we and others study the performance of LM-like models under controlled situations, where we know exactly what is trained, we find that these models can learn generalizable strategies that perform well on truly novel test examples rather than just memorizing + rephrasing. 3/

8

15

150

Andrew Lampinen

@AndrewLampinen

1 year

I think this paper does a good job of highlighting the point that apparently discontinuous transitions in performance may correspond to more continuous change in a smoother metric (also made in ). However, I have a terminological quibble. 1/5

Beyond the Imitation Game: Quantifying and extrapolating the...

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet...

arxiv.org

Aran Komatsuzaki

@arankomatsuzaki

1 year

Are Emergent Abilities of Large Language Models a Mirage? Presents an alternative explanation for emergent abilities: one can choose a metric which leads to the inference of an emergent ability or another metric which does not.

24

187

978

3

26

148

Andrew Lampinen

@AndrewLampinen

1 year

For example, see the grokking paper by @exteriorpower (), works led by @scychan on few-shot learning and generalization ( and ), @oswaldjoh work on in-context gradient descent () etc. 4/

Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets

In this paper we propose to study generalization of neural networks on small algorithmically generated datasets. In this setting, questions about data efficiency, memorization, generalization, and...

arxiv.org

3

7

148

Andrew Lampinen

@AndrewLampinen

7 months

Nice, though not too shocking, to see that human inductive biases on these sequential tasks can be meta-learned. I particularly appreciate the emphasis on the ability of the transformers to learn subtler biases in the human behavior. 1/4

Brenden Lake

@LakeBrenden

7 months

Today in Nature, we show how a standard neural net, optimized for compositional skills, can mimic human systematic generalization (SG) in a head-to-head comparison. This is the capstone of a 5 year effort with Marco Baroni to make progress on SG. (1/8)

24

395

2K

3

13

146

Andrew Lampinen

@AndrewLampinen

3 years

Excited that our work on "Transforming task representations to perform novel tasks" is now published online @PNASNews ! We provide a new perspective on flexible adaptation, which we hope will be inspiring for both AI and cognitive science. Summary thread:

Transforming task representations to perform novel tasks | Proceedings of the National Academy of...

An important aspect of intelligence is the ability to adapt to a novel task without any direct experience (zero shot), based on its relationship to...

www.pnas.org

2

24

128

Andrew Lampinen

@AndrewLampinen

3 years

Honored to share that I have been awarded one of the CogSci 2021 Glushko Prizes for my dissertation on learning to transform task representations to adapt to novel tasks! And I'm especially honored to share it with these incredible other scholars.

CogSci Society

@cogsci_soc

3 years

The #CogSci2021 Glushko Dissertation Prize Winners have been announced! Congratulations to all - very well deserved. Visit the website for further info on our winners:

2

8

76

16

3

117

Andrew Lampinen

@AndrewLampinen

2 years

I admire the authors, and I find their work interesting. However, I think this piece makes subtle equivocations between strong and weak notions of compositionality that may be problematic (and are common in this space). Thread:

Tom McCoy

@RTomMcCoy

2 years

🤖🧠NEW PAPER🧠🤖 What explains the dramatic recent progress in AI? The standard answer is scale (more data & compute). But this misses a crucial factor: a new type of computation. Shorter opinion piece: Longer tutorial: 1/5

8

112

601

5

15

113

Andrew Lampinen

@AndrewLampinen

1 year

This is not to imply that LMs never do (lossy) memorization, of course; that definitely happens in many instances, is important to study, and may underlie their success in some cases. But saying that is all they do seems out of keeping with the state of the science. 7/7

4

2

112

Andrew Lampinen

@AndrewLampinen

1 year

In case you missed our paper "Can language models learn from explanations in context?" and want a 7.5 minute video overview of the experiments and main findings, I've now put one on YouTube!

Can language models learn from explanations in context?

This is a short talk on our paper "Can language models learn from explanations in context?" (https://arxiv.org/abs/2204.02329), recorded for Findings of EMNL...

www.youtube.com

4

19

110

Andrew Lampinen

@AndrewLampinen

1 year

But it's also a general finding that neural networks generalize better from more diverse learning experiences. E.g. the open-ended learning work () or our work on how richer environments improve generalization (). 5/

Open-Ended Learning Leads to Generally Capable Agents

In this work we create agents that can perform well beyond a single, individual task, that exhibit much wider generalisation of behaviour to a massive, rich space of challenges. We define a...

arxiv.org

2

5

93

Andrew Lampinen

@AndrewLampinen

2 years

Why? Because evolution made some pretty hefty tradeoffs in energy cost, soft/fractured skulls at birth, etc. in order to scale the human brain. As a consequence, we have perhaps ~1 quadrillion synapses (). 2/

The human brain in numbers: a linearly scaled-up primate brain

The human brain has often been viewed as outstanding among mammalian brains: the most cognitively able, the largest-than-expected from body size, endowed with an overdeveloped cerebral cortex that...

www.frontiersin.org

1

2

90

Andrew Lampinen

@AndrewLampinen

7 months

Definitely out here attacking anyone who believes in probabilities less than zero, ngl

Gary Marcus

@GaryMarcus

7 months

It’s almost impossible to be in the middle on p(doom). If you think it is greater than zero, you get attacked, if you think it is less than zero, you get it attacked. If you think we don’t know, you get attacked. For the record, I think we don’t know.

50

6

140

2

1

93

Andrew Lampinen

@AndrewLampinen

1 year

So if anything, we should expect LMs with internet training to generalize better than models trained on toy tasks. And indeed, LMs perform well at BIG-Bench () — tasks researchers chose specifically to be hard for a model that memorized the internet. 6/

Beyond the Imitation Game: Quantifying and extrapolating the...

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet...

arxiv.org

2

4

93

Andrew Lampinen

@AndrewLampinen

3 months

This is a valuable paper for helping to understand how general the inductive biases that give rise to common language structures may be. But furthermore, I think the prior claim that these languages are "impossible" for humans is weak at best. Thread:

Julie Kallini ✨

@JulieKallini

4 months

Do LLMs learn impossible languages (that humans wouldn’t be able to acquire) just as well as they learn possible human languages? We find evidence that they don’t! Check out our new paper… 💥 Mission: Impossible Language Models 💥 ArXiv: 🧵

12

114

478

2

9

90

Andrew Lampinen

@AndrewLampinen

2 years

If you're curious about my thoughts on language models and reasoning (and more), I'll be speaking at the London Machine Learning Meetup on November 16th! (It's virtual, so you don't need to be in London to attend.)

3

12

87

Andrew Lampinen

@AndrewLampinen

8 months

Disappointed, but not all that surprised to see these kinds of opinions going around. A few years ago, we wrote a paper called "Publishing fast and slow: a path towards generalizability in psychology and AI" that seems relevant again. Short thread:

Thomas Steinke

@shortstein

8 months

I'm shocked that some people hold such backward views and saddened that this is coming from a person with significant power and influence in the research community. I ❤️ @arXiv

64

71

980

4

5

90

Andrew Lampinen

@AndrewLampinen

1 year

One important approach to the scientific study of complex phenomena like human intelligence or the behavior of language models is to create a simplified model which captures the key elements, while maintaining full control over the system, and study its behavior. 2/

3

88

Andrew Lampinen

@AndrewLampinen

2 years

We're not going to be able to run brain scale models anytime soon, and I think innovations in architecture, algorithms and training will contribute to advances in AI. But I think dismissing scale may lead to fundamentally misunderstanding the basis of intelligence. 9/9

5

85

Andrew Lampinen

@AndrewLampinen

3 years

I'm pleased to share that we've open-sourced two environments and the hierarchical attention mechanism for our "Towards mental time travel: A hierarchical memory for RL" paper: Paper summary thread/links:

Andrew Lampinen

@AndrewLampinen

3 years

How can RL agents recall the past in detail, in order to behave appropriately in the present? In our new preprint "Towards mental time travel: A hierarchical memory for RL agents" () we propose a memory architecture that steps in this direction.

3

69

340

1

9

86

Andrew Lampinen

@AndrewLampinen

5 months

Really cool results! It makes sense in light of the fact that LMs have to model lots of bad reasoning and inaccurate info that stripping out some of what they've learned can help. Pretty surprising that just targeted rank reduction works though! I'm curious though, 1/2

Pratyusha Sharma @ ICLR

@pratyusha_PS

5 months

What if I told you that you can simultaneously enhance an LLM's task performance and reduce its size with no additional training? We find selective low-rank reduction of matrices in a transformer can improve its performance on language understanding tasks, at times by 30% pts!🧵

37

281

2K

1

13

84

Andrew Lampinen

@AndrewLampinen

3 years

Super excited to announce the lineup of panelists and discussion moderators for our Neural Network Models of Cognition Affinity Group at CogSci this year! An awesome collection of legends in the field and rising stars.

1

13

85

Andrew Lampinen

@AndrewLampinen

2 years

So our largest models are *nowhere near* the scale/complexity of the human brain, which the evolutionary evidence suggests may be important to our intelligence. Sometimes more is different (), and you get emergent phenomena (cf. ) 4/

More Is Different

www.science.org

2

82

Andrew Lampinen

@AndrewLampinen

4 years

Interested in the relationship between cognition and AI or neural networks? I have a new @cogsci_soc blog post describing some factors that allow neural networks to be more flexible, and how those factors relate to human cognition and development:

Flexibility in Neural Network Models - Cognitive Science Society

By Andrew Lampinen Banner image sketch by Natalia Vélez. The flexibility of human cognition is remarkable. We can perform a new task (for example, playing a new videogame) decently after a few...

cognitivesciencesociety.org

1

20

83

Andrew Lampinen

@AndrewLampinen

1 year

If you're interested in recent discussions about what language models can and can't do, you may be interested in this paper on the challenges of comparing LM and human capabilities! I've recently updated it with some minor edits and some new experiments.

Andrew Lampinen

@AndrewLampinen

2 years

How should we compare the capabilities of language models and humans? Is the answer different for LMs than cognitive models? In I offer some thoughts, focusing on a case study of LM processing of recursively nested grammatical structures. Thread: 1/11

6

56

227

1

15

81

Andrew Lampinen

@AndrewLampinen

1 year

Will be at NeurIPS! Let me know if you want to chat about explanations, reasoning, language models, or language-augmented RL!

8

3

76

Andrew Lampinen

@AndrewLampinen

2 months

I like the empirical results, but have some questions about the way they're being discussed. In particular, what is reasoning? By this bar, humans also don't learn to reason, because they tend to be less accurate in out of distribution situations as well. 1/7

Kevin Patrick Murphy

@sirbayes

2 months

@fchollet claimed that "learning to reason" (robustly across problem instances) is hard because of the nature of the MLE objective. This paper proves that claim. The reason is "statistical features inherently exist in data distributions, but can hinder model generalization…

9

72

391

1

10

75

Andrew Lampinen

@AndrewLampinen

5 months

Very excited to head to NeurIPS! Feel free to reach out if you want to chat about any of our recent work on LMs, agents, interpretability, representational alignment, etc. You can find me:

2

0

72

Andrew Lampinen

@AndrewLampinen

2 years

That's many orders of magnitude more than the largest models I'm aware of. And synaptic interactions are likely far more complex than a single parameter can capture (e.g. ). And this all neglects the increasing evidence for glial contributions etc. 3/

2

1

69

Andrew Lampinen

@AndrewLampinen

1 year

See also this thoughtful thread from @raphaelmilliere !

Raphaël Millière

@raphaelmilliere

1 year

I don't think lossy compression is a very helpful analogy to convey what (linguistic or multimodal) generative models do – at least if "blurry JPEGs" is the leading metaphor. It might work in a loose sense, but it doesn't tell the whole story. 1/

22

71

410

3

1

68

Andrew Lampinen

@AndrewLampinen

7 months

Some great analysis and thoughtful comments here! However, are these really embers of *autoregression*? It seems to me that they are quite general phenomena, that are also observed in humans (as the paper notes), CNNs, RL agents, ... any system that learns a prior from data 1/

Tom McCoy

@RTomMcCoy

8 months

🤖🧠NEW PAPER🧠🤖 Language models are so broadly useful that it's easy to forget what they are: next-word prediction systems Remembering this fact reveals surprising behavioral patterns: 🔥Embers of Autoregression🔥 (counterpart to "Sparks of AGI") 1/8

36

299

1K

6

1

69

Andrew Lampinen

@AndrewLampinen

2 years

And as the scaling laws paper showed, larger models generalize *better* from the same amount of data (). Larger models *can* use more data, but they can also make better use of less! 5/9

Scaling Laws for Neural Language Models

We study empirical scaling laws for language model performance on the cross-entropy loss. The loss scales as a power-law with model size, dataset size, and the amount of compute used for training,...

arxiv.org

1

3

67

Andrew Lampinen

@AndrewLampinen

1 year

And though some particularly challenging BIG-Bench tasks are harder, models can perform better on many of them if given time to produce some reasoning steps before their answer (). Memorization + rephrasing is not a good explanation of this. 7/

Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them

BIG-Bench (Srivastava et al., 2022) is a diverse evaluation suite that focuses on tasks believed to be beyond the capabilities of current language models. Language models have already made good...

arxiv.org

2

67

Andrew Lampinen

@AndrewLampinen

1 year

Always fun chatting with @MLStreetTalk ! Check this out for my current thoughts on language, grounding, NeurIPS, and more!

Machine Learning Street Talk

@MLStreetTalk

1 year

First interview from #NeurIPS2022 uploaded! This is a conversation with @AndrewLampinen about NLP, symbols, grounding and even Chomsky! We recorded a LOT of footage this week, keep an eye out!

4

21

108

1

5

67

Andrew Lampinen

@AndrewLampinen

3 months

Since there continues to be some confusion about the term "passive," let me unpack this in more detail. Part of the motivation for the use of the term here was to draw attention to two dimensions that get conflated in discussions of langauge models, but that should not be. 1/

Andrew Lampinen

@AndrewLampinen

1 year

What can be learned about causality and experimentation from passive data? What could language models learn from simply passively imitating text? We explore these questions in our new paper: “Passive learning of active causal strategies in agents and language models” Thread: 1/

11

75

403

1

7

67

Andrew Lampinen

@AndrewLampinen

1 year

As someone who is both interested in the successes of language models, and also in the contributions of rich learning environments to generalization, I've been hoping to see some results like this for a while! The evaluations are relatively simple, but it's very suggestive.

Jiannan Xiang

@szxiangjn

1 year

🤔Humans can learn from embodied experiences in the physical world. Can Language Models also do that? 🔥Check out our new paper about enhancing Language Models with World Models! 👇 1/n

5

67

356

1

7

66

Andrew Lampinen

@AndrewLampinen

2 years

Pleased to share that our paper "Can language models learn from explanations in context?" () was accepted to Findings of EMNLP 2022! Camera ready version coming soon, with some better control experiments for the tuning conditions.

Can language models learn from explanations in context?

Language Models (LMs) can perform new tasks by adapting to a few in-context examples. For humans, explanations that connect examples to task principles can improve learning. We therefore...

arxiv.org

Andrew Lampinen

@AndrewLampinen

2 years

Can language models learn from explanations of answers in a few-shot prompt? In our new preprint (), we explore this question! Thread:

3

45

191

2

7

65

Andrew Lampinen

@AndrewLampinen

2 years

@NicoleCRust Something about NNs brings nasty responses out! A few excerpts from an email I got from a full professor (at a reasonably prominent institution) who didn't like my answer to their question in a talk I gave about my dissertation:

19

5

65

Andrew Lampinen

@AndrewLampinen

21 days

Our SIMA tech report is now on arXiv!

Andrew Lampinen

@AndrewLampinen

2 months

Excited to share one of the main things I’ve been working on: scaling towards grounded language agents that can follow natural language instructions across many video game environments—using the same pixels to keyboard/mouse controls as humans do. 1/8

12

61

325

1

9

64

Andrew Lampinen

@AndrewLampinen

3 years

I've often thought that some of what people consider a separate "system 1 vs system 2" might be well captured with a single architecture that just decides how long to think. Glad to see some progress in that direction, and excited for further work in this area!

Google DeepMind

@GoogleDeepMind

3 years

Introducing PonderNet, a new algorithm that allows artificial neural networks to learn to “think for a while” before answering, like humans do. This improves the ability of neural networks to generalize outside of the training distribution:

24

297

1K

2

4

62

Andrew Lampinen

@AndrewLampinen

2 years

In our paper "Symbolic Behaviour in AI" () we discuss a perspective on symbols, that we suggest 1) better describes human capacities 2) provides a more promising direction for AI. Briefly, we focus on symbols as subjective, following C.S. Peirce.

Symbolic Behaviour in Artificial Intelligence

The ability to use symbols is the pinnacle of human intelligence, but has yet to be fully replicated in machines. Here we argue that the path towards symbolically fluent artificial intelligence...

arxiv.org

Gary Marcus

@GaryMarcus

2 years

Deep Learning Is Hitting a Wall. What would it take for artificial intelligence to make real progress? #longread in ⁦ @NautilusMag ⁩ on one of the key technical questions in AI.

62

248

843

2

10

62

Andrew Lampinen

@AndrewLampinen

9 months

The most thorough and compelling paper yet in the line of work suggesting that there is signal in even minor adversarial perturbations, that humans are somewhat sensitive to; but with time they can override that signal (but still detect it if asked).

Gamaleldin Elsayed

@gamaleldinfe

9 months

Nature Comms paper: Subtle adversarial image manipulations influence both human and machine perception! We show that adversarial attacks against computer vision models also transfer (weakly) to humans, even when the attack magnitude is small.

12

87

389

2

7

62

Andrew Lampinen

@AndrewLampinen

3 years

Our RL agent memory was accepted at #NeurIPS2021 ! Thanks to the reviewers and AC for thoughtful reviews. We'll have a revised version out soon!

Andrew Lampinen

@AndrewLampinen

3 years

How can RL agents recall the past in detail, in order to behave appropriately in the present? In our new preprint "Towards mental time travel: A hierarchical memory for RL agents" () we propose a memory architecture that steps in this direction.

3

69

340

3

1

61

Andrew Lampinen

@AndrewLampinen

1 year

Emergence is the idea that a large system composed of many small parts can have fundamentally different properties than those parts do — or that“more is different” as Anderson described it (). 2/

More Is Different

www.science.org

2

4

60

Andrew Lampinen

@AndrewLampinen

8 months

Got pretty lucky this year with 3/3 papers I was involved in accepted to NeurIPS! Our work on passive learning of causal strategies:

Andrew Lampinen

@AndrewLampinen

1 year

What can be learned about causality and experimentation from passive data? What could language models learn from simply passively imitating text? We explore these questions in our new paper: “Passive learning of active causal strategies in agents and language models” Thread: 1/

11

75

403

2

0

59

Andrew Lampinen

@AndrewLampinen

2 years

But again, I don't think scale is everything; architecture, algorithms and learning experiences obviously matter. E.g. transformers scale much better than LSTMs and may also have better inductive biases for certain interesting emergent behaviors () 6/9

Data Distributional Properties Drive Emergent In-Context Learning...

Large transformer-based models are able to perform in-context few-shot learning, without being explicitly trained for it. This observation raises the question: what aspects of the training regime...

arxiv.org

2

1

57

Andrew Lampinen

@AndrewLampinen

2 years

I think this is a really exciting time for cognitive science, as @davisblalock highlights from our paper—for the first time we have various models that can exhibit many of the complex behaviors humans do, on naturalistic stimuli like those that are used to test humans. 1/3

Davis Blalock

@davisblalock

2 years

"Language models show human-like content effects on reasoning" So the specific results are interesting and actionable, but my main takeaway from this paper is that cognitive science is becoming adjacent to AI in a way it hasn't been since the 50s or 60s. [1/6]

6

65

430

1

5

57

Andrew Lampinen

@AndrewLampinen

8 months

This is an interesting case of generalization from LMs, where they clearly cannot just "parrot" training data, but rather use explicit statements in tuning to non-trivially alter their behavior in a related test context. However, I find the overall framing misleading. 1/

Owain Evans

@OwainEvans_UK

8 months

Our experiment: 1. Finetune an LLM on descriptions of fictional chatbots but with no example transcripts (i.e. only declarative facts). 2. At test time, see if the LLM can behave like the chatbots zero-shot. Can the LLM go from declarative → procedural info?

1

3

51

3

10

56

Andrew Lampinen

@AndrewLampinen

1 year

Cool work! Some more nice evidence that you shouldn't just *assume* what humans would do when evaluating models, you should actually compare. Really like the conclusion "future research should not idealize human behaviors as a monolith"

Brown NLP

@Brown_NLP

1 year

Last year, we criticized LMs for performing “too well” with pathological prompts, and many papers have now shown similar results with corrupted ICL or CoT. In our new work, we find that *humans* also perform surprisingly well with irrelevant prompts! (But not misleading ones.) ⅕

2

24

137

3

7

56

Andrew Lampinen

@AndrewLampinen

2 years

If you're interested in my thoughts on cognition and AI, or how my research interests have evolved (or if you just need something to listen to while hiding from omicron), this may be the podcast for you!

The Thesis Review Podcast

@thesisreview

2 years

Episode 38 of The Thesis Review: Andrew Lampinen ( @AndrewLampinen ), "A Computational Framework for Learning and Transforming Task Representations" We discuss learning to rapidly adapt to new tasks, cognitive flexibility in minds & machines, and more

2

5

35

1

4

55

Andrew Lampinen

@AndrewLampinen

1 year

There's a trend recently of arguing that we shouldn't describe language models (and other AI systems) with words like "understand" or "see" that anthropomorphize them. A thread on benefits + costs of metaphor in language and thought, and why I'm ambivalent about these arguments:

6

11

55

Andrew Lampinen

@AndrewLampinen

2 years

DeepMind internship applications open tomorrow!

Google DeepMind

Artificial intelligence could be one of humanity’s most useful inventions. We research and build safe artificial intelligence systems. We're committed to solving intelligence, to advance science...

deepmind.google

2

18

53

Andrew Lampinen

@AndrewLampinen

2 years

What does it take to make symbols useful? We think it's embracing their status as subjective, constructed tools. If you want to find out more you can check out this fun conversation with @MLStreetTalk about our paper on Symbolic Behaviour! Thanks for having me on!

Machine Learning Street Talk

@MLStreetTalk

2 years

We discuss @AndrewLampinen from @DeepMind paper on symbolic behaviour in artificial intelligence with @ecsquendor @DoctorDuggar

3

5

23

2

4

52

Andrew Lampinen

@AndrewLampinen

9 months

This claim seems, ironically, to be over-hyped. At a glance, it appears to rely on some very strong assumptions, including a crucial one that is unstated as far as I can see. 1/

Olivia Guest · Ολίβια Γκεστ

@o_guest

9 months

My fav part of this paper — we're not able to stop climate change directly, but we can rain a bit on this silly AI summer: "as we formally prove herein, creating systems with human(-like or -level) cognition is intrinsically computationally intractable."

1

23

61

4

5

53

Andrew Lampinen

@AndrewLampinen

4 years

Interested in integration of information in memory? We have a new paper from a complementary learning systems perspective out in @RSocPublishing : (Open preprint: ). #neuroscience #memory

Integration of new information in memory: new insights from a complementary learning systems...

According to complementary learning systems theory, integrating new memories into the neocortex of the brain without interfering with what is already known depends on a gradual learning process,...

royalsocietypublishing.org

3

14

53

Andrew Lampinen

@AndrewLampinen

3 years

New commentary on "The Generalizability Crisis" by @talyarkoni : "Publishing fast and slow: A path toward generalizability in psychology and AI." We argue that these fields share similar generalizability challenges, and could learn from each other.

Publishing fast and slow: A path toward generalizability in psychology and AI

Commentary on "The Generalizability Crisis" (Yarkoni, 2020). Artificial intelligence (AI) shares many generalizability challenges with psychology. But the fields publish differently. AI publishes...

osf.io

2

11

51

Andrew Lampinen

@AndrewLampinen

1 year

This will indeed probably be a point of confusion. But also the Chinchilla paper only considers training compute in their optimality calculations afaiu; considering inference cost/speed would lead to different trade-offs, and presumably more "overtrained" models.

Stella Biderman

@BlancheMinerva

1 year

@MetaAI @GuillaumeLample Another thing I anticipate being a massive source of confusion: Their smaller models are massively overtrained. The fact that their 13B model meets or exceeds GPT-3 in performance is NOT contradictory to anything in the Chinchilla paper, because it's not compute-optimally trained

5

9

94

1

2

49

Andrew Lampinen

@AndrewLampinen

4 years

We've updated our preprint on zero-shot adaptation by transforming task representations! Now including more carefully controlled experiments, and analyses of how task space transforms under various meta-mappings. (See attached pic!)

1

5

52

Andrew Lampinen

@AndrewLampinen

1 year

Some reflections on our Symbolic Behaviour in AI paper, after two more years of rapid progress in the field:

2

7

52

Andrew Lampinen

@AndrewLampinen

7 months

@yudapearl @EliSennesh @ylecun @tdietterich @GaryMarcus @geoffreyhinton Addressed in our recent paper, now accepted at NeurIPS (updated version coming soon):

Andrew Lampinen

@AndrewLampinen

1 year

What can be learned about causality and experimentation from passive data? What could language models learn from simply passively imitating text? We explore these questions in our new paper: “Passive learning of active causal strategies in agents and language models” Thread: 1/

11

75

403

2

4

47

Andrew Lampinen

@AndrewLampinen

1 year

Dataset contamination is an increasing issue. Was also really disappointed to see BIG-Bench eval omitted due to contamination — there are lots of interesting tasks there.

Horace He

@cHHillee

1 year

I suspect GPT-4's performance is influenced by data contamination, at least on Codeforces. Of the easiest problems on Codeforces, it solved 10/10 pre-2021 problems and 0/10 recent problems. This strongly points to contamination. 1/4

81

696

4K

2

1

48

Andrew Lampinen

@AndrewLampinen

4 years

Interested in automated curricula for RL? Check out our ICLR poster in sessions 3 & 4 tomorrow. Our method works for goal-conditioned agents in complex environments. Paper (open access): Poster: #ICLR2020 #ReinforcementLearning

1

10

45

Andrew Lampinen

@AndrewLampinen

11 months

Interesting to contrast with our recent work studying causal strategies () — a key difference is that rather than inference from correlations, we focus on the fact that LMs can *intervene* at test time, e.g. via a tool/API, which can unlock generalization

Sepp Hochreiter

@HochreiterSepp

11 months

ArXiv : Benchmark dataset to test causal inference of large language models (LLMs). A key shortcoming of LLMs is causal inference. LLMs achieve close to random performance on the causal tasks. Causal inference works only for already known tasks.

2

59

260

2

5

44

Andrew Lampinen

@AndrewLampinen

1 month

Really excited for this interview to come out! Check it out if you're interested in LMs, causality, and agency! Also recorded just before I left London, so a bit of a throwback for me to see :)

Aleksander Molak (CausalPython.io)

@AleksanderMolak

1 month

Causal Inference, Agents and LLMs Join Andrew Lampinen and me for a conversation on... 1/n #CausalBanditsPodcast #causalinference #causaltwitter @machinelearning

1

3

12

1

2

44

Andrew Lampinen

@AndrewLampinen

1 year

Great demonstration of an important point that can't be repeated enough — it's hard to interpret LM evaluations if you don't know whether examples appeared in the training data.

Oscar Sainz @ICLR 2024

@osainz59

1 year

As an example, for popular datasets like CoNLL03, ChatGPT is capable of generating the training, validation, and even test splits. It turns out that ChatGPT has been evaluated as a zero-shot or few-shot NER system on this dataset by multiple papers. 🧵2/5

3

14

56

3

44

Andrew Lampinen

@AndrewLampinen

2 years

And as my colleagues and I have explored in lots of prior work, experiences like explanations () or richer, more embodied environments () can fundamentally change learning & generalization. 7/9

Tell me why! Explanations support learning relational and causal structure

Inferring the abstract relational and causal structure of the world is a major challenge for reinforcement-learning (RL) agents. For humans, language{—}parti...

proceedings.mlr.press

1

43

Andrew Lampinen

@AndrewLampinen

11 months

If you're interested in what LMs or agents can learn about causal strategies from passive data, a recording of a talk I gave at @ic_arl a few weeks ago is out now!

Passive Learning of Active Causal Strategies in Agents and Language...

ICARL Seminar Series - 2023 SpringPassive Learning of Active Causal Strategies in Agents and Language ModelsAbstract:What can be learned about causality and ...

www.youtube.com

1

7

43

Andrew Lampinen

@AndrewLampinen

1 year

It’s not surprising that recent developments in AI are eliciting a range of skeptical and fearful reactions. This almost always happens when technology develops. Thread: 1/

4

3

42

Andrew Lampinen

@AndrewLampinen

11 months

Very cool paper! A bit reminiscient of our work () where we likewise showed that learning to output explanations (~=thoughts) could improve learning and OOD generalization; but letting the agent observe its thoughts is a key idea we didn't get to!

Jeff Clune

@jeffclune

11 months

Introducing Thought Cloning: AI agents learn to *think* & act like humans by imitating the thoughts & actions of humans thinking out loud while acting, enhancing performance, efficiency, generalization, AI Safety & Interpretability. Led by @shengranhu 1/5

14

145

604

1

6

42

Andrew Lampinen

@AndrewLampinen

1 year

Form-meaning arguments are trending again, guess it's time to reboost this (), and also remind everyone that distributional semantics existed long before current LLMs (e.g. "Distributional semantics in linguistic and cognitive research", Lenci, 2008).

Andrew Lampinen

@AndrewLampinen

2 years

It's often claimed that learning language alone can't lead to understanding because understanding requires relating language to external meaning. I'm all for grounding and social learning of language, but I think that argument is wrong, or at least uninteresting. 1/n

14

30

174

4

8

42

Andrew Lampinen

@AndrewLampinen

1 year

Will shortly be presenting our paper on how data distributions give rise to few-shot learning in transformers! Come by poster 524 this afternoon, and/or check out @scychan_brains oral presentation next week!

Stephanie Chan

@scychan_brains

2 years

Intriguingly, transformers can achieve few-shot learning (FSL) without being explicitly trained for it. Very excited to share our new work, showing that FSL emerges in transformers only when the training data is distributed in particular ways! 🧵👇

14

190

1K

0

6

42

Andrew Lampinen

@AndrewLampinen

4 years

A paper supporting our observation () that egocentric perspective improves generalization in RL!

Environmental drivers of systematicity and generalization in a...

The question of whether deep neural networks are good at generalising beyond their immediate training experience is of critical importance for learning-based approaches to AI. Here, we consider...

arxiv.org

hardmaru

@hardmaru

4 years

Rotation, Translation, and Cropping for Zero-Shot Generalization Makes a lot of sense. Try playing Doom not from an agent-centric perspective! I think agent-centric view is a better prior for encoding useful information using fewer bits for the policy.

2

20

73

0

6

41

Andrew Lampinen

@AndrewLampinen

10 months

Lots of interesting results! Most importantly, data dominates the match to neural data much more than architecture (*as long as you use a decent architecture)!

talia konkle

@talia_konkle

10 months

5/ Surprisingly, across both measures, we found that even major differences in model architecture (e.g. CNNs vs. Transformers) did not significantly lead to better or worse brain alignment…

2

10

59

4

6

41

Andrew Lampinen

@AndrewLampinen

5 months

How far can we trust representation analyses or mechanistic interpretations? @danfriedman0 's new work shows that analyses based on simplifying the model or its representations can be misleading about how the model will behave out of distribution! Check out his thread:

Dan Friedman

@danfriedman0

5 months

We often interpret neural nets by studying simplified representations (e.g. low-dim visualization). But how faithful are these simplifications to the original model? In our new preprint, we found some surprising "interpretability illusions"... 1/6

3

49

298

2

4

39

Andrew Lampinen

@AndrewLampinen

1 month

Very cool! Reminds me a bit of the hypernetwork meta-learning architecture we found was beneficial in — awesome to see some theoretical justification for why it might be useful!

Transforming task representations to perform novel tasks | Proceedings of the National Academy of...

An important aspect of intelligence is the ability to adapt to a novel task without any direct experience (zero shot), based on its relationship to...

www.pnas.org

Simon Schug

@ssmonsays

1 month

Many tasks are compositional. Their constituent parts can be recombined in exponentially many ways. How could neural networks possibly learn all those tasks? That is what our ICLR 2024 paper shows: How to fight exponentials with exponentials 🧵️

3

22

118

0

4

41

Andrew Lampinen

@AndrewLampinen

1 year

@kareem_carr Are you familiar with the PDP books ()? They focused on NNs as cognitive models more constrained by some aspects of neural processing, and also introduce concepts like representation learning by backpropagation which are rather fundamental to modern DL.

Parallel Distributed Processing

What makes people smarter than computers? These volumes by a pioneering neurocomputing group suggest that the answer lies in the massively parallel architect...

mitpress.mit.edu

2

0

39