The original text contained 1 image which was described by AI.

---

First published:

Source:

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

How four accounts have shifted the election probabilities narrative

As the US Presidential Election approaches, all eyes are on who will become the next president - Vice President Kamala Harris or former President Donald Trump. In early October, many analysts believed both candidates had a relatively equal chance of winning. However, a significant shift in betting markets occurred following Trump's rally in Butler, PA on October 6, with odds heavily moving in his favor.

On Polymarket, traders began noticing a sharp increase in bets on Trump. This shift was driven by a group of relatively new accounts, which have, as of today, deposited approximately $35 million USDC into Polymarket and deployed $33 million across election-related markets, always betting on Trump / Republican outcomes. This has sparked a lot of online speculation, leading to misconceptions. Below, I aim to clarify these points.

1. Can [...]

---

Outline:

(00:07) How four accounts have shifted the election probabilities narrative

(01:14) 1. Can a Group of Accounts Really Shift the Market This Much?

(02:09) 2. Is Trump Simply Gaining Momentum, and Did These Accounts Predict That?

(02:54) 3. Why Aren’t Sharp Traders Betting the Other Side to Normalize the Odds?

---

First published:

Source:

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

This is a linkpost for a new research paper from the Alignment Evaluations team at Anthropic and other researchers, introducing a new suite of evaluations of models' abilities to undermine measurement, oversight, and decision-making. Paper link.

Abstract:

Sufficiently capable models could subvert human oversight and decision-making in important contexts. For example, in the context of AI development, models could covertly sabotage efforts to evaluate their own dangerous capabilities, to monitor their behavior, or to make decisions about their deployment. We refer to this family of abilities as sabotage capabilities. We develop a set of related threat models and evaluations. These evaluations are designed to provide evidence that a given model, operating under a given set of mitigations, could not successfully sabotage a frontier model developer or other large organization's activities in any of these ways. We demonstrate these evaluations on Anthropic's Claude 3 Opus and Claude 3.5 Sonnet models. [...]

---

Outline:

(03:33) Human decision sabotage

(05:21) Code sabotage

(06:36) Sandbagging

(08:28) Undermining oversight

(10:00) Conclusions

---

First published:

Source:

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

It's been a busy season at the Nucleic Acid Observatory, and we have a lot to share since our last update. As always, If anything here is particularly interesting or if you’re working on similar problems, please reach out!

Wastewater Sequencing

We performed an initial analysis of untargeted sequencing data from aggregated airplane lavatory waste and municipal treatment plant influent that we collected and processed during our Fall 2023 partnership with CDC's Traveler-based Genomic Surveillance program and Ginkgo Biosecurity. We've now analyzed viral abundance and diversity in sequencing data across multiple sample types and wastewater-processing and sequencing protocols. Next steps include further investigating how protocol and sample-type affect specific viruses and bacteria, as well as understanding pathogen temporal dynamics seen in airport versus treatment-plant samples.

We have continued to work with Pardis Sabeti's Lab at the Broad Institute to develop an optimized protocol for preparing RNA sequencing libraries from [...]

---

Outline:

(00:19) Wastewater Sequencing

(02:33) Pooled Individual Sequencing

(03:33) Other Sampling Strategies

(04:12) Nucleic Acid Tracers

(04:47) Analysis of Sequencing Data

(07:27) Organizational Updates

The original text contained 1 image which was described by AI.

---

First published:

Source:

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Your mind is a prediction machine, constantly trying to anticipate the world around you and altering its forecasts based on new information. It's always doing this as a background process. But what would happen if you deliberately trained this skill? Could you get better at predicting your projects, your life, and the future?

Sadly, I don’t have a crystal ball, but I have found trying to see the future valuable. I can't be certain about what will happen, but by deliberately practising the skill of prediction, I can create feedback loops that allow me to better anticipate what the future will bring. This has indeed helped me become happier, better to interact with, and more impactful.

Concretely, making probabilistic predictions has helped me:

• Improve my productivity
• Make better decisions about my career and living situation
• Set and make progress towards my goals
• Prioritise work tasks
• Communicate project timelines [...]

---

Outline:

(01:34) Which questions are worth asking?

(10:53) How to track your predictions

The original text contained 20 images which were described by AI.

---

First published:

Source:

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Audio note: this article contains 149 uses of latex notation, so the narration may be difficult to follow. There's a link to the original text in the episode description.

ARC recently released our first empirical paper: Estimating the Probabilities of Rare Language Model Outputs. In this work, we construct a simple setting for low probability estimation — single-token argmax sampling in transformers — and use it to compare the performance of various estimation methods. ARC views low probability estimation as a potential technique for mitigating worst-case behavior from AI, including deceptive alignment; see our previous theory post on estimating tail risks in neural networks for further discussion.

Outline of this post:

---

Outline:

(01:32) Problem statement

(03:24) Estimation methods

(04:57) Importance sampling

(05:57) Independent Token Gradient Importance Sampling (ITGIS)

(07:16) Metropolis—Hastings Importance Sampling (MHIS)

(08:50) Activation extrapolation

(09:21) Quadratic Logit Decomposition (QLD)

(11:06) Gaussian Logit Difference

(11:41) Empirical results

(14:08) Discussion

(14:11) Motivation

(16:15) Importance sampling versus activation extrapolation

(17:15) Limitations

(18:54) Related work

(19:56) Future directions

The original text contained 7 footnotes which were omitted from this narration.

The original text contained 2 images which were described by AI.

---

First published:

Source:

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Of all the cognitive tools our ancestors left us, what's best? Society seems to think pretty highly of arithmetic. It's one of the first things we learn as children. So I think it's weird that only a tiny percentage of people seem to know how to actually use arithmetic. Or maybe even understand what arithmetic is for. Why?

I think the problem is the idea that arithmetic is about “calculating”. No! Arithmetic is a world-modeling technology. Arguably, it's the best world-modeling technology: It's simple, it's intuitive, and it applies to everything. It allows you to trespass into scientific domains where you don’t belong. It even has an amazing error-catching mechanism built in.

One hundred years ago, maybe it was important to learn long division. But the point of long division was to enable you to do world-modeling. Computers don’t make arithmetic obsolete. If anything, they do the opposite. Without [...]

---

Outline:

(01:17) Chimps

(06:18) Big blocks

(09:34) More big blocks

The original text contained 5 images which were described by AI.

---

First published:

Source:

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Dario Amodei is thinking about the potential. The result is a mostly good essay called Machines of Loving Grace, outlining what can be done with ‘powerful AI’ if we had years of what was otherwise relative normality to exploit it in several key domains, and we avoided negative outcomes and solved the control and alignment problems. As he notes, a lot of pretty great things would then be super doable.

Anthropic also offers us improvements to its Responsible Scaling Policy (RSP, or what SB 1047 called an SSP). Still much left to do, but a clear step forward there.

Daniel Kokotajlo and Dean Ball have teamed up on an op-ed for Time on the need for greater regulatory transparency. It's very good.

Also, it's worth checking out the Truth Terminal saga. It's not as scary as it might look at first glance, but it is definitely [...]

---

Outline:

(01:01) Language Models Offer Mundane Utility

(05:10) Language Models Don’t Offer Mundane Utility

(11:21) Deepfaketown and Botpocalypse Soon

(19:52) They Took Our Jobs

(20:33) Get Involved

(20:48) Introducing

(21:58) In Other AI News

(26:08) Truth Terminal High Weirdness

(34:54) Quiet Speculations

(44:45) Copyright Confrontation

(45:02) AI and the 2024 Presidential Election

(46:02) The Quest for Sane Regulations

(51:00) The Week in Audio

(53:40) Just Think of the Potential

(01:15:09) Reactions to Machines of Loving Grace

(01:25:32) Assuming the Can Opener

(01:32:32) Rhetorical Innovation

(01:35:41) Anthropic Updates its Responsible Scaling Policy (RSP/SSP)

(01:41:35) Aligning a Smarter Than Human Intelligence is Difficult

(01:43:36) The Lighter Side

The original text contained 11 images which were described by AI.

---

First published:

Source:

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

1. The amount of effort is not proportional to the result. One lucid dream (LD) can take hours or even dozens of hours of effort. On average, a practitioner experiences several dozen LDs in their lifetime before quitting. If they don't quit, they dedicate their entire life to it, day and night, trying endless techniques, practicing reality checks, etc.
2. All techniques are absolutely useless compared to the effectiveness of galantamine. The effectiveness of galantamine is directly proportional to its dosage. Thus, 16 mg will almost certainly send you into a powerful lucid dream or out-of-body experience. Therefore, all techniques like WILD are absolutely meaningless.
3. Non-lucid dreams are often more interesting than lucid ones. In a lucid dream, I find myself as my usual self, just in some virtual reality, which is only slightly more interesting than a computer game or watching a movie. But in a non-lucid dream, I [...]

---

First published:

Source:

Narrated by TYPE III AUDIO.

Here's a new paper on cancer developing in patients after receiving gene therapy with eli-cel, brand name Skysona.

What is eli-cel?

Stem cells are isolated and modified by using a lentiviral vector to add the ABCD1 gene. The stem cells in bone marrow are killed with chemotherapy, and replaced with the modified stem cells.

How often did cancer develop?

Hematologic cancer developed in 7 of 67 patients after the receipt of eli-cel

Yep, that's about what I expected. How long did it take?

myelodysplastic syndrome (MDS) with unilineage dysplasia in 2 patients at 14 and 26 months; MDS with excess blasts in 3 patients at 28, 42, and 92 months; MDS in 1 patient at 36 months; and acute myeloid leukemia (AML) in 1 patient at 57 months.

Makes sense. What does genetic analysis of the cancer cells indicate?

In the [...]

---

First published:

Source:

Narrated by TYPE III AUDIO.