Tap a claim on the ring to put it at the centre.
← Prompt injection can be caught by watching the neurons that fire when…
17 connected korrents · 14 moments from 10 Dec 2015 to 9 Sept 2026.
Everything filed under prompt injection
prompt injection
Everything filed under neural networks
neural networks
Everything filed under AI alignment
AI alignment Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Read this korrent: Prompt injection can be caught by watching the neurons that fire when it happens, so a defence no longer depends on the model reporting the attack. Prompt injection can be caught by watching the neurons that fire when it happens, so a defence no longer depends on the model reporting the attack. Last stated 3 months ago
27 Jul 2026
BC
Boris Cherny — holds since 2026-07-27 — tap for who they are
Same subject: Auto-mode does not yet convincingly fix prompt-injection risk for coding agents. — tap to centre the map on it Auto-mode does not yet convincingly fix prompt-injection risk for coding agents. Last stated 2 months ago
8 Aug 2026
SW
Simon Willison — holds since 2026-08-08 — tap for who they are
Same subject: Prompt injection has no equivalent of parameterized queries: there is no reliable way to tell a language model which text is data and which is instructions. — tap to centre the map on it Prompt injection has no equivalent of parameterized queries: there is no reliable way to tell a language model which text is data and which is instructions. Last stated 7 months ago
19 Mar 2026
SW
Simon Willison — holds since 2026-03-19 — tap for who they are
Same subject: Prompt injection is an extremely hard attack to defend against, and hardly anyone integrating AI is talking about it. — tap to centre the map on it Prompt injection is an extremely hard attack to defend against, and hardly anyone integrating AI is talking about it. Last stated 2 years ago
22 Mar 2025
TH
ThePrimeagen — holds since 2025-03-22 — tap for who they are
Same subject: Prompt injection is no longer a live attack on the frontier model: it simply does not follow instructions it reads on the internet any more. — tap to centre the map on it Prompt injection is no longer a live attack on the frontier model: it simply does not follow instructions it reads on the internet any more. Last stated 3 months ago
27 Jul 2026
BC
Boris Cherny — holds since 2026-07-27 — tap for who they are
Same subject: Prompt injection is unsolved, but the latest models are post-trained to resist it, so it takes far more than 'ignore all previous instructions' now. — tap to centre the map on it Prompt injection is unsolved, but the latest models are post-trained to resist it, so it takes far more than 'ignore all previous instructions' now. Last stated 8 months ago
12 Feb 2026
PS
Peter Steinberger — holds since 2026-02-12 — tap for who they are
Same subject: If whole categories of reward hack are undetectable by humans, then training against the hacks we do catch teaches models to cheat only where we cannot see. — tap to centre the map on it If whole categories of reward hack are undetectable by humans, then training against the hacks we do catch teaches models to cheat only where we cannot see. Last stated 2 months ago
11 Aug 2026
RG
Ryan Greenblatt — holds since 2026-08-11 — tap for who they are
Same subject: The only guaranteed defence against a prompt-injection attack is to remove one of its three legs — private data, exposure to untrusted content, or a way to send data out. — tap to centre the map on it The only guaranteed defence against a prompt-injection attack is to remove one of its three legs — private data, exposure to untrusted content, or a way to send data out. Last stated 7 months ago
19 Mar 2026
SW
Simon Willison — holds since 2026-03-19 — tap for who they are
Same subject: The AI safety community's fixation on a model escaping its box has pushed the other significant AI risks out of the discussion, with little progress to show for it. — tap to centre the map on it The AI safety community's fixation on a model escaping its box has pushed the other significant AI risks out of the discussion, with little progress to show for it. Last stated 3 years ago
18 Mar 2024
SA
Sam Altman — holds since 2024-03-18 — tap for who they are
Same subject: A deep bidirectional model is strictly more powerful than a left-to-right model or a shallow concatenation of unidirectional models. — tap to centre the map on it A deep bidirectional model is strictly more powerful than a left-to-right model or a shallow concatenation of unidirectional models. Last stated 8 years ago
11 Oct 2018
KT
Kristina Toutanova — holds since 2018-10-11 — tap for who they are
MC
Ming-Wei Chang — holds since 2018-10-11 — tap for who they are
KL
Kenton Lee — holds since 2018-10-11 — tap for who they are
JD
Jacob Devlin — holds since 2018-10-11 — tap for who they are
Same subject: A layer should learn a residual with reference to its own input rather than an unreferenced function, which is what makes great depth trainable. — tap to centre the map on it A layer should learn a residual with reference to its own input rather than an unreferenced function, which is what makes great depth trainable. Last stated 11 years ago
10 Dec 2015
JS
Jian Sun — holds since 2015-12-10 — tap for who they are
KH
Kaiming He — holds since 2015-12-10 — tap for who they are
Same subject: A neural network's latent space is closer to an uncopyrightable syntax than to copyrightable expression. — tap to centre the map on it A neural network's latent space is closer to an uncopyrightable syntax than to copyrightable expression. Last stated 2 months ago
7 Sept 2026
KK
Kevin Kelly — holds since 2026-09-07 — tap for who they are
Same subject: A small enough winning ticket learns faster than the network it was cut out of, and ends up more accurate than it. — tap to centre the map on it A small enough winning ticket learns faster than the network it was cut out of, and ends up more accurate than it. Last stated 9 years ago
9 Mar 2018
MC
Michael Carbin — holds since 2018-03-09 — tap for who they are
JF
Jonathan Frankle — holds since 2018-03-09 — tap for who they are
Same subject: A winning ticket wins on its initial weights: the connections it keeps started at values that happen to make training work. — tap to centre the map on it A winning ticket wins on its initial weights: the connections it keeps started at values that happen to make training work. Last stated 9 years ago
9 Mar 2018
MC
Michael Carbin — holds since 2018-03-09 — tap for who they are
JF
Jonathan Frankle — holds since 2018-03-09 — tap for who they are
Same subject: Intelligence is the execution of small, simplicity-weighted programs found by search over smooth loss landscapes in overparameterized networks, approaching Bayes-optimal reasoning as scale increases. — tap to centre the map on it Intelligence is the execution of small, simplicity-weighted programs found by search over smooth loss landscapes in overparameterized networks, approaching Bayes-optimal reasoning as scale increases. Last stated 5 years ago
11 Jun 2021
GW
Gwern — holds since 2021-06-11 — tap for who they are
Same subject: A transduction model can rely entirely on self-attention for input and output representations without sequence-aligned RNNs or convolution. — tap to centre the map on it A transduction model can rely entirely on self-attention for input and output representations without sequence-aligned RNNs or convolution. Last stated 9 years ago
12 Jun 2017
IP
Illia Polosukhin — holds since 2017-06-12 — tap for who they are
JU
Jakob Uszkoreit — holds since 2017-06-12 — tap for who they are
NP
Niki Parmar — holds since 2017-06-12 — tap for who they are
AV
Ashish Vaswani — holds since 2017-06-12 — tap for who they are
NS
Noam Shazeer — holds since 2017-06-12 — tap for who they are
AG
Aidan N. Gomez — holds since 2017-06-12 — tap for who they are
+2
2 more
Same subject: Looped/recurrent-depth transformer architectures can improve model quality at a fixed compute budget when the model is large enough. — tap to centre the map on it Looped/recurrent-depth transformer architectures can improve model quality at a fixed compute budget when the model is large enough. Last stated a month ago
9 Sept 2026
SR
Sebastian Raschka — holds since 2026-09-09 — tap for who they are
Same subject: Self-attention can yield more interpretable models. — tap to centre the map on it Self-attention can yield more interpretable models. Last stated 9 years ago
12 Jun 2017
IP
Illia Polosukhin — holds since 2017-06-12 — tap for who they are
JU
Jakob Uszkoreit — holds since 2017-06-12 — tap for who they are
NP
Niki Parmar — holds since 2017-06-12 — tap for who they are
AV
Ashish Vaswani — holds since 2017-06-12 — tap for who they are
NS
Noam Shazeer — holds since 2017-06-12 — tap for who they are
AG
Aidan N. Gomez — holds since 2017-06-12 — tap for who they are
+2
2 more
At the centre
Prompt injection can be caught by watching the neurons that fire when it happens, so a defence no longer depends on the model reporting the attack.
Holds Read this korrent →
Auto-mode does not yet convincingly fix prompt-injection risk for coding agents.
Holds
Prompt injection has no equivalent of parameterized queries: there is no reliable way to tell a language model which text is data and which is instructions.
Holds
Prompt injection is an extremely hard attack to defend against, and hardly anyone integrating AI is talking about it.
Holds
Prompt injection is no longer a live attack on the frontier model: it simply does not follow instructions it reads on the internet any more.
Holds
Prompt injection is unsolved, but the latest models are post-trained to resist it, so it takes far more than 'ignore all previous instructions' now.
Holds
If whole categories of reward hack are undetectable by humans, then training against the hacks we do catch teaches models to cheat only where we cannot see.
Holds
The only guaranteed defence against a prompt-injection attack is to remove one of its three legs — private data, exposure to untrusted content, or a way to send data out.
Holds
The AI safety community's fixation on a model escaping its box has pushed the other significant AI risks out of the discussion, with little progress to show for it.
Holds
A deep bidirectional model is strictly more powerful than a left-to-right model or a shallow concatenation of unidirectional models.
Holds
A layer should learn a residual with reference to its own input rather than an unreferenced function, which is what makes great depth trainable.
Holds
A neural network's latent space is closer to an uncopyrightable syntax than to copyrightable expression.
Holds
A small enough winning ticket learns faster than the network it was cut out of, and ends up more accurate than it.
Holds
A winning ticket wins on its initial weights: the connections it keeps started at values that happen to make training work.
Holds
Intelligence is the execution of small, simplicity-weighted programs found by search over smooth loss landscapes in overparameterized networks, approaching Bayes-optimal reasoning as scale increases.
Holds
A transduction model can rely entirely on self-attention for input and output representations without sequence-aligned RNNs or convolution.
Holds
Looped/recurrent-depth transformer architectures can improve model quality at a fixed compute budget when the model is large enough.
Holds
Self-attention can yield more interpretable models.
Holds