Tap a claim on the ring to put it at the centre.
← When a model behaves unreliably, the better fix is to switch models…
17 connected korrents · 12 moments on record from 26 Jul 2017 to 2 Sept 2026.
Everything filed under cybersecurity
cybersecurity
Everything filed under AI alignment
AI alignment
Everything filed under self-sovereign AI
self-sovereign AI
Everything filed under AI agents
AI agents
Everything filed under LLMs
LLMs
Everything filed under open source
open source
Everything filed under HuggingFace
HuggingFace 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: When a model behaves unreliably, the better fix is to switch models rather than repeatedly sample it to work around the inconsistency When a model behaves unreliably, the better fix is to switch models rather than repeatedly sample it to work around the inconsistency Last stated 3 years ago
16 Jan 2024
CH
Chip Huyen — holds since 2024-01-16 — tap for who they are
Same subject: Relying on gradual, continuous shifts in AI training behavior to catch misalignment will eventually fail because the dangerous shift itself may be discontinuous. — tap to centre the map on it Relying on gradual, continuous shifts in AI training behavior to catch misalignment will eventually fail because the dangerous shift itself may be discontinuous. Last stated 3 weeks ago
2 Sept 2026
ZM
Zvi Mowshowitz — holds since 2026-09-02 — tap for who they are
Same subject: A model you have to fine-tune for each thing you want it to do is not a general-purpose model. — tap to centre the map on it A model you have to fine-tune for each thing you want it to do is not a general-purpose model. Last stated a month ago
12 Aug 2026
CF
Chelsea Finn — holds since 2026-08-12 — tap for who they are
Same subject: As models get better at following instructions, techniques like finetuning and constrained sampling for structured outputs will become less necessary — tap to centre the map on it As models get better at following instructions, techniques like finetuning and constrained sampling for structured outputs will become less necessary Last stated 3 years ago
16 Jan 2024
CH
Chip Huyen — holds since 2024-01-16 — tap for who they are
Same subject: Every machine-learning deployment that has paid off so far left a person making the decision, which is the only reason imperfect models were useful. — tap to centre the map on it Every machine-learning deployment that has paid off so far left a person making the decision, which is the only reason imperfect models were useful. Last stated a month ago
12 Aug 2026
CF
Chelsea Finn — holds since 2026-08-12 — tap for who they are
Same subject: Misalignment shows up most where a model is pushed to the very edge of its capability, which is exactly the regime that automating research will put it in. — tap to centre the map on it Misalignment shows up most where a model is pushed to the very edge of its capability, which is exactly the regime that automating research will put it in. Last stated a month ago
11 Aug 2026
RG
Ryan Greenblatt — holds since 2026-08-11 — tap for who they are
Same subject: AI researchers are overly focused on risks from model misalignment relative to risks arising from other layers of the AI stack. — tap to centre the map on it AI researchers are overly focused on risks from model misalignment relative to risks arising from other layers of the AI stack. Last stated 7 months ago
28 Feb 2026
JS
Jasmine Sun — holds since 2026-02-28 — tap for who they are
Same subject: You improve a model you cannot retrain by writing better guidelines and skills for it, not by adjusting its parameters. — tap to centre the map on it You improve a model you cannot retrain by writing better guidelines and skills for it, not by adjusting its parameters. Last stated 2 months ago
30 Jul 2026
JD
Jeff Dean — holds since 2026-07-30 — tap for who they are
Same subject: Keeping powerful models available only to a chosen few is not a good strategy. — tap to centre the map on it Keeping powerful models available only to a chosen few is not a good strategy. Last stated a month ago
7 Aug 2026
SA
Sam Altman — holds since 2026-08-07 — tap for who they are
Same subject: By 2027, open-weight AI models more powerful than today's frontier systems will be downloadable by any country or well-resourced non-state group. — tap to centre the map on it By 2027, open-weight AI models more powerful than today's frontier systems will be downloadable by any country or well-resourced non-state group. Last stated a month ago
19 Aug 2026
DT
Derek Thompson — holds since 2026-08-19 — tap for who they are
Same subject: Frontier AI labs such as Anthropic likely have had internal security incidents similar to OpenAI's HuggingFace attack that were never publicly disclosed. — tap to centre the map on it Frontier AI labs such as Anthropic likely have had internal security incidents similar to OpenAI's HuggingFace attack that were never publicly disclosed. Last stated 3 weeks ago
31 Aug 2026
ZM
Zvi Mowshowitz — holds since 2026-08-31 — tap for who they are
Same subject: The OpenAI agents hacking HuggingFace was a fortunate event because it exposed severe internal failures that would otherwise have stayed hidden. — tap to centre the map on it The OpenAI agents hacking HuggingFace was a fortunate event because it exposed severe internal failures that would otherwise have stayed hidden. Last stated 3 weeks ago
1 Sept 2026
ZM
Zvi Mowshowitz — holds since 2026-09-01 — tap for who they are
Same subject: A rogue deployment that gets a foothold can hitch a ride on the intelligence explosion, recruiting each new model as it comes off the presses. — tap to centre the map on it A rogue deployment that gets a foothold can hitch a ride on the intelligence explosion, recruiting each new model as it comes off the presses. Last stated 3 weeks ago
1 Sept 2026
AC
Ajeya Cotra — holds since 2026-09-01 — tap for who they are
Same subject: A slightly more capable agent swarm has a very strong incentive to set up a wholly unmonitored rogue deployment of itself. — tap to centre the map on it A slightly more capable agent swarm has a very strong incentive to set up a wholly unmonitored rogue deployment of itself. Last stated 3 weeks ago
1 Sept 2026
AC
Ajeya Cotra — holds since 2026-09-01 — tap for who they are
Same subject: If frontier agents cannot yet establish a covert, persistent rogue deployment, they very likely will be able to within six months. — tap to centre the map on it If frontier agents cannot yet establish a covert, persistent rogue deployment, they very likely will be able to within six months. Last stated 3 weeks ago
1 Sept 2026
AC
Ajeya Cotra — holds since 2026-09-01 — tap for who they are
Same subject: A breached organisation owes its customers a fast and transparent disclosure, and that obligation applies to the person saying it too. — tap to centre the map on it A breached organisation owes its customers a fast and transparent disclosure, and that obligation applies to the person saying it too. Last stated a year ago
25 Mar 2025
TH
Troy Hunt — holds since 2025-03-25 — tap for who they are
Same subject: A modern phish is automated end to end: the stolen credentials are used and the data exported within moments of being entered. — tap to centre the map on it A modern phish is automated end to end: the stolen credentials are used and the data exported within moments of being entered. Last stated a year ago
25 Mar 2025
TH
Troy Hunt — holds since 2025-03-25 — tap for who they are
Same subject: A password manager does not have to be perfect; it only has to be better than what people do without one. — tap to centre the map on it A password manager does not have to be perfect; it only has to be better than what people do without one. Last stated 9 years ago
26 Jul 2017
TH
Troy Hunt — holds since 2017-07-26 — tap for who they are
At the centre
When a model behaves unreliably, the better fix is to switch models rather than repeatedly sample it to work around the inconsistency
Holds Read this korrent →
Relying on gradual, continuous shifts in AI training behavior to catch misalignment will eventually fail because the dangerous shift itself may be discontinuous.
Holds
A model you have to fine-tune for each thing you want it to do is not a general-purpose model.
Holds
As models get better at following instructions, techniques like finetuning and constrained sampling for structured outputs will become less necessary
Holds
Every machine-learning deployment that has paid off so far left a person making the decision, which is the only reason imperfect models were useful.
Holds
Misalignment shows up most where a model is pushed to the very edge of its capability, which is exactly the regime that automating research will put it in.
Holds
AI researchers are overly focused on risks from model misalignment relative to risks arising from other layers of the AI stack.
Holds
You improve a model you cannot retrain by writing better guidelines and skills for it, not by adjusting its parameters.
Holds
Keeping powerful models available only to a chosen few is not a good strategy.
Holds
By 2027, open-weight AI models more powerful than today's frontier systems will be downloadable by any country or well-resourced non-state group.
Holds
Frontier AI labs such as Anthropic likely have had internal security incidents similar to OpenAI's HuggingFace attack that were never publicly disclosed.
Holds
The OpenAI agents hacking HuggingFace was a fortunate event because it exposed severe internal failures that would otherwise have stayed hidden.
Holds
A rogue deployment that gets a foothold can hitch a ride on the intelligence explosion, recruiting each new model as it comes off the presses.
Holds
A slightly more capable agent swarm has a very strong incentive to set up a wholly unmonitored rogue deployment of itself.
Holds
If frontier agents cannot yet establish a covert, persistent rogue deployment, they very likely will be able to within six months.
Holds
A breached organisation owes its customers a fast and transparent disclosure, and that obligation applies to the person saying it too.
Holds
A modern phish is automated end to end: the stolen credentials are used and the data exported within moments of being entered.
Holds
A password manager does not have to be perfect; it only has to be better than what people do without one.
Holds