Elliot Murphy (@ElliotMurphy91) header
Elliot Murphy (@ElliotMurphy91) profile picture

Elliot Murphy

@ElliotMurphy91

• Postdoc, Department of Neurosurgery, UTHealth • Neural mechanisms of compositional syntax

 Houston, TX
elliot-murphy.com
Joined May 2026
  • Tweets246
  • Following96
  • Followers138
48Media
Photo by Elliot Murphy (@ElliotMurphy91)Photo by Elliot Murphy (@ElliotMurphy91)Photo by Elliot Murphy (@ElliotMurphy91)Photo by Elliot Murphy (@ElliotMurphy91)Photo by Elliot Murphy (@ElliotMurphy91)Photo by Elliot Murphy (@ElliotMurphy91)Photo by Elliot Murphy (@ElliotMurphy91)Photo by Elliot Murphy (@ElliotMurphy91)Photo by Elliot Murphy (@ElliotMurphy91)Photo by Elliot Murphy (@ElliotMurphy91)
New work out today! The result of a 15 year project to offer genuine linking hypotheses between language & brain: a revised research program, and its first result - a novel neural binding operation ('Meld') capturing core properties of syntax 🧵 arxiv.org/abs/2609.14384
Image from Elliot Murphy (@ElliotMurphy91)
4
24
117
7,721
Disrupting the semantic network shifts activity to domain-general brain regions nature.com/articles/s41467-0…
Disrupting the semantic network shifts activity to domain-general brain regions

Disrupting the semantic network shifts activity to domain-general brain regions

Nature Communications - Disrupting a key language region prompts increased activity in widespread brain regions. Stronger engagement of frontal control regions helps preserve cognitive performance,...

nature.com
2
6
578
The scramble for your brain data science.org/content/article/…
Wearable device makers are harvesting brain data to train AI—without telling users

Wearable device makers are harvesting brain data to train AI—without telling users

Science investigation reveals firms have amassed tens of millions of hours of brain wave data from neurotech headbands, and face few constraints in using it

science.org
77
Recursion in the wild
36
Elliot Murphy retweeted
Chris Collins (@ChrisTCollins) profile picture
Chris Collins@ChrisTCollinsNov 7
The video for "Merge: Past, Present, and Future" is now out. ordinaryworkinggrammarian.bl…
5
20
549
This is an old hat. That's exactly what the neural fast weight programmers did in the 1990s (including the 1991 unnormalised linear Transformer). Overview with links to the original papers: people.idsia.ch/~juergen/fas…
2021: 30th anniversary of linear Transformer principle

2021: 30th anniversary of linear Transformer principle

1991: Nets learn to program nets with outer product fast weights

people.idsia.ch
Rikiya Takehi (@rikiyatakehi) profile pictureRikiya Takehi@rikiyatakehiNov 2
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Image from Rikiya Takehi (@rikiyatakehi)
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
101
199
99
2,098
229,929
Elliot Murphy retweeted
James Zou (@james_y_zou) profile picture
James Zou@james_y_zouNov 2
How long does a finding about LLMs stay true? We re-tested 50 widely cited findings about LLM behavior (various biases, hallucination, etc). Within 15 months, most no longer held or had substantially reduced effects. LLM findings has short shelf life zou-group.github.io/llm-clai… Great job by @oq_35!
Image from James Zou (@james_y_zou)
Owen Queen (@oq_35) profile pictureOwen Queen@oq_35Nov 2
1/n Within 15 months of publication, most of the LLM claims we tested either no longer held or showed substantially diminished effects. How long does our knowledge of AI hold up? In our new study, we revisit 50 findings to see what fades, what persists, and what gets stronger. 🧵
Image from Owen Queen (@oq_35)
49
239
42
1,139
110,647
A domain-general basal ganglia mechanism for inhibitory control during language comprehension biorxiv.org/content/10.64898…
A domain-general basal ganglia mechanism for inhibitory control during language comprehension

A domain-general basal ganglia mechanism for inhibitory control during language comprehension

Whether controlled language processing involves domain-general or language-specific neural circuits is subject of controversial debate. We tested whether the inhibitory control of linguistic activity...

biorxiv.org
1
175
Interpretable embeddings of speech explain and enhance the brain encoding performance of audio models journals.plos.org/ploscompbi…
Interpretable embeddings of speech explain and enhance the brain encoding performance of audio...

Interpretable embeddings of speech explain and enhance the brain encoding performance of audio...

Author summary Modern AI systems trained on large amounts of speech, known as speech foundation models, are remarkably good at predicting how the human brain responds when people listen to spoken...

journals.plos.org
1
1
106
Elliot Murphy retweeted
André M. Bastos (@BastosLabNeuro) profile picture
André M. Bastos@BastosLabNeuroNov 2
Following up this, a new preprint from my lab finds a human cousin with iEEG recordings: when an identical sound is presented but creates a different percept due to expectation, we see a rhythmic interplay of brain wave activity reflecting the mismatch: biorxiv.org/content/10.64898…
Doris Tsao (@doristsao) profile pictureDoris Tsao
@doristsaoOct 17
1/ We discovered something totally unexpected about the neural structure of consciousness that overturns decades of thinking on its biological basis and brings us significantly closer to the recipe for generating conscious awareness. New preprint: biorxiv.org/content/10.64898… (led by @HesseJanis, with @FrankLanfranchi, @_conor_mcgrory, @Yuelin_Shi, and @EngelTatiana)
Rapid alternation of conscious and suppressed stimulus representations in the primate brain

Rapid alternation of conscious and suppressed stimulus representations in the primate brain

The neural correlates of consciousness remain elusive[1][1]. Previous studies found that high-level visual cortex is weakly modulated by conscious perception[2][2],[3][3], yet a pure signal of...

biorxiv.org
4
16
2,076
Brilliant paper 🧠 💥 Large language models poorly capture how word order shapes representations of sentence meaning in the human brain elifesciences.org/articles/1…
Large language models poorly capture how word order shapes representations of sentence meaning in...

Large language models poorly capture how word order shapes representations of sentence meaning in...

Representations from models that explicitly encode semantic roles (who did what to whom) align better with activity in language regions of the brain than representations from transformers do.

elifesciences.org
3
15
1
87
4,761
“Our results … cast doubt on the claim that transformers represent sentence meaning similarly to the human brain.”
3
315
New paper out today in Cognitive Science assessing some recent biocultural and interactional accounts of human language evolution that we find to be incomplete. onlinelibrary.wiley.com/doi/…
Communication Without Design: Language Is Not Merely Ornamental to Cognition

Communication Without Design: Language Is Not Merely Ornamental to Cognition

Viewing language as a dynamic system shaped by interaction raises questions about how Hockett’s design features should be interpreted.While this shift successfully broadens the functional and...

onlinelibrary.wiley.com
2
270
Independent basal ganglia neural populations encode speech production and ambient-noise levels nature.com/articles/s41467-0…
Independent basal ganglia neural populations encode speech production and ambient-noise levels

Independent basal ganglia neural populations encode speech production and ambient-noise levels

Nature Communications - The role of basal ganglia nuclei in speech production has remained unclear. Here, the authors discover two separate basal ganglia neural populations related to acoustic...

nature.com
4
168
Functional architecture of the human insula nature.com/articles/s41586-0…
Functional architecture of the human insula

Functional architecture of the human insula

Nature - A functional architecture of the human insula links cytoarchitecture to directed and asymmetric electrophysiological interactions, with mechanistic implications for cognitive modulation of...

nature.com
59
I'm giving a talk on the power of language this Thursday (10/22) at 3:30pm (New York time). It's on Zoom, so feel free to join!
Image from Ryan Rhodes ⚙️🧠 (@langofmind)
2
18
3
56
2,147
Elliot Murphy retweeted
Anil Seth (@anilkseth) profile picture
Anil Seth
@anilksethOct 19
Three reasons why AI is extremely unlikely to be conscious - and why this matters. Joint opinion in @nature with Giulio Tononi and Ned Block @De_dicto
nature (@Nature) profile picturenature
@NatureOct 19
LLMs can mimic thought and feeling with uncanny skill, but intelligence alone is not enough for subjective experience go.nature.com/3VSXWQ4
Three reasons why AI is not conscious and perhaps never will be

Three reasons why AI is not conscious and perhaps never will be

Nature - Large language models can mimic thought and feeling with uncanny skill. But evidence from neuroscience suggests that intelligence alone is not enough for subjective experience.

nature.com
83
90
25
476
66,578
Important results! A language model's linguistic performance can deteriorate without a corresponding deterioration in its ability to predict human brain activity. proceedings.mlr.press/v306/o…

Linguistic Properties and Model Scale in Brain Encoding: From Small to Compressed Language Models

Recent work has shown that scaling large language models (LLMs) improves their alignment with human brain activity, yet it remains unclear what drives these ...

proceedings.mlr.press
4
101
Assessing agrammatic speech: Insights from perceptual ratings, statistical metrics, and lesion-symptom mapping sciencedirect.com/science/ar…
1
78
Elliot Murphy retweeted
Elliot Murphy (@ElliotMurphy91) profile picture
Elliot Murphy
@ElliotMurphy91Oct 6
New paper out today! We extend Rational Speech Act models using coupled active inference. A number of clear implications for theories of pragmatics and the language/thought debate follow 🧵 arxiv.org/abs/2610.04347
Evidence and Intervention: A Coupled Active-Inference Extension of...

Evidence and Intervention: A Coupled Active-Inference Extension of...

Identical utterance choices can arise from different communicative causes, and identical interpretations can leave different traces in what a listener learns. Rational Speech Act (RSA) models...

arxiv.org
1
1
3
194
Frequency-specific oscillations track parallel processing of linguistic structural and lexical-semantic information from perception to decision nature.com/articles/s42003-0…

Frequency-specific oscillations track parallel processing of linguistic structural and lexical-se...

Communications Biology - Frequency-specific oscillations track parallel processing of linguistic structural and lexical-semantic information from perception to decision

nature.com
4
131
Load more