#Neural
LightNews LightNews
philpax.me
what the fuck, frontier LLMs can directly interpret and edit the weights of smaller models rikiyat.github.io/blog/weight-...
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 2, 2026 at 7:06 PM
alicegodden.bsky.social
The final paper from my PhD is out in Developmental Biology! 🐸

We show that the #microRNA miR-219 is needed for neural border and neural crest development in #Xenopus neurulas 🧬

Highlights
•MiR-219 depletion expands the neural border territory and disrupts neural crest specification.

More below!
MicroRNA miR-219 is required for neural border and neural crest development in Xenopus neurulas
At head levels, neural crest (NC) multipotent stem cells are specified in the ectoderm territory located between the anterior neural plate ectoderm an…
www.sciencedirect.com
November 1, 2026 at 3:27 PM
esqueer.net
A new article discusses how the latest LLM models can now directly edit and manipulate neural network model weights to achieve targeted outcomes such as improving task completion. This is a new emergent feature.

rikiyat.github.io/blog/weight-...
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 3, 2026 at 2:10 AM
jlp.systems
Warm Neural Kitchen
November 2, 2026 at 5:50 AM
adler.dev
it's Neural Net November (NNN)
November 3, 2026 at 6:55 AM
easterncoyote.bsky.social
Would people want to read your mind?

“BCI.”
“Neural dust.”
“Neural data.”
“Brain mapping.”

Those spooky phrases weren’t invented by a cyberpunk novelist. They’re real technologies.

Artificial telepathy isn’t conspiracy bullshit. Your neural data is up for sale.

Enjoy your weekend!

#neuroethics
karenhao.bsky.social Karen Hao @karenhao.bsky.social · May 25
The Pope has weighed in: AI isn't magic. It's currently produced by a deeply exploitative, extractive supply chain. It challenges human dignity; could fuel a new colonialism. And among the greatest challenges of our time will be to redirect its path of development.
www.vaticannews.va/en/pope/news...
Breaking the chains of new forms of slavery 173. This distorted view of the human person is reflected today in various forms of servitude directly linked to the digital economy. Nothing in the world of AI is immaterial or magical. Every seemingly immediate and flawless response is the result of a long chain of mediation, involving vast networks of natural resources, energy infrastructure and, above all, people. A significant part of the digital economy’s functioning relies on the silent work of millions of people engaged in essential yet largely unseen activities, such as data labeling, model training and content moderation, often involving disturbing material. In many cases, these workers are young people, predominantly women, working under demanding conditions for minimal wages. Added to this invisible labor is the even harsher work of extracting the resources required for the production of the devices and microprocessors on which AI depends. In some regions of the world, children and adolescents work in dangerous conditions, crushing the materials from which rare earth elements are extracted. The bodies of these people are scarred, injured and worn down so that computational flow may continue uninterruptedly. Furthermore, criminal networks use online platforms, messaging systems, anonymous payment methods and profiling techniques in order to recruit, control and transport victims of trafficking — very often minors — reducing men and women to “data” to be tracked and “packages” to be moved around within the same digital circuits that support much of the global economy. This reality deeply challenges the moral conscience of our time. It is not enough to invoke efficiency, nor to celebrate the benefits of innovation, if they are built on a chain of exploitation that remains deliberately hidden. If technology promises emancipation, yet produces new forms of global subordination, it stands in contradiction to the fundamental principle of human dignity. 178. Even today, colonialism assumes new forms. It no longer dominates only bodies, but appropriates data, transforming personal lives into exploitable information. Entire regions, especially those marked by structural fragility and limited geopolitical relevance, are currently subjected to a new mindset of extraction: that of health data, epidemiological profiles, genetic maps and demographic information. These have become the new “rare earths” of power: vital data which, once aggregated and analyzed, can be used to train predictive models, guide investment strategies, anticipate crises and, above all, determine who and what is deemed to matter. Those who control the health data of entire peoples — often collected under the pretext of aid, research or innovation — possess a structural leverage over the future, for they can shape needs and markets. They can also decide, before others, to whom medicines, investments and protections will be allocated. Here lies one of the most urgent moral challenges of our time: to ensure that shared knowledge becomes a true common good rather than an instrument of dominance. This requires restoring to individuals not only the data that describes them, but also the ability to decide how it is used, by whom and for whose benefit. Otherwise, the digital age will not be post-colonial, but colonial in another form.
November 1, 2026 at 8:03 AM
earlkmiller.bsky.social
Brain-like robustness likely comes from feedback and randomness working together

Robust inference via top-down feedback and neural variability in recurrent vision models
www.cell.com/patterns/ful...
#neuroscience
Robust inference via top-down feedback and neural variability in recurrent vision models
Greco et al. show that top-down feedback improves the sensory robustness of recurrent vision models when combined with neural variability via dropout. Neural variability prevents unit-level co-adaptat...
www.cell.com
November 1, 2026 at 5:06 PM
ponder.ooo
interesting
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 2, 2026 at 7:41 PM
eriskii.net
this is absurd‽ What the fuck do you mean Astra can look at a network's weights, with no tools, and go "yeah this guy plays breakout"????
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 2, 2026 at 7:12 PM
rainboweyeddog.bsky.social
New Neural pathways November #digidognsfw
November 1, 2026 at 4:23 PM
thomasfuchs.at
an anagram of Samuel Harris Altman is "Alarmist Neural Sham"
November 1, 2026 at 2:21 PM
basin.zone
So uh, on RSI.

rikiyat.github.io/blog/weight-...
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 2, 2026 at 7:14 PM
gutlin.bsky.social
Privileged representational axes in biological and artificial neural networks

www.nature.com/articles/s41...
Privileged representational axes in biological and artificial neural networks - Nature Human Behaviour
Khosla et al. show that both brains and neural networks consistently favoured certain axes for representing the natural world.
www.nature.com
November 1, 2026 at 9:49 AM
emollick.bsky.social
If this holds up, it is big: "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." rikiyat.github.io/blog/weight-...
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 2, 2026 at 9:00 PM
hikikomorphism.bsky.social
I have experimental evidence that, by using metacog, LLMs are able to both model and modify their own activations in unexpected ways. I wonder if I can leverage that: rikiyat.github.io/blog/weight-...
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 2, 2026 at 11:55 PM
thebadcode.com
on any given day there's maybe 10x as much stuff that i need to pay attention to as i have time for and that ratio is slipping at an accelerating rate
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 2, 2026 at 7:25 PM
vcarchidi.bsky.social
Counter-hype was fast on this one...

www.seangoedecke.com/llms-are-not...
LLMs are not superhuman at reading and writing neural network weights
A recent paper argues that frontier LLMs can work with neural network weights directly. This seems superhuman, but it isn’t. The paper shows that GPT-6 Astra…
www.seangoedecke.com
November 3, 2026 at 3:38 PM
nature.com
Nature research paper: Emotion regulation capacity lacks a shared neural signature

go.nature.com/4klnhMD
Emotion regulation capacity lacks a shared neural signature - Nature
Sensorimotor and amygdala activation, but not canonical prefrontal control circuits, indicate emotion regulation capacity.
go.nature.com
November 2, 2026 at 11:23 AM
rachel.supply
?
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 2, 2026 at 7:10 PM
cee.wtf
Neural network nut November
November 3, 2026 at 3:40 AM
trippystachio.bsky.social
sometimes my neural network makes some connections and i just gotta go with the flow man
November 2, 2026 at 7:49 PM
faz.ms
It’s just autocompleting from plagiarized stack overflow answers where people where asking which specific neurons to tweak to make their neural net drive a car better
November 2, 2026 at 10:43 PM
ophelia168.bsky.social
a neural interface cable is a kind of leash if you think about it
November 2, 2026 at 2:59 PM
realmotleysue.bsky.social
Fire those neural synapses
Yea
ALT: Yea
static.klipy.com
November 3, 2026 at 2:36 AM
cara.city
this is kinda bonkers? LLMs being used to natively edit and improve the weights of other LLMs for a given task:

rikiyat.github.io/blog/weight-...
Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
rikiyat.github.io
November 2, 2026 at 8:53 PM