:max_bytes(150000):strip_icc()/Bayes_Theorem-v2-6d6b2a2293c648ec99f16f78a3c80f09.png)
Bayesian probability
192
MENTIONS
24
EPISODES
21
PODCASTS
Search complete. 192 mentions across 24 episodes found for "Bayesian probability".
Sep 10, 2026
Type Three AudioNARRATOR
It is, in some sense, the structure that makes in-context learning both necessary and powerful.
Type Three AudioNARRATOR
Six argued that when pre-training data is a mixture of latent generative concepts, in-context learning should be thought of as the model implicitly performing Bayesian inference over those concepts.
Type Three AudioNARRATOR
In E7 and E5, we showed that NextToken pre-training forces this kind of inference even within a single concept.
Type Three AudioNARRATOR
Non-agonic data calls for both at once.
Type Three AudioNARRATOR
We will tackle that case momentarily, but here we have two memoryless components.
Type Three AudioNARRATOR
The only memory is a hidden which coin latent that is set once for each sequence generation and never changed.
Type Three AudioNARRATOR
This choice is hidden because you never see the initial selection directly, but Bayesian inference eventually resolves your uncertainty from observation statistics alone.
Type Three AudioNARRATOR
Together, the two coins can be thought of as one generator with two hidden states and no way to move between them.
David KippingHOST
A high probability of 0%, and equally so for 100%, but crucially, a valley of improbability between them.
David KippingHOST
This thought experiment provides some conceptual understanding, but Jaynes went far further than this and rigorously proved that this must be true, and he is now credited as one of the founders of what we now call objective Bayesianism in monstrosity.
David KippingHOST
Now, Jaynes never really thought about applying this to astrobiology, but for me, ever since I heard about this thought experiment, the connection was obvious.
David KippingHOST
But instead of those flasks of water, we would now have possible seats for life like star systems.
speaker_1HOST
They had to build an ultimate, perfectly objective machine, an algorithmic pipeline completely devoid of bias, hope, or the biological desire to see a pattern.
speaker_0HOST
They built a ruthless Bayesian detection pipeline, and I wanna dive into what Bayesian statistics actually means in this context because it is the ultimate buzzkill for pareidolia.
speaker_1HOST
It really is.
speaker_0HOST
Bayesian analysis fundamentally incorporates the principle of Occam's razor, doesn't it?
speaker_1HOST
It does, and it does so mathematically.
speaker_1HOST
Occam's razor states that the simplest explanation is usually the correct one.
speaker_1HOST
The bubble collision model introduces five new specific geometric parameters for every single collision it proposes, the radius, the depth, the center point, et cetera.
speaker_0HOST
It's a lot of extra math.
Jamie WestHOST
we must get back to business.
Jamie WestHOST
So last episode, we were discussing Bayesian probability.
Jamie WestHOST
And of course, it's a very straightforward topic and easy to understand.
Jamie WestHOST
And I can't imagine anybody having difficulty with it at all.
Robert WestHOST
when we draw conclusions, we should be doing so with at least implicitly, but some sense of the degree of confidence that we should have in whatever it is that we're saying.
Robert WestHOST
And that degree of confidence should be based on information that's coming in, and it should always be ready to be changed.
Robert WestHOST
So if you're someone who is very, very, very, very confident on the basis of your judgment what you've heard and beliefs about the situation that for example you know e-cigarettes are incredibly harmful then you should be willing to if new evidence comes in revise your judgment and similarly if you take the view that e-cigarettes are perfectly safe then you should be willing to revise your judgment as new evidence comes in now unfortunately humans are not good Bayesians in the sense that we tend to hold on very often to our priors, our pre-existing beliefs, rather more tenaciously than we should.
Robert WestHOST
That's a bias that we all have.
David KippingHOST
It's actually one of the papers that I'm most proud of, not only accounting for this covariance effect, but also employing some elegant statistical theorems to yield a result which is robust to choices that one might make during such an analysis.
David KippingHOST
And that's a technique known as objective Bayesianism.
David KippingHOST
But here's the punchline, it's that we can express the odds ratio between two hypotheses with this simple formula.
David KippingHOST
In the first hypothesis, life is, as Allwood would say, a quick and easy process.
Aaron ZalikowiczHOST
I'm really pleased to be here today with Bhaskar Roy from Yale University.
Aaron ZalikowiczHOST
Bhaskar is the author of a neurology article titled, Comparison of the Ethicacy and Safety of Myasthenia Gravis Treatments, a Bayesian Network Meta-analysis.
Aaron ZalikowiczHOST
In his paper, him and his authors provide a novel analysis comparing newly approved treatments for myasthenia gravis using publicly available data from published clinical trials.
Aaron ZalikowiczHOST
The authors use a Bayesian network meta-analysis to compare treatments across trials and mechanisms of action with a goal of beginning to answer the challenging question of how do we navigate the expanding number of myasthenia gravis treatments in clinical practice.
Aaron ZalikowiczHOST
Welcome Bhaskar to the Neurology Podcast.
Bhaskar RoyGUEST
Thank you, Arun.
Bhaskar RoyGUEST
However, when we are doing it, it is not as same as direct head-to-head comparison, but it can provide us some idea regarding the relative efficacy.
Bhaskar RoyGUEST
And we always have to be careful that this is just a meta-analysis and not a real-world evidence, and it is not equivalent to randomized clinical trials.
Kevin ScharpGUEST
all new every single thing i say is brand new he's never heard it he's never read it nothing ever and so
Miles K. DonahueHOST
that's exactly what you did
Kevin ScharpGUEST
Yeah, so the big move was to put Bayesianism into the theism-atheism debate.
Kevin ScharpGUEST
It's not an original move.
Kevin ScharpGUEST
Lots of people have done it.
Type Three AudioNARRATOR
Links Personal Website Google Scholar Profile DBLP Page MIT ECS Profile Vanessa Kosoy is Director of AI Research at Alta and Principal Research Scientist at Coral and a former research associate at the Machine Intelligence Research Institute.
Type Three AudioNARRATOR
She leads the learning theoretic agenda for AI alignment, seeking provable guarantees for safe agents and originated infra-Bayesianism with Alexander Apple, a mathematical framework generalizing Bayesian decision theory to handle non-realizability, logical uncertainty, and adversarial environments.
Type Three AudioNARRATOR
Her work spans reinforcement learning theory, decision theory, and the foundations of embedded agency.
Type Three AudioNARRATOR
Links Less Wrong Profile Alignment Forum Profile Google Scholar Profile Jan Kulveit is co-founder and principal investigator of the Alignment of Complex Systems ACS research group, part of the Centre for Theoretical Study at Charles University in Prague.
Robert WestHOST
Well, this is the thing.
Robert WestHOST
I like to claim things as 100%, but as a Bayesian, I would never do that.
Robert WestHOST
Because what you know as a Bayesian is once you...
Robert WestHOST
What is a Bayesian?
Jamie WestHOST
A
Robert WestHOST
Bayesian is someone who just follows Bayesian statistics and Bayes' theorem, which we're going to get into.
Robert WestHOST
We have talked about it before, but I'm sure if we did, people won't have either remembered it or remembered what it was that they understood about it if they did understand it.
Robert WestHOST
But the thing about being a Bayesian, and it sounds a bit weird, is that you must...
Laura GilbertGUEST
Um, but actually it doesn't have much to do with expertise in the topic.
Laura GilbertGUEST
Um, uh, the people that do really well at predicting the future are very good at what we call Bayesian updating.
Laura GilbertGUEST
Um-
Steph McGovernHOST
Right
14 more episodes mention Bayesian probability.
Create an account to see the whole feed, search across every transcript, and follow the entities you care about.