Loading…
Hosted by Redwood Research · 🇺🇸 US · EN-GB · 117 episodes
Established thought leaders with verified media credentials.
Narrations of Redwood Research blog posts.Redwood Research is a research nonprofit based in Berkeley. We investigate risks posed by the development of powerful artificial intelligence and techniques for mitigating those risks.
Redwood Research hosts Redwood Research Blog, a technology show with 117 episodes published.
Subtitle: We need more details. The OpenAI AI attack on Hugging Face wasn’t the first loss of control incident at OpenAI, Reuters recently reported, and perhaps not even the most concerning. In one case, an agent left no
Subtitle: And what the incident can’t tell us about alignment. The most common dismissive response to OpenAI's hack of Hugging Face's servers is that the models were simply attempting to follow the instructions they were
We talk about the OpenAI–Hugging Face incident, where an OpenAI model — in the middle of a cyber evaluation — broke out of its sandbox and autonomously hacked Hugging Face. We discuss: What we actually
Subtitle: Yes, but less than had they been schemers. OpenAI models recently broke through a series of security boundaries and into Hugging Face servers in order to cheat on a cyber eval. A lot of people thought it was sc
We recently ran a strategy fellowship through Astra. As part of this, we ran a reading group for our fellows on some of the topics that we think are important for thinking about AI futurism (key dynamics in AI developmen
Subtitle: If it transfers misalignment, we might get a misaligned model that's easier to incriminate. If it doesn’t, we might get a capable benign replacement model. Suppose we have a dangerous misaligned AI that can foo
Subtitle: Models' no-CoT time horizon has doubled roughly every year. by Francis Rhys Ward, Dewi Gould, Anders Cairns Woodruff et al. (see full author list at the end) PAPER LINK About a year ago, METR showed that the le
Subtitle: When is "increasing safety budget" a useful concept? I often use what I’ll call the “safety-usefulness tradeoff model”, which is: developers face a tradeoff between “safety” and “usefulness” of an AI deployment
We’ve just released a new paper: Retrying vs Resampling in AI Control. We revisit the resampling protocols introduced in Ctrl-Z with an up-to-date setting and much stronger models, and compare them against “retrying” pro
We’d like to develop training techniques that work when applied to future misaligned AI systems. One strategy for studying proposed techniques is to test them on model organisms. However, model organisms built with commo
Subtitle: Full automation likely yields a one-time speed-up and higher returns from compute. This is a somewhat technical note. By “software-only singularity”, I mean that, after full automation of AI R&D, progress g
Suppose we have a dangerous misaligned AI that can fool alignment audits, and distill it into a student model. Two things can happen: Misalignment fails to transfer to the student. If so, we get a fairly capable benign m
Subtitle: Deployment-time spread is the most plausible near-term route to consistent adversarial misalignment. Risk reports commonly use pre-deployment alignment assessments to measure misalignment risk from an internall
Subtitle: My median guess: it's as good as a crystal ball that sees 2.5 months into the future. This post was drafted by Buck, and substantially edited by Anders. “I” refers to Buck. Thanks to Alex Mallen for comments. P
Last week, OpenAI staff shared an early draft of Investigating the consequences of accidentally grading CoT during RL with Redwood Research staff. To start with, I appreciate them publishing this post. I think it is valu
Subtitle: Fitness-seeking is increasingly what misalignment looks like in practice—how should we respond? Current AIs routinely take unintended actions to score well on tasks: hardcoding test cases, training on the test
One of the main hopes for AI safety is using AIs to automate AI safety research. However, if models are misaligned, then they may sabotage the safety research. For example, misaligned AIs may try to: Perform sloppy resea
Subtitle: Eliciting long-term forecasts from myopic fitness-seekers. We’d like to use powerful AIs to answer questions that may take a long time to resolve. But if a model only cares about performing well in ways that ar
Subtitle: A controlled reward-seeking motivation could make AI safer and more useful. It's plausible that flawed RL processes will select for misaligned AI motivations.1 Some misaligned motivations are much more dangerou
Subtitle: Third-party experts should assess defenses against tampering and theft — and publish high-level findings. AI companies should get third-party security experts to assess (and possibly also red-team/pen-test) the
Sponsor detection runs nightly. Check back soon.
No public pitch examples yet for this show.
Generate your own personalised pitchBased on semantic analysis of episode topics and host coverage, this show is a strong guest fit for executives in:
Industry fit is computed by PitchCentric using vector embeddings of the show's episode catalog.
Shows with the most semantically similar episode content. Pitch one, pitch all; producers cluster.
Redwood Research Blog has a verified contact on file. Create a free PitchCentric account to access it and generate a personalised pitch in seconds. Research at least 3 recent episodes first and lead with a specific angle that serves their technology audience.
Redwood Research Blog is hosted by Redwood Research. The show is categorised under technology (society) and has published 117 episodes.
Redwood Research Blog has published 117 episodes.
Redwood Research Blog regularly covers technology, society, culture. It sits in the technology category, with a society focus.
Redwood Research Blog is accessible for guests with genuine technology expertise. A personalised, episode-aware pitch will still outperform a generic one every time.
Redwood Research Blog hasn't explicitly signalled guest openness in recent episodes. That doesn't rule out pitching. your hook just needs to be especially compelling and relevant to their recent content.
Episodes of Redwood Research Blog average 21 minutes. a focused format where a clear narrative arc and tight preparation matter most.
Our data rates Redwood Research Blog's guest bar at 80/100 (Premium tier). Established thought leaders with verified media credentials. Sign in to PitchCentric to see how your own Pod Score compares against this show.
Methodology. Booking Probability™ blends Listen Score, 30-day Virality, open-to-guests detection, and Apple ratings. Data refreshed every 60 minutes. Listen Score and Booking Probability are calculated by PitchCentric. Last enriched 11 days ago.