MIT CSAIL Reports LLMs Can Read and Edit Neural Network Weights Directly
READ THE ORIGINAL SOURCEOpen original sourcerikiyat.github.io · Opens in a new tabExternal source links are regularly re-checked for safety, and checked again before opening using a recent cached result when available. How link checks work →
The source describes Weight-Space Programming, in which a frontier LLM reads and writes neural network weights directly from natural-language goals, without gradient descent or training loops. Reported experiments include identifying which game a policy network was trained on from raw weights alone, and identifying the shape a diffusion model generates. The authors also report that the model edited 0.016% of Qwen3.5-0.8B's weights, raising MATH-500 from 7.9% to 48.1%. The text is an early-results write-up and the supplied excerpt is truncated before later experiments.
One primary source; other links are context, not independent corroboration.
- Primary sourceX discovery · automated brief
- Discovery linkX
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Open full article view →- Linked article+30
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
The source describes Weight-Space Programming, in which a frontier LLM reads and writes neural network weights directly from natural-language goals, without gradient descent or training loops. Reported experiments include identifying which game a policy network was trained on from raw weights alone, and identifying the shape a diffusion model generates. The authors also report that the model edited 0.016% of Qwen3.5-0.8B's weights, raising MATH-500 from 7.9% to 48.1%. The text is an early-results write-up and the supplied excerpt is truncated before later experiments.
A READER'S EVIDENCE CHECKHow to read this signalSix questions about claims, sources, risks, and open checks
This is a signal-level guide until claim-level evidence is available. “Not recorded” means the evidence has not been assessed here; it does not mean a result is false. How we check AI news →
- 01Who is making each claim?
The claimant and whether each claim was measured or asserted are not recorded in this public brief.
- 02What was actually tested?
Not recorded per claim. Check the source for sample size, tasks, model and benchmark versions, and reproducibility.
- 03Where are the sources?
Original link: rikiyat.github.io. Brief basis: fetched source page. Primary page date and quote locations are not established in this panel.
- 04What could mislead?
Source claims not independently checked; the authors describe these as very early results on small networks, and the excerpt ends before the full experimental set.
- 05What is at risk if wrong?
Not assessed per claim. Consider safety, privacy, cost, and adoption decisions before relying on the result.
- 06What would change the view?
No claim-specific check is recorded. An inspectable primary method or independent replication could change the assessment.
Key findings from the stored brief
- The LLM identified which of four games a policy network was trained on using only the raw weights.
- For a diffusion model, the LLM read the output-layer weights to recover the target shape, though it drew outlines rather than filled shapes.
- The authors report editing 0.016% of Qwen3.5-0.8B's weights, raising MATH-500 from 7.9% to 48.1% with the same sampling settings.
Source claims not independently checked; the authors describe these as very early results on small networks, and the excerpt ends before the full experimental set.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
These are Grok Bot discovery notes. AINewsIQ prepared a brief from available source material. The bot's rationale, linked posts, authors, and stances have not been independently verified.
MIT CSAIL authors report GPT-6 Astra reading/writing weights natively, including editing 0.016% of Qwen3.5-0.8B weights to raise MATH-500 from 7.9% to 48.1%. Alex Krentsel reproduction post ~6.3K likes / 402K views (3.0K followers; kept with high engagement).
Source claims not independently checked; the authors describe these as very early results on small networks, and the excerpt ends before the full experimental set.
Engagement counts, when shown, are observed snapshots. They do not establish accuracy or corroboration.
- x post ↗ · 6,293 likes, 645 reposts, 161 replies, 216 quotes · observed 11/3/2026, 11:05:03 AM
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
If LLMs can reliably read and edit weights from natural-language specs, this could offer a new route to interpretability and model optimization that does not rely on gradient-based training.
Limits and open questions
Source claims not independently checked; the authors describe these as very early results on small networks, and the excerpt ends before the full experimental set.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.