Skip to content

Topic

Knowledge Distillation

A machine-learning technique that trains smaller "student" models to mimic larger "teacher" models' outputs, transferring capability cheaply.

Current stories

buildOne report1 publisher

Distillation pulls hidden quirks out of AuditBench models when the student shares their base

Distilling AuditBench's secret-keeping Llama fine-tunes into their instruct base yields students that admit hidden quirks far more often than the originals. The paper finds the method works poorly when the student comes from a different pretrained base.

Publishers:lesswrong.com

Reality

Evidence45
Adoption
Not scored
Hype gap+10
Incentives
Not scored
Confidence40
investConfirmed4 publishers

PewDiePie says OpenAI banned him twice as he trained his Ajax model on GPT-5.6 Sol outputs

PewDiePie says OpenAI banned him twice while he used GPT-5.6 Sol outputs to train Ajax, a 9-billion-parameter model for home PCs. OpenAI has not commented publicly, but his account shows that a team training its own model on a frontier lab's answers can lose API access partway through the build.

Perspective Coverage

4 publishers
Builder
Builder 45%
Operator
Operator 35%
Investor
Investor 20%

Reality

Evidence45
Adoption5
Hype gap+35
Incentives60
Confidence55
buildOne report1 publisher

One-word answers leak a fine-tuned model's coding skill to a student

Researchers from five institutions lifted a student model 5.34 points over a control on HumanEval+ using 5,664 one-word answers from a code-tuned teacher. None of those answers contained code, so a filter that screens outputs for task content would have passed every one.

Publishers:theneuron.ai

Reality

Evidence45
Adoption
Not scored
Hype gap+15
Incentives
Not scored
Confidence40
securityConfirmed8 publishers

NSA, CISA and FBI name six Chinese AI companies as industrial-scale distillers of US models

The conduct described is authorised API traffic at scale rather than an intrusion, the detection asked for is billing telemetry providers already hold, and on Beijing's role the advisory goes no further than likely government awareness.

Perspective Coverage

8 publishers
Builder
Builder 30%
Operator
Operator 50%
Investor
Investor 20%

Reality

Evidence55
Adoption
Not scored
Hype gap+20
Incentives55
Confidence60
buildOne report1 publisher

Sony and UMG's second Suno suit goes after the preference clicks behind v6

Sony and UMG argue that a licensed retrain does not clear what Suno's older models left behind in user preference data and distilled behavior. A Massachusetts federal court now has to rule on that.

Publishers:runtimewire.com

Reality

Evidence58
Adoption52
Hype gap+6
Incentives78
Confidence57
buildOne report1 publisher

Caliber scales its extraction noise by each model's median top-two logit margin

An arXiv defense adds Gaussian noise to a served model's logits and solves for the scale that hits a chosen accuracy target, and it bounds how many repeated queries an attacker needs to average that noise away.

Publishers:arxiv.org

Reality

Evidence44
Adoption8
Hype gap+12
Incentives
Not scored
Confidence42
investConfirmed3 publishers

Apple reportedly pays Google $1 billion a year for cloud infrastructure powering Siri's new AI model

iOS 27 shipped Monday with a Siri built by distilling Gemini's answers into Apple's own on-device models, gated behind an English-only waitlist and unavailable in the EU and China. Paid cloud limits come later.

Perspective Coverage

3 publishers
Builder
Builder 33%
Operator
Operator 40%
Investor
Investor 27%

Reality

Evidence58
Adoption38
Hype gap+24
Incentives76
Confidence57
buildOne report1 publisher

LinkedIn's job-search ranker queries its teacher models live during training

A custom SGLang framework feeds teacher outputs into the loss while training runs. The reported eightfold speedup comes from a stack of separate optimizations, and the named multipliers reach about 5.5x on their own.

Publishers:infoq.com

Reality

Evidence45
Adoption50
Hype gap+15
Incentives60
Confidence50
productOne report1 publisher

CISA, NSA and FBI warn US AI firms of industrial-scale distillation by Chinese companies

CISA, the NSA and the FBI want US model providers to catch industrial-scale distillation hiding inside traffic that looks like a busy enterprise customer, then quietly degrade the answers, which makes false positives a product decision.

Publishers:techradar.com

Reality

Evidence46
Adoption
Not scored
Hype gap+32
Incentives70
Confidence42
securityOne report1 publisher

NSA advisory tells AI providers to subtly degrade answers for suspected distillation accounts

The joint NSA, CISA and FBI advisory names six China-based companies and describes billions of tokens taken through paid APIs, cloud resellers and proxy transfer stations, which turns billing telemetry into a detection surface.

Publishers:cisa.gov

Reality

Evidence48
Adoption30
Hype gap+32
Incentives76
Confidence55
buildOne report1 publisher

The judge went synthetic first, which tells you which part of your pipeline is next

Latent Space dates five flips from human-made to model-made since 2022. The order they happened in predicts more than the annual cadence the essay claims for it.

Publishers:latent.space

Reality

Evidence42
Adoption58
Hype gap+32
Incentives55
Confidence48