Pith Number

pith:PYE5AVJQ

pith:2026:PYE5AVJQ7LLVLKZHZ3CL2Q4JIZ

not attested not anchored not stored refs resolved

Representation Without Reward: A JEPA Audit for LLM Fine-Tuning

Biswa Sengupta

JEPA-style auxiliaries change LLM hidden-state geometry but leave task accuracy unchanged on language-to-regex generation

arxiv:2605.15394 v1 · 2026-05-14 · cs.LG · cs.AI · stat.ML

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{PYE5AVJQ7LLVLKZHZ3CL2Q4JIZ}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

Hidden-state representation work and decoded-task accuracy are therefore weakly coupled in this regime; we accordingly reframe LLM-domain JEPA evaluation as a coupling problem.

C2weakest assumption

The natural-language-to-regex generation task with exact-match metric is sufficiently representative that a null result on it generalizes to the broader claim of weak coupling between hidden geometry and task signal in LLM fine-tuning.

C3one line summary

An empirical audit of 22 JEPA-style training auxiliaries on Llama-3.2-1B fine-tuning for regex generation finds no statistically significant task improvement after multiple-testing correction, even when auxiliaries visibly alter hidden-state geometry.

References

29 extracted · 29 resolved · 6 Pith anchors

[1] Making the world differentiable: On using self- supervised fully recurrent neural networks for dynamic reinforcement learning and planning in non-stationary environments, 1990

[2] A path towards autonomous machine intelligence, ver- sion 0.9.2, 2022

[3] Curious model-building control systems, 1991

[4] Self-supervised learning from images with a joint-embedding predictive architecture.arXiv preprint arXiv:2301.08243 2023

[5] arXiv preprint arXiv:2603.14482 (2026) 2026

Receipt and verification

First computed	2026-05-20T00:00:56.350056Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

7e09d05530fad755ab27cec4bd43894671e5ec255d7498f107a4968e91815db3

Aliases

arxiv: 2605.15394 · arxiv_version: 2605.15394v1 · doi: 10.48550/arxiv.2605.15394 · pith_short_12: PYE5AVJQ7LLV · pith_short_16: PYE5AVJQ7LLVLKZH · pith_short_8: PYE5AVJQ

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/PYE5AVJQ7LLVLKZHZ3CL2Q4JIZ \
  | jq -c '.canonical_record' \
  | python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 7e09d05530fad755ab27cec4bd43894671e5ec255d7498f107a4968e91815db3

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "3de13ac3eea96cb0c4d0c8c3dc748ab683b560aa36b140508fb53cfc0630ac65",
    "cross_cats_sorted": [
      "cs.AI",
      "stat.ML"
    ],
    "license": "http://creativecommons.org/licenses/by-nc-nd/4.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-05-14T20:27:32Z",
    "title_canon_sha256": "8921e762b44b31f71bf28a3d2f0d0d6aa84b965a52da3e407da36f114436b3b8"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.15394",
    "kind": "arxiv",
    "version": 1
  }
}