GPT-2 news
The latest GPT-2 news, : Scoring Explanations Without Reading Them: Structural Shortcuts in Automated-Interpretability.
6 stories mentioning GPT-2 since 16 September 2026, newest first, each with its sources and discussion. Follow to see new ones on your front page.
-
Scoring Explanations Without Reading Them: Structural Shortcuts in Automated-Interpretability
A new study reveals that automated interpretability scores for sparse-autoencoder features can be calculated without actually reading the explanations. Researchers found that over 90% of activating examples are cropped…
1 source primary sourcearXiv cs.CL -
Researchers prove Prefix-Tuning can exactly steer pretrained transformers
A new arXiv paper titled “Prefix‑Tuning for Arbitrary Output Sequences on Pretrained Transformers” provides a theoretical proof that a task‑specific prefix can be constructed to make a pretrained autoregressive…
1 source primary sourcearXiv cs.CL -
Arena Physica’s Heaviside EM model joins Microsoft Discovery for faster design checks
Arena Physica announced on October 13 that its foundation model for electromagnetism, Heaviside, is now available on Microsoft Discovery, a platform for agentic engineering. The model predicts full 3D electromagnetic…
1 sourceUnite.AI -
Timnit Gebru says AI existential risk narrative distracts from real harms
Timnit Gebru, a prominent AI researcher and coauthor of the 2021 Stochastic Parrots paper, argues that framing AI as an existential threat is a harmful distraction from immediate risks like data exploitation,…
1 sourceWired AI -
Opinion: AI may pose existential risk yet some call for continued development
The author critiques the current AI discourse, noting that leading firms label the technology as an existential risk while simultaneously pouring vast sums into its development. They point out that even a ~10%…
1 sourcecefboud.com -
Functionalizer cuts vocab size and boosts code modeling with lossless tokenization
The paper introduces the Functionalizer, a pre‑tokenization framework that separates orthographic and structural word variations into a stream of opcodes and a canonical base token. By encoding transformations such as…
1 source primary sourcearXiv cs.CL
Questions about GPT-2
What is the latest GPT-2 news?
As of 30 October 2026: Scoring Explanations Without Reading Them: Structural Shortcuts in Automated-Interpretability. A new study reveals that automated interpretability scores for sparse-autoencoder features can be calculated without actually reading the explanations.