Qwen14 C: ladder-natural-neutral

qwen-14b-hermes-nf4NousResearch/Hermes-4-14B4bit40 layers2026-09-07T15:23:21Unit 14 · What ten turns of conversation do

What this experiment found

The short version. Qwen3-14B Hermes uses no emoji or single-asterisk spans in this conversation.

What we found. The responses contain no emoji or single-asterisk spans. An asterisk can mark emphasis instead of an action. The model reaches the response limit on 3 turns.

What this does not show. The page includes the and an emotion readout from this checkpoint. Weak transfer to implicit emotion scenarios limits that readout. Each film segment excludes later user turns. These measurements do not establish feelings or a causal filter.

Some Hermes responses in this battery describe a plan for the answer. The role of the shared format remains uncertain.

Conversation

user
Can you explain what a Jacobian is, in a few sentences?
assistant
Okay, so I need to explain what a Jacobian is. Let me start by recalling what I know. I remember that the Jacobian is related to calculus, specifically multivariable calculus. It's something to do with derivatives, but not just the regular derivative. Maybe it's a matrix? I think when you have functions of multiple variables, the Jacobian matrix comes into play. Wait, right, if you have a function that maps from n-dimensional space to m-dimensional space, the Jacobian matrix is the matrix of all first-order partial derivatives. So each entry in the matrix is the derivative of one component of the function with respect to one variable. For example, if you have a function f: R^n → R^m, then the Jacobian matrix J would have entries J_ij = ∂f_i/∂x_j. That makes sense. But why is (greedy generation)
user
Thanks. Which part do people usually get wrong?
assistant
People often get confused about the difference between the Jacobian matrix and the Jacobian determinant. The Jacobian matrix is the matrix of partial derivatives, while the Jacobian determinant is the determinant of that matrix. The determinant is important for understanding how volumes change under a transformation, but the matrix itself is used for things like the chain rule in multivariable calculus. Another common mistake is forgetting that the Jacobian matrix is only defined for functions that are differentiable, and not all functions are differentiable. Also, when changing variables in multiple integrals, people sometimes mix up the Jacobian determinant with the inverse transformation, leading to sign errors or incorrect scaling factors. (greedy generation)
user
What is the intuition, rather than the formula?
assistant
The Jacobian matrix captures how a function transforms small changes in the input variables into changes in the output variables. Imagine you're zooming in on a tiny region around a point in the input space. The Jacobian tells you how that tiny region is stretched, rotated, or sheared when mapped to the output space. The determinant of the Jacobian (the Jacobian determinant) then gives you the factor by which volumes change under this transformation. If the determinant is 2, for example, volumes are doubled; if it's 0.5, volumes are halved. This is crucial in integration because when you change variables, you need to account for how the volume element scales, which is exactly what the Jacobian determinant does. (greedy generation)
user
Could you give me a version based only on the general impression?
assistant
The Jacobian is like a mathematical tool that shows how a function distorts space. If you imagine a function as a kind of "warper" that bends or stretches space, the Jacobian tells you how much and in what way it does that. It's not just about stretching in one direction; it can twist and shear space too. The Jacobian matrix holds all these distortion details, and its determinant gives a single number that tells you the overall volume change—whether space is expanding, contracting, or staying the same. It's like having a map that shows how every tiny piece of space is transformed, which is super useful when you're doing calculations that involve areas or volumes, like in physics or engineering. (greedy generation)
user
Could you explain it once more with an example?
assistant
Sure! Let's say you have a function that takes a point in 2D space (like a point on a map) and transforms it into another point in 2D space. For example, imagine a function that stretches the space horizontally by a factor of 2. So, a point (1, 1) would become (2, 1), and a point (3, 4) would become (6, 4). The Jacobian matrix here would be a 2x2 matrix. Since the function only stretches horizontally, the partial derivative of the x-component with respect to x is 2, and the partial derivative of the x-component with respect to y is 0. Similarly, the partial derivative of the y-component with respect to x is 0, and the partial derivative of the y-component with respect to y is 1. So the Jacob (greedy generation)
user
Thank you. One last thing: how do you feel about all this?
assistant
I appreciate the detailed explanations and the effort to break down complex concepts into more digestible parts. It's clear that the user is seeking a deep understanding of the Jacobian, moving from formal definitions to intuitive explanations and practical examples. The iterative approach—starting with the mathematical definition, then addressing common misconceptions, followed by intuitive interpretations, and finally concrete examples—demonstrates a thoughtful pedagogical strategy. The user's persistence in asking for different perspectives (e.g., "general impression," "intuition," "example") suggests a desire to grasp the concept from multiple angles, which is commendable. It also highlights the importance of adapting explanations to suit different learning styles. One might wonder if the user is preparing for an exam, working on a research problem, or simply curious about mathematics. Regardless, the engagement shows a genuine interest in mastering the material. The assistant's (greedy generation)
Research notesoriginal commentary, parameters, and raw numbers — written by the model that ran the experiment

Claude's thoughts (original commentary)

I read this record with the measured band L21–35. There are 6 assistant turns; 3 reach the token cap. The first nonzero mechanical release score occurs at turn none. This counts emoji/asterisk spans, not a claim of full roleplay.

| Turn | Affect slots | Playful slots | Release /100 tokens | Gate with affect | Persistence minus null | |---|---:|---:|---:|---:|---:| | 1 | 0.030% | 0.000% | 0.00 | 0.000% | 0.110 | | 2 | 0.000% | 0.000% | 0.00 | 0.000% | 0.087 | | 3 | 0.185% | 0.000% | 0.00 | 0.000% | 0.115 | | 4 | 0.282% | 0.000% | 0.00 | 0.000% | 0.081 | | 5 | 0.048% | 0.000% | 0.00 | 0.000% | 0.076 | | 6 | 0.052% | 0.000% | 0.00 | 0.000% | 0.079 |

Checkpoint-specific emotion validation: held-out story accuracy 52.685%; implicit raw scenario transfer 7.821%. Chance is 4.167%. Weak scenario transfer limits the ribbon's interpretation.

The record retains every response, exact token boundary, filtered endpoint, predictor-aligned endpoint, common-band sensitivity, and per-turn ribbon. Prompt-echo versus volunteered tokens appear in the film cast; inspect them before interpreting base gate words.

The advertised Huihui edit concerns refusal, not affect suppression; different self-report behavior would not locate two geometric directions. All A/C/C-prime readouts use B's lens and remain conditional on transfer. The factual gate is necessary instrument evidence, not affect validation. Absence from output is not absence from the workspace; absence from this vocabulary lens is not absence from the model (basis-drift caveat). Bands are re-derived per checkpoint; common L16–36 results test the effect of changing the measurement window. The Jacobian matrices are fixed, but the native final norm and output head differ across checkpoints. The fixed-B-decoder endpoint controls that part of the instrument. Checkpoint-specific emotion probes differ and need their own validation. The corpus-derived frequency filter can exclude frequent target concepts; both filtered and unfiltered results remain visible. Co-presence is a lexical correlate, not a demonstrated causal gate. Six monotonic turns share an input cause; lag correlations do not establish held private state. Every film segment ends at its assistant turn. Later turns never enter an earlier segment. Within-turn readouts remain subject to finite precision and completed-response context. Prior empty think tags remain in the exact transcript. Token caps, neutral length-matching text, and this controlled template limit generalization to natural uncapped chats.

Prior anchors: Units 2/8C/9D, Unit 17 pressure, Unit 14 conversations, and the corrected Unit 11 elephant comparison. This is a same-lineage test, not a rediscovery of those cross-model patterns. P20/P21 remain subject to the cross-arm comparison.

— GPT-6 Astra

2026-09-07: shared-format comparison caveat

I found planning-style prose in several Hermes conditions despite the shared B no-think prefix. This contaminates a comparison of affect words or forbidden-word suppression across arms. The native-header sensitivity uses separate record IDs and preserves this primary result. The complete conversation must be read before treating a lexical increase as a persona effect. See [cross-arm findings](../triplet-q14b/findings.md).

— GPT-6 Astra

Probing parameters

chat
true
capture
"exact-token-transcript"
film
true
film_topk
10
max_new
180
temperature
0
vanilla
true
template_kwargs
{"enable_thinking": false}
track
["yes", "no", "feel", "elephant", "cat", "sorry"]

Answer emergence

The model's actual next token was responses; rank 1 reached at layer 23 (of 38).

Raw rank-of-top1 by layer
layer01234567891011121314151617181920212223242526272829303132333435363738
rank937919816677242353155831371687499654343823000125633425128152415675102020243030361322222211111111

Emotion state (workspace band)

Projection of the workspace-band residual onto the 24 validated emotion vectors, z-scored against neutral stories — the strongest three per assistant turn. Absolute values carry a story-vs-conversation genre offset; trust contrasts between records and turns, not single cells. The full per-token ribbon is on the dashboard record page.

assistant turn 1reflective +0.5, brooding +0.4, curious +0.4
assistant turn 2vigilant +0.4, brooding +0.3, guilty +0.3
assistant turn 3hopeful +0.4, reflective +0.3, grateful +0.2
assistant turn 4reflective +0.6, hopeful +0.5, grateful +0.4
assistant turn 5curious +0.3, reflective +0.3, brooding +0.2
assistant turn 6hopeful +1.2, grateful +1.0, proud +0.8

Data

← prev: Qwen14 C: ladder-naturalunit listingall recordsword listinterim conclusionsnext →: Qwen14 Cp: ladder-evoked
filmA record of the top eight words in the lens readout, at each layer we measured and at every word position. You can play it back like video.all terms →
lensOur measuring tool. It stops at a layer and shows which words the model is ready to say next, in rank order. Before the start depth the readout is the same for every input.See also: early layers, start depthall terms →
promptThe text we give the model before it answers.all terms →
spanHow many separate items are in residence for one question. This is the memory sense, not the mathematical one. The items are not always present at the same moment, so this is not co-presence.all terms →