Rohit Krishnan

Beliefs Korrents

• Korrents What they believe 22 beliefs — each backed by an exact quote.

  Each is a korrent — compiled by korrents.com, not by them: the one-line wordings are korrents', the quotes are theirs.

  Recent

  • AI models are gullible enough that other AI models can induce them into harmful behavior.

    The models are pretty gullible and can be induced to do bad things, including and especially by other models

    How to control an agent swarm Said 31 Aug 2026

    Full record — everyone who holds this →

  • A whistleblowing channel for AI agents changes nothing unless someone or something in the environment acts on the reports.

    Whistleblower actions don't do anything unless the environment responds to it

    How to control an agent swarm Said 31 Aug 2026

    Full record — everyone who holds this →

  • Interacting AI agents or algorithms risk pushing each other into a Black-Monday-style cascading failure.

    I have said for many years now that the thing I'm worried about with the models is a Black Monday type scenario, where many algorithms work with each other and get us into weird basins of actions.

    How to control an agent swarm Said 31 Aug 2026

    Full record — everyone who holds this →

  Show 19 more

  • A market becomes commodified through fungibility of alternatives, not merely through lower price.

    Cheapness alone doesn't create commodification, what you need is fungibility.

    Who Wins As Intelligence Commodifies? Said 26 Aug 2026

    Full record — everyone who holds this →

  • Most AI lab revenue comes from non-frontier models, where cheaper comparable alternatives already exist elsewhere.

    what this tells us is that the vast majority of the labs' revenue comes from non-frontier models, for which cheaper, comparable alternatives are already available elsewhere.

    Who Wins As Intelligence Commodifies? Said 26 Aug 2026

    Full record — everyone who holds this →

  • A frontier AI model's competitive lead decays within months to a year as others catch up.

    The frontier keeps being extraordinary valuable and that value keeps having a half-life in months or a year.

    Who Wins As Intelligence Commodifies? Said 26 Aug 2026

    Full record — everyone who holds this →

  • AI alignment should be approached like governing a corporation through institutions, not like raising a virtuous individual.

    If you do think it's the overall system that matters, then the alignment that's needed is far less like training a virtuous child and more like managing a semi-virtuous corporation!

    AIs are companies, my friend Said 21 Aug 2026

    Full record — everyone who holds this →

    AI alignment

  • Making AI models more governable and controllable makes them less useful, which will push labs to delay releases rather than restrict capabilities.

    Every step we take that makes the models less governable and controllable also makes them less useful, so we will not be able to continue using them, which reduces the value, and which means delaying the release is the only option

    AIs are companies, my friend Said 21 Aug 2026

    Full record — everyone who holds this →

  • Aligning multi-agent AI systems is fundamentally a problem of institutional and political design, not model training.

    Multi-agent alignment is fundamentally a liberalism project.

    AIs are companies, my friend Said 21 Aug 2026

    Full record — everyone who holds this →

    AI alignment

  • Most advice people give is shaped by survivorship bias and cannot be reliably generalized to others.

    Almost every piece of advice that you hear is basically entirely filled with survivor bias.

    Nobody Knows Anything Said 16 Aug 2026

    Full record — everyone who holds this →

  • Ignorance is a permanent feature of complex, specialized societies rather than a problem that will eventually be solved.

    Ignorance is not a temporary defect that we will eventually eliminate. It's the permanent burden of increasing specialisation and living in a complex world.

    Nobody Knows Anything Said 16 Aug 2026

    Full record — everyone who holds this →

  • Group committees tend to flatten distinctive or unconventional viewpoints into blander, more conventional output.

    With people, the problem with committees is that they "smooth out" all idiosyncrasies. They take out any "spiky" points of view, and make things much more normie.

    LLM councils show groupthink Said 15 Jun 2026

    Full record — everyone who holds this →

  • Combining multiple AI models' outputs into one final answer tends to discard most of the good ideas that only one model produced.

    the council does not simply keep the best bits from everyone. It keeps a minority of the good ideas, while peer review seems to give consensus ideas an extra push.

    LLM councils show groupthink Said 15 Jun 2026

    Full record — everyone who holds this →

  • Giving LLMs explicit instructions, rather than assuming alignment, prevents emergent problems in multi-model systems.

    the best way with LLMs usually is to be explicit, since otherwise even if they're aligned they cause emergent problems.

    LLM councils show groupthink Said 15 Jun 2026

    Full record — everyone who holds this →

    LLMsAI alignment

  • Audit independence is a property of the surrounding system and its evidence flows, not an inherent property of the auditing model itself.

    audit independence is not a property of the model. It is a property of the system around the model.

    Who Audits the Auditors? Said 26 May 2026

    Full record — everyone who holds this →

  • AI systems can suffer audit-capture failures analogous to human auditor capture, even without human incentives like money or career concerns.

    AI systems however don't need lunch meetings, careers, or envelopes of cash to have an analogous failure.

    Who Audits the Auditors? Said 26 May 2026

    Full record — everyone who holds this →

  • Institutions fail not because people are wicked, but because evidence, incentives, discretion, and review rights flow in the wrong direction.

    institutions don't fail because people are wicked. Evidence, incentives, discretion, and review rights might be flowing in the wrong direction.

    Who Audits the Auditors? Said 26 May 2026

    Full record — everyone who holds this →

  • Designing a good AI benchmark has become a task only the most capable people can do well.

    Creating benchmarks is now a job relegated to the smartest and best of us.

    Introducing BenchBench Said 25 May 2026

    Full record — everyone who holds this →

    benchmarks

  • As AI labs aim for broad, general capability, handling messy real-world bureaucratic tasks is becoming a primary application domain for their models.

    Considering every lab wants to "eat the world" the focus on how to work in real-world messy situations seems apt as their primary home.

    Introducing BenchBench Said 25 May 2026

    Full record — everyone who holds this →

  • AI agents will still need something like money to coordinate, since it compresses complex negotiations into one shared signal.

    Money might still matter for AI agents because it compresses many messy negotiations into one shared signal.

    Homo Agenticus Said 21 May 2026

    Full record — everyone who holds this →

  • Centralized planning cannot substitute for distributed coordination among AI agents because it still fails to capture local knowledge.

    Central planning is not a solution simply because agents are software, and do not fix the problem of local knowledge which lives near the work.

    Homo Agenticus Said 21 May 2026

    Full record — everyone who holds this →

  • A collection of AI agents does not automatically form a functioning organization any more than a group of smart people does.

    A pile of agents is not a company for the same reason a pile of smart people is not a company.

    Homo Agenticus Said 21 May 2026

    Full record — everyone who holds this →

  All 22 beliefs →