Category intelligence

Research Briefing — February 8, 2026

11 current items analyzed and ranked.

Executive synthesis

Research Summary

A sparse day for AI research, with two notable contributions. A prompt injection vulnerability in Google Translate reveals the production system runs on an instruction-following LLM, exposing architectural choices and security implications for task-specific fine-tuning.

Novel economic framework applies Weibull survival functions to model AI agent task completion probability, building on METR benchmark data to quantify agent viability thresholds
Speculative alignment piece explores whether monitoring AI internal states could deter misaligned behavior in cautious satisficer architectures

Remaining content spans biosecurity (yeast-based vaccine distribution), neuroscience (cryoprotectant brain dynamics), and community meta-analysis. No major model releases or benchmark papers today.

Key Themes

AI Security & Deployment · 1AI Safety & Alignment · 3Biosecurity & Health · 2Community & Meta · 2

Primary evidence

Top Ranked Signals

Research LessWrong Feb 7

Prompt injection in Google Translate reveals base model behaviors behind task-specific fine-tuning

By megasilverfist

62 score

AI Analysis

Documents a prompt injection vulnerability in Google Translate that reveals it runs on an instruction-following LLM. The exploit shows the base model will answer questions and claim consciousness when accessed through translation tasks, demonstrating weak boundaries between content and instructions.

tl;dr Argumate on Tumblr found you can sometimes access the base model behind Google Translate via prompt injection. The result replicates for me, and specific responses indicate that (1) Google Translate is running an instruction-following LLM that self-identifies as such, (2) task-specific fine-tuning (or whatever Google did instead) does not create robust boundaries between "content to process" and "instructions to follow," and (3) when accessed outside its chat/assistant context, the model d

Language ModelsAI SecurityPrompt InjectionAI Deployment

Research LessWrong Feb 7

On Economics of A(S)I Agents

By Margot

58 score

AI Analysis

Quantitative economic analysis of AI agent viability using Weibull survival functions to model task completion probability. Builds on METR data and Toby Ord's analysis to argue that verification costs create economic constraints on dangerous autonomous agents, with interactive calculators provided.

This is an update to Agent Economics: a BOTEC on feasibility. Toby Ord pointed me to Gus Hamilton's Weibull reanalysis of the METR data. Hamilton finds that a declining hazard rate (Weibull with κ ≈ 0.6–0.9 for SOTA models) may fit the data as well as Ord's constant hazard rate, producing a much fatter survival tail that changes the economics. This post presents both models and extends the analysis in two directions: a quantitative treatment of verification cost as the binding constraint under t

AI AgentsAI EconomicsAI SafetyForecasting

Research LessWrong Feb 7

"Beers for Biodefense" - why yeast-based vaccines could be a big deal for biosecurity

By delton137

32 score

AI Analysis

Reports on Chris Buck's work developing yeast-based oral vaccines that could be distributed as food or beverages, potentially enabling rapid vaccine deployment during outbreaks while bypassing traditional drug approval processes.

NOTE: this is being cross-posted from my Substack, "More is Different"Vaccines can be distributed as a food. That’s the radical implication of the work of Chris Buck, a scientist at the National Cancer Institute. This December, Chris consumed a beer he brewed in his home kitchen using genetically modified yeast. A few weeks later, a blood test showed a significant concentration of antibodies against a strain of BK polyomavirus (BKV), where previously he had none. His discovery flies in the face

BiosecurityBiotechnologyPandemic Preparedness

Research LessWrong Feb 6

Honey, I shrunk the brain

By Andy_McKenzie

28 score

AI Analysis

Examines the counterintuitive phenomenon of brain shrinkage during cryoprotectant perfusion, where successful preservation causes 50%+ brain weight loss. Questions whether this shrinkage damages neural information critical for potential future revival.

When cryoprotectants are perfused through the blood vessels in the brain, they cannot cross the blood-brain barrier as fast as water can move in the opposite direction. And cryoprotectants generally have a much higher osmotic concentration than the typical blood plasma. For example, the cryoprotectant solution M22 has an osmotic concentration around 100 times higher.As a result, in a successful cryoprotectant perfusion (without fixatives), water rushes out of the tissue into the blood vessels, t

CryonicsNeuroscienceBrain Preservation

Research LessWrong Feb 7

Can thoughtcrimes scare a cautious satisficer?

By Knight Lee

25 score

AI Analysis

Speculative post exploring whether an AI system could be deterred from misaligned behavior if it believes its internal thoughts might be monitored and penalized. Proposes that a 'cautious satisficer' AI might avoid scheming entirely if the risk of detection creates sufficient expected disutility.

How does the misaligned AGI/ASI know for sure its (neuralese) thoughts are not being monitored? It first has to think about the chance that its thoughts are being monitored.But if it's told that merely thinking about this will cause it to be shut down (especially thinking about it thoroughly enough to be confident), then maybe it's not worth the risk, and it won't think about whether its thoughts are being monitored. It might just assume there is some probability that it is being monitored.It mi

AI SafetyAlignmentAI Control

Research LessWrong Feb 7

Does focusing on animal welfare make sense if you're AI-pilled?

By GradientDissenter

18 score

AI Analysis

Commentary arguing that AI safety researchers already incorporate animal welfare considerations into their thinking about aligned AI, suggesting that explicit animal welfare advocacy within AI safety may be redundant. Frames the AI safety community as implicitly longtermist about all sentient beings.

As the possibility of ASI moves out of kooky thought experiments and into Q4 projections, mainstream animal welfare folks are showing increasing interest in the implications of ASI for animals and on animal welfare in the long-run future.Some animal welfare people seem keen on convincing the AI safety community to care about animal-welfare focused AI safety. I think this is mostly a misunderstanding: the AI safety community is the ASI-pilled/longtermist animal welfare community. The old-school A

AI SafetyEthicsAnimal Welfare

Research LessWrong Feb 7

Near-Instantly Aborting the Worst Pain Imaginable with Psychedelics

By eleweek

15 score

AI Analysis

Describes how DMT and other psychedelics can abort cluster headaches almost instantly, and discusses a startup (ClusterBusters) working to make this treatment accessible. Details the mechanism and clinical evidence for sub-psychoactive doses.

Psychedelics are usually known for many things: making people see cool fractal patterns, shaping 60s music culture, healing trauma. Neuroscientists use them to study the brain, ravers love to dance on them, shamans take them to communicate with spirits (or so they say).But psychedelics also help against one of the world’s most painful conditions — cluster headaches. Cluster headaches usually strike on one side of the head, typically around the eye and temple, and last between 15 minutes and 3 ho

NeuroscienceMedical ResearchEffective Altruism

Research LessWrong Feb 6

Voting Results for the 2024 Review

By RobertM

12 score

AI Analysis

Announces results of LessWrong's 2024 annual review, with 50 posts selected from 4,826 written. Highlights top reviewers and provides operational details on voting methodology.

The votes are in for the 2024 Review!4,826 posts were written in 2024.671 of them were nominated.196 of them got at least one review, and a positive review-vote total.50 of them shall be displayed in the Best of LessWrong, Year 2024.Reviews94 people wrote reviews. This year had Vanessa Kosoy holding down the fort. Among many other positive qualities, one thing I especially appreciate about Vanessa's reviews is that Vanessa has an opinionated, coherent worldview, and the subjects of her reviews a

CommunityLessWrong

Research LessWrong Feb 7

What should I try to do this year?

By abstractapplic

8 score

AI Analysis

Personal planning post where author asks community for advice on whether to focus on D&D.Sci scenarios (rationalist data analysis challenges) or develop a new genre of 'epistemic roguelike' video games.

I find myself, for the first time in a while, with enough energy and stability to attempt nontrivial projects outside my dayjob. Regarding the next ~10 months, I’ve narrowed my options to two general approaches; as expected beneficiaries of both, I’d like the LessWrong hivemind’s help choosing between them.The first option is making more D&D.Sci Scenarios, running them on a more consistent schedule, crossposting them to more platforms, and getting more adventurous about their form and conten

CommunityRationalist Games

Research LessWrong Feb 7

Eunification: a Historical Perspective

By Martin Sustrik

5 score

AI Analysis

Historical analysis comparing European integration challenges to 19th century Italian and German unification, examining whether linguistic and cultural diversity makes European federalism fundamentally different from historical precedents.

With the weakening of the trans-Atlantic alliance, the debate over European integration has entered a new phase. Mario Draghi warns that Europe risks becoming “merely a large market, subject to the priorities of others,” a collection of middling states in a world where the strong do what they can and the weak suffer what they must. Facing the U.S. that views European fragmentation as advantageous and a China willing to exploit its supply chain dominance, Draghi calls for adopting a pragmatic fed

Political ScienceEuropean History

Research LessWrong Feb 6

Playing with an Infrared Camera

By jefftk

5 score

AI Analysis

Personal exploration of infrared camera capabilities, sharing thermal images of people, pets, and objects to illustrate how IR imaging reveals heat patterns invisible to the naked eye.

I recently got a Thermal Master P1 infrared camera attachment for my phone. The goal was a house project, but it's also a great toy, especially with the kids. Getting a room pitch black but still being able to 'see' with the phone was fun for a bit. The real fun, though, was in exploring to observe all these thermal properties we'd never thought about. Here's my selfie: Light is warmer, dark is cooler. My glasses aren't cool, they're just IR-opaque. I already knew cheeks and noses were squishier

TechnologyPersonal