Sign in

Bradley Love

@profdata.bsky.social
4.3K followers 804 following 62 posts

Senior research scientist at Los Alamos National Laboratory. Former UCL, UTexas, Alan Turing Institute, Ellis EU. CogSci, AI, Comp Neuro, AI for scientific discovery bradlove.org

PostsRepliesMedia
Bradley Love @profdata.bsky.social · 02/06/2026
Last month Dawkins declared Claude conscious and was mocked. @garymarcus.bsky.social cleverly called it The Claude Delusion. My (and Nagel's?) take: Both are wrong for the same reason. Here's why the question of machine consciousness will never be settled scientifically. arxiv.org/abs/2606.00226
arxiv.org
Consciousness, AI, and the Limits of Scientific Explanation
Science is constitutively third-personal: its findings are in principle reproducible by any observer, independent of perspective, and answerable to measurement. This is the source of its power and als...
1164
Bradley Love @profdata.bsky.social · 23/03/2026
Thanks @toddgureckis.bsky.social for the good memories; certainly foundational times for me too learning tons. Good points on writing helping to think through stuff. thanks @robmok.bsky.social for the kind words and apologies for the times the emails were annoying :-)
020
Reposted by Bradley Love
Alison Preston @aliprestonphd.bsky.social · 06/03/2026
Every time you experience something new, your brain faces a decision: Should it update an existing memory or create a new one? In our new paper in @sfnjournals.bsky.social #JNeurosci, we isolate that exact decision, moment-by-moment during learning 🧵
313847
Bradley Love @profdata.bsky.social · 28/01/2026
Personally, I will be looking to mentor projects with Mahindra Rautela on (1) Search and Evaluation for test-time AI Reasoning, and (2) model distillation to compress large physics foundation models. Please feel free to get in touch with questions or to express interest.
001
Bradley Love @profdata.bsky.social · 28/01/2026
Are you a graduate student interested in working at Los Alamos National Laboratory (LANL) this summer? LANL has student internships, apply here: lanl.jobs/search/jobde... Please apply ASAP and before February 13th (sorry for the rush) 1/2
lanl.jobs
Computing & Artificial Intelligence (CAI) Division Graduate Intern at Los Alamos National Laboratory
Los Alamos National Laboratory is Hiring! Search available jobs or submit your resume now by visiting this link. Please share with anyone you feel would be a great fit.
120
Bradley Love @profdata.bsky.social · 25/11/2025
with @robmok.bsky.social and Xiaoliang "Ken" Luo
051
Bradley Love @profdata.bsky.social · 25/11/2025
Intuitive cell types don't necessarily play the ascribed functional role in the overall computation. This is not a message the field wants to hear as it suggests better baselines, controls, and some reflection. elifesciences.org/reviewed-pre... 2/2
elifesciences.org
1143
Bradley Love @profdata.bsky.social · 25/11/2025
"The inevitability and superfluousness of cell types in spatial cognition". Intuitive cell types are found in random artificial networks using the same selection criteria neuroscientists use with actual data. elifesciences.org/reviewed-pre... 1/2
elifesciences.org
44515
Bradley Love @profdata.bsky.social · 25/11/2025
Working with monkey data, we found neural representations stretched across brain regions to emphasize task relevant features on a trial-by-trial basis. Spike timing mattered over spike rate. Deep nets did the same. nature.com/articles/s41... 2/2
nature.com
Adaptive stretching of representations across brain regions and deep learning model layers - Nature Communications
How the brain adapts its representations to prioritize task-relevant information remains unclear. Here, the authors show that both monkey brains and deep learning models stretch neural representations...
020
Bradley Love @profdata.bsky.social · 25/11/2025
Exciting "new" work illustrating our broken publishing system. Seb presented this work online at neuromatch 2.0 at the height of the pandemic. Then, Xin-Ya worked years on addressing reviewer comments, which added some rigor but didn't change the message. 1/2
nature.com
Adaptive stretching of representations across brain regions and deep learning model layers - Nature Communications
How the brain adapts its representations to prioritize task-relevant information remains unclear. Here, the authors show that both monkey brains and deep learning models stretch neural representations...
1212
Bradley Love @profdata.bsky.social · 25/11/2025
We developed a straightforward method of combining confidence-weighted judgments for any number of humans and AIs. w Felipe Yáñez, Omar Valerio Minero, @ken-lxl.bsky.social 2/2
000
Bradley Love @profdata.bsky.social · 25/11/2025
When AI surpasses human performance, what's left for humans? We find that human judgment boosts performance of human-AI teams because humans and machines make different errors. cell.com/patterns/ful... 1/2
cell.com
Confidence-weighted integration of human and machine judgments for superior decision-making
When AI surpasses human performance, what can humans offer? We demonstrate that the performance of teams increases by integrating human judgments with those of machines. Integration is achieved by a s...
151
Reposted by Bradley Love
The Transmitter @thetransmitter.bsky.social · 04/11/2025
Researchers are using LLMs to analyze the literature, brainstorm hypotheses, build models and interact with complex datasets. Hear from @mschrimpf.bsky.social, @neurokim.bsky.social, @jeremymagland.bsky.social, @profdata.bsky.social and others. #neuroskyence www.thetransmitter.org/machine-lear...
thetransmitter.org
How neuroscientists are using AI
Eight researchers explain how they are using large language models to analyze the literature, brainstorm hypotheses and interact with complex datasets.
0269
Reposted by Bradley Love
PessoaBrain @pessoabrain.bsky.social · 06/10/2025
Michael X Cohen on why he left academia/neuroscience. mikexcohen.substack.com/p/why-i-left...
mikexcohen.substack.com
Why I left academia and neuroscience
Don't worry, this isn't yet another story of rage-quitting.
79636
Bradley Love @profdata.bsky.social · 18/07/2025
moderation@blueskyweb.xyz, send to me, or send directly to the Met (London police) who are investigating www.met.police.uk. I could see this being super distressing for a vulnerable person, so hope this does not become more common. For me, it's been an exercise in rapidly learning to not care! 2/2
met.police.uk
Home
Your local police force - online. Report a crime, contact us and other services, plus crime prevention advice, crime news, appeals and statistics.
030
Bradley Love @profdata.bsky.social · 18/07/2025
Some UK dude is trying to extort me, demanding money to not spread made-up stories. I reported to the poilice after getting flooded with phone messages I never listen to, etc. @bsky.app has been good about deleting his posts and accounts. If contacted, don't interact, but instead report to...1/2
240
Bradley Love @profdata.bsky.social · 13/06/2025
New blog w @ken-lxl.bsky.social, “Giving LLMs too much RoPE: A limit on Sutton’s Bitter Lesson”. The field has shifted from flexible data-driven position representations to fixed approaches following human intuitions. Here’s why and what it means for model performance bradlove.org/blog/positio...
bradlove.org
Giving LLMs too much RoPE: A limit on Sutton’s Bitter Lesson — Bradley C. Love
Introduction Sutton’s Bitter Lesson (Sutton, 2019) argues that machine learning breakthroughs, like AlphaGo, BERT, and large-scale vision models, rely on general, computation-driven methods that prior...
341
Bradley Love @profdata.bsky.social · 28/05/2025
New blog, "Backwards Compatible: The Strange Math Behind Word Order in AI" w @ken-lxl.bsky.social It turns out the language learning problem is the same for any word order, but is that true in practice for large language models? paper: arxiv.org/abs/2505.08739 BLOG: bradlove.org/blog/prob-ll...
https://bradlove.org/blog/prob-llm-consistency
341
Bradley Love @profdata.bsky.social · 14/05/2025
Bonus: I found it counterintuitive that (in theory) the learning problem is the same for any word ordering. Aligning proof and simulation was key. Now, new avenues open to address positional biases, better training and knowing when to trust LLMs. w @ken-lxl.bsky.social arxiv.org/abs/2505.08739
arxiv.org
Probability Consistency in Large Language Models: Theoretical Foundations Meet Empirical Discrepancies
Can autoregressive large language models (LLMs) learn consistent probability distributions when trained on sequences in different token orders? We prove formally that for any well-defined probability ...
210
Bradley Love @profdata.bsky.social · 14/05/2025
When LLMs diverge from one another because of word order (data factorization), it indicates their probability distributions are inconsistent, which is a red flag (not trustworthy). We trace deviations to self-attention positional and locality biases. 2/2 arxiv.org/abs/2505.08739
arxiv.org
Probability Consistency in Large Language Models: Theoretical Foundations Meet Empirical Discrepancies
Can autoregressive large language models (LLMs) learn consistent probability distributions when trained on sequences in different token orders? We prove formally that for any well-defined probability ...
300
Bradley Love @profdata.bsky.social · 14/05/2025
"Probability Consistency in Large Language Models: Theoretical Foundations Meet Empirical Discrepancies" Oddly, we prove LLMs should be equivalent for any word ordering: forward, backward, scrambled. In practice, LLMs diverge from one another. Why? 1/2 arxiv.org/abs/2505.08739
arxiv.org
Probability Consistency in Large Language Models: Theoretical Foundations Meet Empirical Discrepancies
Can autoregressive large language models (LLMs) learn consistent probability distributions when trained on sequences in different token orders? We prove formally that for any well-defined probability ...
340
Bradley Love @profdata.bsky.social · 17/02/2025
with @ken-lxl.bsky.social , @robmok.bsky.social , Brett Roads
210
Bradley Love @profdata.bsky.social · 17/02/2025
"Coordinating multiple mental faculties during learning" There's lots of good work in object recognition and learning, but how do we integrate the two? Here's a proposal and model that is more interactive than perception provides the inputs to cognition. www.nature.com/articles/s41...
nature.com
Coordinating multiple mental faculties during learning - Scientific Reports
Scientific Reports - Coordinating multiple mental faculties during learning
3319
Reposted by Bradley Love
Erin Grant @eringrant.me · 21/01/2025
Last year, we funded 250 authors and other contributors to attend #ICLR2024 in Vienna as part of this program. If you or your organization want to directly support contributors this year, please get in touch! Hope to see you in Singapore at #ICLR2025!
13714
Bradley Love @profdata.bsky.social · 13/12/2024
Thanks @hossenfelder.bsky.social for covering our recent paper, doi.org/10.1038/s415... Also, I want to spotlight this excellent podcast (19 minutes long) with Nicky Cartridge covering how AI will impact science and healthcare in the coming years, touchneurology.com/podcast/brai...
2141
Bradley Love @profdata.bsky.social · 27/11/2024
A 7B is small enough to train efficiently on 4 A100s (thanks Microsoft) and at the time Mistral performed relatively well for its size.
170
Bradley Love @profdata.bsky.social · 27/11/2024
Yes, the model weights and all materials are openly available. We really want to offer easy to use tools people can use through the web without hassle. To do that, we need to do more work (will be announcing an open source effort soon) and need some funding for hosting a model endpoint.
190
Bradley Love @profdata.bsky.social · 27/11/2024
While BrainBench focused on neuroscience, our approach is science general, so others can adopt our template. Everything is open weight and open source. Thanks to the entire team and the expert participants. Sign up for news at braingpt.org 8/8
braingpt.org
BrainGPT
This is the homepage for BrainGPT, a Large Language Model tool to assist neuroscientific research.
3110
Bradley Love @profdata.bsky.social · 27/11/2024
Finally, LLMs can be augmented with neuroscience knowledge for better performance. We tuned Mistral on 20 years of the neuroscience literature using LoRA. The tuned model, which we refer to as BrainGPT, performed better on BrainBench. 7/8
170
Bradley Love @profdata.bsky.social · 27/11/2024
Indeed, follow-up work on teaming finds that joint LLM and human teams outperform either alone, because LLMs and humans make different types of errors. We offer a simple method to combine confidence-weighted judgements. arxiv.org/abs/2408.08083 6/8
arxiv.org
Confidence-weighted integration of human and machine judgments for superior decision-making
Large language models (LLMs) have emerged as powerful tools in various domains. Recent studies have shown that LLMs can surpass humans in certain tasks, such as predicting the outcomes of neuroscience...
2100
Bradley Love @profdata.bsky.social · 27/11/2024
In the Nature HB paper, both human experts and LLMs were well calibrated - when they were more certain of their decisions, they were more likely to be correct. Calibration is beneficial for human-machine teaming. 5/8
160
Bradley Love @profdata.bsky.social · 27/11/2024
There were no signs of leakage from the training to test set. We performed standard checks. In follow-up work, we trained an LLM from scratch to rule out leakage; even this smaller model was superhuman on BrainBench arxiv.org/abs/2405.09395 4/8
arxiv.org
Matching domain experts by training from scratch on domain knowledge
Recently, large language models (LLMs) have outperformed human experts in predicting the results of neuroscience experiments (Luo et al., 2024). What is the basis for this performance? One possibility...
170
Bradley Love @profdata.bsky.social · 27/11/2024
All 15 LLMs considered crushed human experts at BrainBench's predictive task. LLMs correctly predicted neuroscience results (across all sub areas) dramatically better than human experts, including those with decades of experience. 3/8
1100
Bradley Love @profdata.bsky.social · 27/11/2024
To test, we created BrainBench, a forward-looking benchmark that stresses prediction over retrieval of facts, avoiding LLM's "hallucination" issue. The task was to predict which version of a Journal of Neuroscience abstract gave the actual result. 2/6
180
Bradley Love @profdata.bsky.social · 27/11/2024
"Large language models surpass human experts in predicting neuroscience results" w @ken-lxl.bsky.social and braingpt.org. LLMs integrate a noisy yet interrelated scientific literature to forecast outcomes. nature.com/articles/s41... 1/8
19279109
Bradley Love @profdata.bsky.social · 20/11/2024
Thanks Gary! I have no idea because I don't see how we get anyone to learn over more than a billion tokens. Maybe one could bootstrap some estimate from the perplexity difference between forward and backward, assuming we can get a sense of how that affects learning? Just off the top of my head...
000
Bradley Love @profdata.bsky.social · 19/11/2024
i am not seeing the issue. every method is the same, but the text is reversed. we even tokenize separately for forward and backward to make comparable. Perplexity is calculated over the entire option for the benchmark items. The difficulty doesn't have to be the same - it just turned out that way.
100
Bradley Love @profdata.bsky.social · 19/11/2024
For backward: Everything is reversed at the character level, including the benchmark items. So, the last character of the last word for each passage is the first and the first character of the first word is last. On the benchmark, as in the forward case, the option with lower perplexity is chosen.
100
Bradley Love @profdata.bsky.social · 19/11/2024
Instead of viewing LLMs as models of humans or stochastic parrots, we view them as general and powerful pattern learners that can master a superset of what people can. arxiv.org/abs/2411.11061 2/2
arxiv.org
Beyond Human-Like Processing: Large Language Models Perform Equivalently on Forward and Backward Scientific Text
The impressive performance of large language models (LLMs) has led to their consideration as models of human language processing. Instead, we suggest that the success of LLMs arises from the flexibili...
271
Bradley Love @profdata.bsky.social · 19/11/2024
"Beyond Human-Like Processing: Large Language Models Perform Equivalently on Forward and Backward Scientific Text" Our take is that large language models (LLMs) are neither stochastic parrots nor faithful models of human language processing. arxiv.org/abs/2411.11061 1/2
33611
Bradley Love @profdata.bsky.social · 19/11/2024
Has anyone tried this tool to follow back all of one's followers? github.com/jiftechnify/... It seems legit but I'm weary of giving a password to a third party website. So many people here so suddenly!
github.com
bsky-follow-back-all/ at main · jiftechnify/bsky-follow-back-all
Contribute to jiftechnify/bsky-follow-back-all development by creating an account on GitHub.
120
Reposted by Bradley Love
Blake Richards @tyrellturing.bsky.social · 28/08/2024
I fully support the last sentence of this abstract from @profdata.bsky.social : elifesciences.org/reviewed-pre... "...the complexity of the brain should be respected and intuitive notions of cell type, which can be misleading and arise in any complex network, should be relegated to history." 🧠📈 🧪
elifesciences.org
The inevitability and superfluousness of cell types in spatial cognition
112910
Reposted by Bradley Love
CogCompNeuro @cogcompneuro.bsky.social · 27/03/2024
🚨Submissions for #CCN2024 are now open at ccneuro.org 🚨 We welcome submissions for 2-page papers (deadline: 12 April) and Generative Adversarial Collaborations (GACs), Keynote+Tutorials, and (new this year!) Community Events (deadline: 5 April). Stay tuned: registration will open in early April!
21715
Bradley Love @profdata.bsky.social · 20/03/2024
Here's a recent talk I gave on the braingpt.org project, touching on findings from here, arxiv.org/abs/2403.03230. The video is sectioned for those wishing to skip the bits on explanatory vs. predictive approaches. youtu.be/sDt4-Q_jz7g?...
youtu.be
"Taming the neuroscience literature with explanatory and predictive models" (BrainGPT) - YouTube
Chapters:0:00 Introduction3:30 Process models (theoretical, explanatory)16:18 Models as tools (BrainGPT, predictive)39:03 ConclusionsModels can help scienti...
252
Reposted by Bradley Love
Clare Press @clarepress.bsky.social · 18/03/2024
📢 ERC-funded PhD in our team 📢. Curious about the mechanisms underlying learning and perception? Wish to study them via modelling and neuroimaging? This may be for you! Email me qus. Please circulate 🙏 www.findaphd.com/phds/project...   #HiSciSky #neuroskyence #PsySciSky
findaphd.com
Cognitive Neuroscience PhD at UCL at University College London on FindAPhD.com
PhD Project - Cognitive Neuroscience PhD at UCL at University College London, listed on FindAPhD.com
11123
Bradley Love @profdata.bsky.social · 09/03/2024
Catching up, slowly, on all these interesting comments. Yeah, having something good at prediction doesn't replace human scientists. There's a lot else to expertise and what we do. I do think having good predictive models does end up changing how we work, hopefully for the better.
020
Bradley Love @profdata.bsky.social · 07/03/2024
There were no signs of leakage from the training to test set. Our approach is science general, so others can adopt our template. Open source. Thanks to the entire team and the expert participants, t-shirt raffle for you all soon :) Sign up for news at braingpt.org 6/6
braingpt.org
BrainGPT
This is the homepage for BrainGPT, a Large Language Model tool to assist neuroscientific research.
051
Bradley Love @profdata.bsky.social · 07/03/2024
LLMs can be augmented with neuroscience knowledge for better performance. We tuned Llama-2-7b (chat) on 20 years of the neuroscience literature using LoRA. The tuned model, which we refer to as BrainGPT, performed better on BrainBench. 5/6
Fig 5 in preprint
141
Bradley Love @profdata.bsky.social · 07/03/2024
Both human experts and LLMs were well calibrated - when they were more certain of their decisions, they were more likely to be correct. Calibration is beneficial for human-machine teaming. 4/6
Fig 4 in preprint
130
Bradley Love @profdata.bsky.social · 07/03/2024
All 15 LLMs considered crushed human experts at BrainBench's predictive task. LLMs correctly predicted neuroscience results (across all sub areas) dramatically better than human experts, including those with decades of experience. 3/6
Fig 3 in preprint
151