Wiredframe Radar logo

Category intelligence

Research Briefing — December 27, 2025

8 current items analyzed and ranked.

Executive synthesis

Research Summary

Today's most significant work centers on measuring opaque AI reasoning capabilities. Ryan Greenblatt's empirical research quantifies models' ability to solve math problems without chain-of-thought—a key proxy for detecting potentially dangerous hidden reasoning in future systems.

Note: Limited research volume today—only 8 items available, with top work focused on AI safety measurement and ethics rather than capabilities advances.

Key Themes

AI Capabilities · 1AI Safety · 5Mechanistic Interpretability · 2AI Welfare/Ethics · 1

Primary evidence

Top Ranked Signals

Research LessWrong Dec 26

Measuring no CoT math time horizon (single forward pass)

By ryan_greenblatt

72 score
AI Analysis
Ryan Greenblatt measures AI models' ability to solve math problems without chain-of-thought reasoning as a proxy for opaque reasoning capability—a key risk factor for scheming. Finds Opus 4.5 has a 3.5-minute no-CoT time horizon and that this capability has been doubling approximately every 9 months.
A key risk factor for scheming (and misalignment more generally) is opaque reasoning ability. One proxy for this is how good AIs are at solving math problems immediately without any chain-of-thought (CoT) (as in, in a single forward pass). I've measured this on a dataset of easy math problems and used this to estimate 50% reliability no-CoT time horizon using the same methodology introduced in Measuring AI Ability to Complete Long Tasks (the METR time horizon paper). Important caveat: To get hum
AI SafetyAI CapabilitiesAlignmentScheming RiskEvaluation
Research LessWrong Dec 26

Whole Brain Emulation as an Anchor for AI Welfare

By sturb

45 score
AI Analysis
Argues that Whole Brain Emulations can serve as an anchor point for AI welfare considerations since they would clearly deserve moral status under functionalism while being non-biological. Connects this framework to recent mechanistic interpretability findings showing LLMs have emotional representations with geometric structures matching human affect.
Epistemic status: Fairly confident in the framework, uncertain about object-level claims. Keen to receive pushback on the thought experiments.TL;DR: I argue that Whole Brain Emulations (WBEs) would clearly have moral patienthood, and that the relevant features are computational, not biological. Recent Mechanistic Interpretability (MI) work shows Large Language Models (LLMs) have emotional representations with geometric structure matching human affect. This doesn't prove LLMs deserve moral consid
AI WelfareAI EthicsMechanistic InterpretabilityConsciousness Studies
Research LessWrong Dec 26

The moral critic of the AI industry—a Q&A with Holly Elmore

By Mordechai Rorvig

32 score
AI Analysis
An interview with Holly Elmore discussing AI existential risks and her role as a critic of the AI industry. Explores the tension between corporations marketing AI as consumer technology while acknowledging existential risks and the growing ambiguity around AI safety concerns.
Since AI was first conceived of as a serious technology, some people wondered whether it might bring about the end of humanity. For some, this concern was simply logical. Human individuals have caused catastrophes throughout history, and powerful AI, which would not be bounded in the same way, might therefore pose even worse dangers.In recent times, as the capabilities of AI have grown larger, one might have thought that its existential risks would also have become more obvious in nature. And in
AI SafetyAI GovernanceExistential Risk
Research LessWrong Dec 26

Regression by Composition

By Anders_H

30 score
AI Analysis
This is a linkpost for the preprint “Regression by Composition”, by Daniel Farewell, Rhian Daniel, Mats Stensrud, and myself.The paper introduces Regression by Composition (RBC): a new, modular framew...
This is a linkpost for the preprint “Regression by Composition”, by Daniel Farewell, Rhian Daniel, Mats Stensrud, and myself.The paper introduces Regression by Composition (RBC): a new, modular framework for regression modelling built around the composition of group actions. The manuscript has been accepted as a discussion paper in JRSS-B and will be read to the Royal Statistical Society in London on March 24th, 2026.Background and motivationIn earlier posts on LessWrong, I have argued that an e
Research LessWrong Dec 26

Apply for Alignment Mentorship from TurnTrout and Alex Cloud

By TurnTrout

25 score
AI Analysis
Recruitment announcement for the MATS alignment mentorship program led by Alex Turner and Alex Cloud. Highlights successful alumni placements at Anthropic, MIRI, and Redwood Research, and past research outputs including pioneering work on steering vectors.
Through the MATS program, we (Alex Turner and Alex Cloud[1]) help alignment researchers grow from seeds into majestic trees. We have fun, consistently make real alignment progress, and help scholars tap into their latent abilities.MATS summer '26 applications are open until January 18th!Team Shard in MATS 6.0 during the summer of '24. From left: Evžen Wyitbul, Jacob Goldman-Wetzler, Alex Turner, Alex Cloud, and Joseph Miller.Many mentees now fill impactful roles.Lisa Thiergart (MATS 3.0) moved o
AI SafetyAlignmentCareer Development
Research LessWrong Dec 26

Burnout, depression, and AI safety: some concrete mental health strategies

By KatWoods

20 score
AI Analysis
A practical guide addressing mental health challenges in the AI safety community, arguing that burnout typically stems from hopelessness rather than overwork. Offers strategies for staying motivated when working on low-probability, high-stakes problems by focusing on process enjoyment and reframing expectations.
Cross-posted from my SubstackBurnout and depression in AI safety usually don’t happen because of overwork.From what I've seen, it usually comes from a lack of hope.Working on something you don’t think will work and if it doesn’t work, you’ll die? That's a recipe for misery.How do you fix AI safety hopelessness? First off, rationally assess the likelihood of your work actually helping with AI safety.If you rationally believe that it’s too unlikely to actually help with AI safety, well, then,
AI SafetyCommunity HealthCareer Advice
Research LessWrong Dec 26

Childhood and Education #16: Letting Kids Be Kids

By Zvi

18 score
AI Analysis
Introduces Regression by Composition (RBC), a new statistical framework for regression modeling using composition of group actions. The paper has been accepted as a discussion paper in JRSS-B and addresses limitations of standard GLM approaches for causal effect extrapolation.
The Revolution of Rising Requirements has many elements. The most onerous are the supervisory requirements on children. They have become, as Kelsey Piper recently documented, completely, utterly insane, to the point where: A third of people, both parents and non-parents, responded in a survey that it is not appropriate to leave a 13 year old at home for an hour or two, as opposed to when we used to be 11 year olds babysitting for other neighborhood kids. A third of people said in that same surve
Statistical MethodsCausal Inference
Research LessWrong Dec 26

How hard should I prioritize having kids?

By Recurrented

5 score
AI Analysis
A personal post seeking advice about family planning priorities, weighing various options including single parenthood and relationship choices. Discusses the author's personal circumstances and solicits perspectives from parents in the rationalist community.
I am really not sure how hard I should prioritize having kids in my life. I am posting this because I mostly would like to hear perspectives from other people who have kids. The current worlds that feel possible / close to me right now:not have kidshave kids with someone I like but don't love[1][2]have kids w a known donor and be a single parentmy ideas right now are:get good could try harder to meet people.[3] could change my own visibility so people meet me.could try to actually buil
PersonalCommunity Discussion