Pinned Tweet
Exploring Direct Tensor Manipulation in Language Models: A Case Study in Binary-Level Model Enhancement: areu01or00.github.io/Tensor-…
Turn Qwen TTS perpetually angry
417 earlier mapped neurons, d_L ∈ ℝ¹⁰²⁴, L20–27
Add : v̂_L, L20–26, v̂ ⟂ d_L, v̂ ⟂ <eos>
h_L += 0.03·‖h_L‖·v̂_L every frame @ 12.5 Hz
read_direction: L8#1968↓ L10#888↓ L10#2660↑ L15#3057↓
Result :
f0 : 197 to 379 Hz
Spectral centroid shift: +1135
The Lazarus research by @DimitrisPapail should open fun interpretability objectives.
Neuron space editing Qwen3-TTS
Target : talker MLP neurons
Second-voice trait, β3 (400 mapped + add to activation before down_proj):
L23#1430 ↑ +5.287
L24#527 ↑ +1.674
L21#175 ↑ +0.315
L22#142 ↓ −2.010
L22#555 ↓ −0.220
Reprogramming dormant neurons : Qwen TTS
Ever wondered why Qwen3-TTS sometimes breathes or laughs out of nowhere? Agent found neurons behind it.
The details:
Agent found 417 neurons to be specific (about 0.48% of the talker's MLP neurons) and 61 of them are dormant. Amplified together, they give the following effects effect:
1. Breathes as words are spoken
2. Hushed ASMR
3. Natural laughs as language is being spoken
4. Enhanced whispering
5. 2nd speaker appearing.
Comparisons and neuron map repo in the comments below.
Inspired by recent work on Lazarus by @DimitrisPapail , which reprogrammed dormant neurons.
Reprogramming dormant neurons : Qwen TTS
Ever wondered why Qwen3-TTS sometimes breathes or laughs out of nowhere? Agent found neurons behind it.
The details:
Agent found 417 neurons to be specific (about 0.48% of the talker's MLP neurons) and 61 of them are dormant. Amplified together, they give the following effects effect:
1. Breathes as words are spoken
2. Hushed ASMR
3. Natural laughs as language is being spoken
4. Enhanced whispering
5. 2nd speaker appearing.
Comparisons and neuron map repo in the comments below.
Inspired by recent work on Lazarus by @DimitrisPapail , which reprogrammed dormant neurons.
Invoking Whispering words at will with Neuron Space Edits in Qwen TTS.
Reprogramming dormant neurons : Qwen TTS
Ever wondered why Qwen3-TTS sometimes breathes or laughs out of nowhere? Agent found neurons behind it.
The details:
Agent found 417 neurons to be specific (about 0.48% of the talker's MLP neurons) and 61 of them are dormant. Amplified together, they give the following effects effect:
1. Breathes as words are spoken
2. Hushed ASMR
3. Natural laughs as language is being spoken
4. Enhanced whispering
5. 2nd speaker appearing.
Comparisons and neuron map repo in the comments below.
Inspired by recent work on Lazarus by @DimitrisPapail , which reprogrammed dormant neurons.
Lobotomy on AI is coming soon to a LLM near you – at Bongaigaon, India
Reprogramming dormant neurons : Qwen TTS
Ever wondered why Qwen3-TTS sometimes breathes or laughs out of nowhere? Agent found neurons behind it.
The details:
Agent found 417 neurons to be specific (about 0.48% of the talker's MLP neurons) and 61 of them are dormant. Amplified together, they give the following effects effect:
1. Breathes as words are spoken
2. Hushed ASMR
3. Natural laughs as language is being spoken
4. Enhanced whispering
5. 2nd speaker appearing.
Comparisons and neuron map repo in the comments below.
Inspired by recent work on Lazarus by @DimitrisPapail , which reprogrammed dormant neurons.
Given the Lazarus research, its obvious we need better tooling for observability and interpretability. I urge everyone to develop your own harnesses. One such harness in action of agent mapping speech pattern nuerons of Qwen TTS.
SOUND ON : If you are a Tolkien nerd.
Reprogramming dormant neurons : Qwen TTS
Ever wondered why Qwen3-TTS sometimes breathes or laughs out of nowhere? Agent found neurons behind it.
The details:
Agent found 417 neurons to be specific (about 0.48% of the talker's MLP neurons) and 61 of them are dormant. Amplified together, they give the following effects effect:
1. Breathes as words are spoken
2. Hushed ASMR
3. Natural laughs as language is being spoken
4. Enhanced whispering
5. 2nd speaker appearing.
Comparisons and neuron map repo in the comments below.
Inspired by recent work on Lazarus by @DimitrisPapail , which reprogrammed dormant neurons.
Tensor-Slayer retweeted
very cool work!
Tensor has been at it since last year, way before weight space programming was a thing. Please do check it out, and engage for much deserved reach!
Reprogramming dormant neurons : Qwen TTS
Ever wondered why Qwen3-TTS sometimes breathes or laughs out of nowhere? Agent found neurons behind it.
The details:
Agent found 417 neurons to be specific (about 0.48% of the talker's MLP neurons) and 61 of them are dormant. Amplified together, they give the following effects effect:
1. Breathes as words are spoken
2. Hushed ASMR
3. Natural laughs as language is being spoken
4. Enhanced whispering
5. 2nd speaker appearing.
Comparisons and neuron map repo in the comments below.
Inspired by recent work on Lazarus by @DimitrisPapail , which reprogrammed dormant neurons.
Cc @DimitrisPapail would love to hear your comments on the approach Opus came up with for TTS target : Amplify active and dormant neurons to work with behaviours. I thought <Meta Optimiser> analogy for Opus here is perfect. It built itself a spectrogram tool to “see” the output and sort of reverse engineered and found responsible neurons, which i thought was quite amusing. Atleast in Neuron Space Editing context.
226
I dont want to sound smart ass when top researchers are deliberating on ideas but the timeline i have been advocating for the past 2 years is here and its glorious.
Dimitris is right. Opus 5.5 is great <meta optimiser>. It is also just as important HOW you leverage it , especially in context of WSP.
At present , as things stand :
WSP is a great avenue to “post train” SLMs to improve baselines. Something tagetted, a specific behaviour, or detect and remove inefficiencies by letting Opus reason about weights of the target LLM.
Where this i see things heading :
As networks get smarter and interpretability gets easier, more and more complex weight space edits will start to impact the post train phase. And “WSP” becomes an essential post train step right out the lab gates. Both with safety and improvement lens.
Replying to @DimitrisPapail @_arohan_
ok longer rant:
Researchers love telling stories. “My algorithm is better than yours.” and “Look how clean my idea is compared to yours.” But at the end of the day, if you let a frontier agent build something bespoke, completely untethered to anyone’s feelings, it will crush whatever your favorite thing was: inference stack, your optimizer, your model, whatever.
Agents DO NOT care about a good story. They care about the sweet reward of crushing the goal you gave them. And they will get there by building a frankenstein artifact that works better than anything a human has ever made, because they have a superhuman attention span (pun intended), they never get tired of trying stuff, and they don’t care if the final thing is the ugliest shit ever.
I’ve watched this happen again and again since March, and every time I bet against it I was wrong.
Long live Opus 5.5, the best meta optimizer that ever existed.
7
837
Tensor-Slayer retweeted
The ability to terraform intelligence over a globally distributed swarm has always been the mission.
We present a 30B MOE run,
The Internet is the center .
Higher 🚀
Most of today's leading open models are mixture-of-experts. Crucible, Templar's pre-training platform, has finished its first MoE test run.
The model follows the Qwen3-30B-A3B architecture from @Alibaba_Qwen, with 30B parameters and 128 experts, 8 of them active per token. On 32 A100 GPUs, two replicas of four pipeline stages averaged about 200K tokens per second over more than 50B tokens.
Watch our runs live at tplr.ai/dashboard
Everyone who is interested in coming up with weight space programming edits , few more things i would like to add :
1. It is essential that you develop your own harness with great care
2. Current frontier LLMs, since not trained well on these tasks, tend to bring priors and biases into their approach to edits and so its YOUR responsibility to ensure correctness without drowning yourself into useless check proposed by LLMs on autopilot.
3. Start with a base objective. Test what base model can already do. Establish baselines. Set up activation probes and neuron lens and establish circuits (roughly*)
4. Treat results from steps 1-3 as research corpus to then ask the frontier llm to come up with edits given the corpus.
Replying to @TensorSlay
yeah those are the tricky areas.
I used an open source Q&A dataset of 20k questions and answers and adapted it based on my writings, then narrowed down to 2048 multi-turn conversational evals.
I also had to change @AmpCode ‘s default prompt through a custom plugin to make Astra simply not stop until it reaches the goal (no milestones or half assed results).
It takes a while to generate, test and iterate, but so far the results are nonetheless impressive.







