text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Kamelo: A Rule-Based Constructed Language for Universal, Logical Communication
**Introduction**
Natural languages are messy, ambiguous, and often inefficient for transmitting structured ideas. Kamelo is a proposal for a constructed language designed to address these issues by building words from logical, compositio... | https://www.lesswrong.com/posts/BFfCiTsiziXcnQLLJ/kamelo-a-rule-based-constructed-language-for-universal |
# GPT-4.1 Is a Mini Upgrade
Yesterday’s news alert, nevertheless: The verdict is in. GPT-4.1-Mini in particular is an excellent practical model, offering strong performance at a good price. [The full GPT-4.1](https://openai.com/index/gpt-4-1/) is an upgrade to OpenAI’s more expensive API offerings, it is modestly bett... | https://www.lesswrong.com/posts/t5dMAHrmdXqNaNB6v/gpt-4-1-is-a-mini-upgrade |
# GPT-4.5 is Cognitive Empathy, Sonnet 3.5 is Affective Empathy
This was the thing my friend told me I should make an account here to post, I hope it is appropriate.
Introduction
============
The thrust of this hypothesis is that Anthropic's research direction favors the production of affective empathy while OpenAI'... | https://www.lesswrong.com/posts/5EP9nvN6TNzhy5itN/gpt-4-5-is-cognitive-empathy-sonnet-3-5-is-affective-empathy |
# Prodromes and Biomarkers in Chronic Disease
[
](https://substackcdn.com/image/fetch/f_auto,q_a... | https://www.lesswrong.com/posts/H9wWmuMQ2xzGsmHaz/prodromes-and-biomarkers-in-chronic-disease |
# ALLFED emergency appeal: Help us raise $800,000 to avoid cutting half of programs
**SUMMARY: **
ALLFED is making an emergency appeal here due to a serious funding shortfall. Without new support, ALLFED will be forced to cut half our budget in the coming months, drastically reducing our capacity to help build global... | https://www.lesswrong.com/posts/HL7zzZhCR59CCntpB/allfed-emergency-appeal-help-us-raise-usd800-000-to-avoid |
# Understanding and overcoming AGI apathy
*Crossposting from my* [*substack*](https://dhruvsumathi.substack.com/p/waves-not-mountains-overcoming-apathy?r=b10d1)*.*
**Note for LW:** This post is about the apathy and nonchalance I've been seeing in many peers regarding AGI. There seems to be massive cognitive dissonanc... | https://www.lesswrong.com/posts/PQ3XDmDFwv7WpvNoF/waves-not-mountains |
# AI #112: Release the Everything
OpenAI has upgraded its entire suite of models. By all reports, they are back in the game for more than images. GPT-4.1 and especially GPT-4.1-mini are their new API non-reasoning models. [All reports are](https://thezvi.substack.com/p/gpt-41-is-a-mini-upgrade) that GPT-4.1-mini espec... | https://www.lesswrong.com/posts/nycc4QxQAMkzmXmfz/ai-112-release-the-everything |
# Host Keys and SSHing to EC2
I do a lot of work on EC2, where I [ssh into a few instances I use for specific purposes](https://www.jefftk.com/p/ec2-scripts). Each time I did this I'd get a prompt like:
$ ssh_ec2nf
The authenticity of host 'ec2-54-224-39-217.compute-1.amazonaws.com
(54.224.39.217)' can't be establish... | https://www.lesswrong.com/posts/vioAQaPASNNRNE7nK/host-keys-and-sshing-to-ec2 |
# Announcing Progress Conference 2025
Last fall the Roots of Progress Institute hosted the first annual Progress Conference. 200 people excited about human progress gathered for two days in Berkeley, California, to share ideas in deep conversation, catalyze new projects, and get energized and inspired. Several attende... | https://www.lesswrong.com/posts/YXczyDZNCzrnRAirK/announcing-progress-conference-2025 |
# The Russell Conjugation Illuminator
Bertrand Russell noted how people often describe the same factual behavior using emotionally opposite language depending on perspective — e.g. I am firm, you are obstinate, he is pigheaded. This framing tactic is now called a Russell Conjugation, and once you start noticing them, ... | https://www.lesswrong.com/posts/GjqwHYBrGhCjP7gkH/the-russell-conjugation-illuminator |
# Conditional Forecasting as Model Parameterization
*This work was done jointly with Rajashree Agrawal. We'd like to thank the Epoch AI summer mentorship project, led with Forecasting Research Institute, for introducing us. Thanks to Josh Rosenberg, Tegan McCaslin, Avital Morris and others at FRI, as well as Jaime Sev... | https://www.lesswrong.com/posts/XJkLnRcBfvaQkCs2g/conditional-forecasting-as-model-parameterization |
# Three Months In, Evaluating Three Rationalist Cases for Trump
I recall seeing three “rationalist” cases for Trump:
1. **Richard Ngo** [**on Twitter**](https://x.com/RichardMCNgo/status/1852754804973375588) and elsewhere focused on the realignment of elite coalitions, observing that “most elite institutions have b... | https://www.lesswrong.com/posts/tKhbDBkstMJuBv7jg/three-months-in-evaluating-three-rationalist-cases-for-trump |
# Training AGI in Secret would be Unsafe and Unethical
*Subtitle: Bad for loss of control risks, bad for concentration of power risks*
-------------------------------------------------------------------------------
*I’ve had this sitting in my drafts for the last year. I wish I’d been able to release it sooner, but o... | https://www.lesswrong.com/posts/FGqfdJmB8MSH5LKGc/training-agi-in-secret-would-be-unsafe-and-unethical-1 |
# British and American Connotations
As an American who works with some people who speak British English, the language differences are usually not a problem. Most words mean the same thing, and those that don't are usually concrete enough not to cause confusion (ex: lift, flat, chips). The tricky ones, though, are the ... | https://www.lesswrong.com/posts/3CW5t9LSLy7gZZkB9/british-and-american-connotations |
# Handling schemers if shutdown is not an option
*What if getting strong evidence of scheming isn't the end of your scheming problems, but merely the middle?*
In most of our research and writing on AI control, we’ve emphasized the following situation:
* The AI developer is deploying a model that they think might b... | https://www.lesswrong.com/posts/XxjScx4niRLWTfuD5/handling-schemers-if-shutdown-is-not-an-option |
# Scaffolding Skills
Epistemic status: [Exploratory](https://www.lesswrong.com/posts/Hrm59GdN2yDPWbtrd/feature-idea-epistemic-status)
A scaffold is a lightweight, temporary construction whose point is to make working on other buildings easier. Maybe you could live and work on a scaffold, but why would you? The point ... | https://www.lesswrong.com/posts/68iix7PFHPADhdQYF/scaffolding-skills |
# Inside OpenAI's Controversial Plan to Abandon its Nonprofit Roots
*This is the full text of a post from *[*Obsolete*](https://garrisonlovely.substack.com/)*, a Substack that I write *[*about*](https://garrisonlovely.substack.com/about) *the intersection of capitalism, geopolitics, and artificial intelligence. I’m a*... | https://www.lesswrong.com/posts/SHbEEFHqTuoveZcLr/inside-openai-s-controversial-plan-to-abandon-its-nonprofit |
# Alignment Does Not Need to Be Opaque! An Introduction to Feature Steering with Reinforcement Learning
Reinforcement Learning (RL) has become the driving force behind modern artificial intelligence, as evidenced by [rewarding models from human feedback](https://arxiv.org/abs/2203.02155) or powerful new systems like [... | https://www.lesswrong.com/posts/GgCgRqKkkqsrxZFu5/alignment-does-not-need-to-be-opaque-an-introduction-to |
# What Makes an AI Startup "Net Positive" for Safety?
In light of the recent news from [Mechanize/Epoch](https://x.com/MechanizeWork/status/1912904151874625928) and the [community discussion it sparked](https://www.lesswrong.com/posts/jXjeYYPXipAtA2zmj/jacquesthibs-s-shortform?commentId=ZXdbw9c4PavDq6MTG), I'd like to... | https://www.lesswrong.com/posts/o3sEHE8cqQ5hcqgkG/what-makes-an-ai-startup-net-positive-for-safety |
# AI Control Methods Literature Review
AI Control is a subfield of AI Safety research focused on ensuring the safety of powerful AI systems, particularly under the assumption that they might be misaligned or could become intentionally subversive ([A Brief Explanation of AI Control](https://www.lesswrong.com/posts/7e47... | https://www.lesswrong.com/posts/3PBvKHB2EmCujet3j/ai-control-methods-literature-review |
# o3 Will Use Its Tools For You
[OpenAI has finally introduced us to the full o3](https://openai.com/index/introducing-o3-and-o4-mini/) along with o4-mini.
> [Greg Brockman (OpenAI)](https://x.com/gdb/status/1912561761489940639): Just released o3 and o4-mini! These models feel incredibly smart. We’ve heard from top s... | https://www.lesswrong.com/posts/u58AyZziQRAcbhTxd/o3-will-use-its-tools-for-you |
# Measuring Beliefs of Language Models During Chain-of-Thought Reasoning
*Based on research performed as a* [*PIBBSS*](https://pibbss.ai/fellowship/) *Fellow with Tomáš Gavenčiak as well as work supported by EA Funds and Open Philanthropy.*
*tl;dr: I'm investigating whether LLMs track and update beliefs during chain-... | https://www.lesswrong.com/posts/a86uAnPykqNtmEbDH/measuring-beliefs-of-language-models-during-chain-of-thought-1 |
# AI Advances and Detection Strategy
*Cross-posted from my [NAO Notebook](https://data.securebio.org/jefftk-notebook/ai-and-detection).*
*This is an internal strategy note I wrote in November 2024 that I'm making public with some light editing.*
In my work at the NAO I've been thinking about what I expect to see as ... | https://www.lesswrong.com/posts/fFShACvybctrQBkSh/ai-advances-and-detection-strategy |
# An Introduction to SAEs and their Variants for Mech Interp
*I aim to cover a lot of ground, but only briefly. Click through the links for more detailed explanations.*
*Several of the techniques here see general use outside of interpretability. My definitions/references lean towards how they specifically apply to Me... | https://www.lesswrong.com/posts/vvvJNh8gNE2ScEtQ3/an-introduction-to-saes-and-their-variants-for-mech-interp |
# Why Should I Assume CCP AGI is Worse Than USG AGI?
Though, given my doomerism, I think the natsec framing of the AGI race is likely wrongheaded, let me accept the Dario/Leopold/Altman frame that AGI will be aligned with the national interest of a great power. These people seem to take as an axiom that a USG AGI will... | https://www.lesswrong.com/posts/MKS4tJqLWmRXgXzgY/why-should-i-assume-ccp-agi-is-worse-than-usg-agi-1 |
# How Close We Are to a Complete List of Imprinted Genes
*This post summarizes some of the research I have been doing for
Bootstrap Bio AKA [kman](https://www.lesswrong.com/users/kman) and
[Genesmith](https://www.lesswrong.com/users/genesmith)*
# TL;DR
- **We currently don’t have a list of all imprinted genes that... | https://www.lesswrong.com/posts/FX978FWaNQXeyat9S/how-close-we-are-to-a-complete-list-of-imprinted-genes |
# Impact, agency, and taste
I’ve been thinking recently about what sets apart the people who’ve done the best work at Anthropic.
You might think that the main thing that makes people really effective at research or engineering is technical ability, and among the general population that’s true. Among people hired at A... | https://www.lesswrong.com/posts/DiJT4qJivkjrGPFi8/impact-agency-and-taste |
# Is Gemini now better than Claude at Pokémon?
**Background:** With the release of Claude 3.7 Sonnet, Anthropic promoted [a new benchmark: beating Pokémon](https://www.anthropic.com/research/visible-extended-thinking). Now, [Google claims](https://x.com/sundarpichai/status/1913464625393524967) Gemini 2.5 Pro has subst... | https://www.lesswrong.com/posts/7mqp8uRnnPdbBzJZE/is-gemini-now-better-than-claude-at-pokemon |
# Interesting ACX 2024 Book Review Entries
I have finally got around to updating [codex cc](https://codexcc.neocities.org/) with all the book reviews from 2024!
Here are some interesting books/reviews that I’ve found that did not make the shortlist.
### **A Thousand Ways to Please A Husband (1917) (**[**Goodreads**]... | https://www.lesswrong.com/posts/XEkfRff2P4mtYCydQ/interesting-acx-2024-book-review-entries |
# Spending on Ourselves
The Effective Altruism community has encouraged a range of different approaches to doing good over time. Initially there was more focus on [frugality](https://www.jefftk.com/p/living-frugally-so-we-can-give-away-more) as a way to increase how much you could donate, which was mostly supplanted b... | https://www.lesswrong.com/posts/pGYesWdk57uv5Aok3/spending-on-ourselves |
# How to end credentialism
[The current University System is bad](https://www.cremieux.xyz/p/the-end-of-credentialism). Very bad. Half of young adults spend 3 to 4 years learning nothing of particular value so that they can get a piece of paper that gives them a leg up when applying for a job. They can't stop, because... | https://www.lesswrong.com/posts/dzCFDS98J32exXAqm/how-to-end-credentialism |
# AI 2027 is a Bet Against Amdahl's Law
EDIT: I've written a [followup post](https://www.lesswrong.com/posts/FFKnWk2MGJmyQWEsd/updates-from-comments-on-ai-2027-is-a-bet-against-amdahl-s), summarizing and responding to the key themes raised in the comments.
[AI 2027](https://www.lesswrong.com/posts/TpSFoqoG2M5MAAesg/a... | https://www.lesswrong.com/posts/bfHDoWLnBH9xR3YAK/ai-2027-is-a-bet-against-amdahl-s-law |
# Not All Beliefs Are Created Equal: Diagnosing Toxic Ideologies
*Epistemic status: exploratory but confident. This essay presents a general framework for identifying and analyzing ideologies based on recurring structural patterns. It draws on observation, theory, and examples from across the political spectrum. The g... | https://www.lesswrong.com/posts/yuh9CtkiaBkmuLtFZ/not-all-beliefs-are-created-equal-diagnosing-toxic |
# Research Notes: Running Claude 3.7, Gemini 2.5 Pro, and o3 on Pokémon Red
*Disclaimer: this post was not writ... | https://www.lesswrong.com/posts/8aPyKyRrMAQatFSnG/research-notes-running-claude-3-7-gemini-2-5-pro-and-o3-on |
# Improving CNNs with Klein Networks: A Topological Approach to AI
In our [earlier post](https://www.lesswrong.com/posts/m4fzvBgw95LjkJKjG/from-loops-to-klein-bottles-uncovering-hidden-topology-in), we described how one could parametrize local image patches in natural images by a surface called a Klein bottle. In [Lo... | https://www.lesswrong.com/posts/jtAPDii97mS8msPEr/improving-cnns-with-klein-networks-a-topological-approach-to |
# Crime and Punishment #1
This seemed like a good next topic to spin off from monthlies and make into its own occasional series. There’s certainly a lot to discuss regarding crime.
What I don’t include here, the same way I excluded it from the monthly, are the various crimes and other related activities that may or m... | https://www.lesswrong.com/posts/9TPEjLH7giv7PuHdc/crime-and-punishment-1 |
# Feature-Based Analysis of Safety-Relevant Multi-Agent Behavior
**Introduction**
----------------
### **TL;DR**
[Today’s AI systems are becoming increasingly agentic and interconnected, giving rise to a future of multi-agent (MA) systems (MAS). It is believed that this will introduce unique risks and thus require n... | https://www.lesswrong.com/posts/BFWbufKSPR36k8wc9/feature-based-analysis-of-safety-relevant-multi-agent |
# The Uses of Complacency
[
](https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progress... | https://www.lesswrong.com/posts/cQ2ub5wjdyzoHTbab/the-uses-of-complacency |
# Podcast on “AI tools for existential security” — transcript
About two months ago,[ @finm](https://forum.effectivealtruism.org/users/finm?mention=user) and I recorded a discussion about my now-published[ piece](https://www.lesswrong.com/posts/jDchMg9dcr4a3dJtj/ai-tools-for-existential-security) on “AI tools for Exist... | https://www.lesswrong.com/posts/eMoTJAFFKqEAneqtm/podcast-on-ai-tools-for-existential-security-transcript |
# The US Executive vs Supreme Court Deportations Clash
**Forecaster perspectives**
---------------------------
Sentinel forecasters in aggregate assess as **“83% true”** (65% to 100%) the statement **that the Trump administration has disobeyed the Supreme Court so far**. On the one hand, the Trump administration has ... | https://www.lesswrong.com/posts/hGXvn6GXABHjfzn7C/the-us-executive-vs-supreme-court-deportations-clash |
# $500 Bounty Problem: Are (Approximately) Deterministic Natural Latents All You Need?
**\[**~~**EDIT Aug 22 2025: This bounty is now closed! We have**~~ [~~**solved it ourselves**~~](https://www.lesswrong.com/posts/Gd36HT7qLr684SYuQ/stochastic-natural-latent-implies-deterministic-natural?utm_campaign=post_share&utm_s... | https://www.lesswrong.com/posts/e9KwDDdAxborNSuCd/usd500-bounty-problem-are-approximately-deterministic |
# AE Studio is hiring!
**AE Studio is Hiring: Data Scientists & ML Engineers for AI Alignment Work**
-----------------------------------------------------------------------------
**Location:** Remote or Los Angeles (in-person strongly encouraged)
**Type:** Full-time
**Compensation:** Competitive salary + mean... | https://www.lesswrong.com/posts/fBJQyrgPEjZJ4iZLp/ae-studio-is-hiring |
# Most AI value will come from broad automation, not from R&D
This is a linkpost to an article by [Ege Erdil](https://www.lesswrong.com/users/ege-erdil) and I that we wrote to explain an important perspective that we share regarding AI automation. I'll quote the introduction:
> A popular view about the future impact ... | https://www.lesswrong.com/posts/csFdbopaFQiKma2vF/most-ai-value-will-come-from-broad-automation-not-from-r-and |
# Accountability Sinks
This is a cross-post from [https://250bpm.substack.com/p/accountability-sinks](https://250bpm.substack.com/p/accountability-sinks)
[ Dwarkesh Patel podcast with founders Tamay Besiroglu and Ege Erdil. Mechanize... | https://www.lesswrong.com/posts/j7GgmLwymtJEcnf9L/you-better-mechanize |
# Societal and technological progress as sewing an ever-growing, ever-changing, patchy, and polychrome quilt
**Joel Z. Leibo** \[1\], **Alexander Sasha Vezhnevets** \[1\], **William A. Cunningham** \[1, 2\], **Sébastien Krier** \[1\], **Manfred Diaz** \[3\], **Simon Osindero** \[1\]
\[1\] Google DeepMind, \[2\] Unive... | https://www.lesswrong.com/posts/S8KYwtg5vZqah4SHF/societal-and-technological-progress-as-sewing-an-ever |
# Manifund 2025 Regrants
Each year, Manifund partners with *regrantors*: experts in the field of AI safety, each given an independent budget of $100k+. Regrantors can initiate fast, small grants, seeding early-stage projects with $5k-$50k.
For 2025, we’ve raised $2.25m so far, and are excited to announce our first 10... | https://www.lesswrong.com/posts/LBg7xNuzNGDuQGqKF/manifund-2025-regrants |
# Corrupted by Reasoning: Reasoning Language Models Become Free-Riders in Public Goods Games
### Summary:
* **Traditional LLMs outperform reasoning models in cooperative Public Goods tasks.** Models like Llama-3.3-70B maintain ~90% contribution rates in public goods games, while reasoning-focused models (o1, o3 ser... | https://www.lesswrong.com/posts/M6dXdCbdoLSpHt8v3/corrupted-by-reasoning-reasoning-language-models-become-free |
# The EU Is Asking for Feedback on Frontier AI Regulation (Open to Global Experts)—This Post Breaks Down What’s at Stake for AI Safety
The [European AI Office](https://digital-strategy.ec.europa.eu/en/policies/ai-office) is currently writing the rules for how general-purpose AI (GPAI) models will be governed under the... | https://www.lesswrong.com/posts/AbnfsnEEmHFmprGzm/the-eu-is-asking-for-feedback-on-frontier-ai-regulation-open |
# Is alignment reducible to becoming more coherent?
*Epistemic status: Like all alignment ideas, this one is incomplete and/or wrong, but I am hoping mostly incomplete and not wrong.*
One of the hard subproblems of alignment is constructing a "stable pointer to value" (see this [overview](https://www.lesswrong.com/w/... | https://www.lesswrong.com/posts/nuDJNyG5XLQjtvaeg/is-alignment-reducible-to-becoming-more-coherent |
# To Understand History, Keep Former Population Distributions In Mind
Guillaume Blanc has a [piece in *Works in Progress*](https://worksinprogress.co/issue/frances-baby-bust/) (I assume based on [his paper](https://hummedia.manchester.ac.uk/schools/soss/economics/discussionpapers/EDP-2309.pdf)) about how France’s fer... | https://www.lesswrong.com/posts/gk2aJgg7yzzTXp8HJ/to-understand-history-keep-former-population-distributions |
# Are we "being poisoned"?
I would like to revisit some of the concepts Scott explored in his 2020 article "[For, Then Against, High-Saturated-Fat Diets](https://slatestarcodex.com/2020/03/10/for-then-against-high-saturated-fat-diets/)". I'm hoping someone will have some novel/updated insights or new research to share... | https://www.lesswrong.com/posts/duvBW3rEq9D3QkBC5/are-we-being-poisoned |
# Jaan Tallinn's 2024 Philanthropy Overview
to follow up my [philanthropic pledge](https://www.lesswrong.com/posts/ZMRCpfGD4iXs5Mi5Q/jaan-tallinn-s-philanthropic-pledge) from 2020, i've updated my [philanthropy page](https://jaan.info/philanthropy/) with the 2024 results.
in 2024 my donations funded [$51M worth of en... | https://www.lesswrong.com/posts/8ojWtREJjKmyvWdDb/jaan-tallinn-s-2024-philanthropy-overview |
# Putting up Bumpers
**tl;dr:** Even if we can't solve alignment, we can solve the problem of catching and fixing misalignment.
If a child is bowling for the first time, and they just aim at the pins and throw, they’re almost certain to miss. Their ball will fall into one of the gutters. But if there were beginners’ ... | https://www.lesswrong.com/posts/HXJXPjzWyS5aAoRCw/putting-up-bumpers |
# o3 Is a Lying Liar
I love o3. I’m using it for most of my queries now.
But that damn model is a lying liar. Who lies.
![](https://substackcdn.com/image/fetch/w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa0444c0a-ef24-468c-895d-42dcccd... | https://www.lesswrong.com/posts/KgPkoopnmmaaGt3ka/o3-is-a-lying-liar |
# What AI safety plans are there?
I'm trying to make a list of every written plan for how to make AGI go well. I am interested in hearing about any plans that I missed, or any lists of plans that have already been compiled.
Here are the plans I've found so far, in no particular order:
- Google https://deepmind.googl... | https://www.lesswrong.com/posts/iS4g58qQEJzjMzYZJ/what-ai-safety-plans-are-there |
# OpenAI Alums, Nobel Laureates Urge Regulators to Save Company's Nonprofit Structure
### Converting to a for-profit model would undermine the company's founding mission to ensure AGI "benefits all of humanity," argues new letter
*This is the full text of a post from *[*Obsolete*](https://garrisonlovely.substack.com/... | https://www.lesswrong.com/posts/rN8tHAJnRYgN7hfoF/openai-alums-nobel-laureates-urge-regulators-to-save-company |
# My Favorite Productivity Blog Posts
I’ve read at least a few hundred blog posts, maybe upwards of a thousand. [Agreeing](https://www.gleech.org/essays) with Gavin Leech, I believe I’ve gained from essays more than any other medium. I’m an intellectually curious college student with a strong interest in both the prac... | https://www.lesswrong.com/posts/ArizxGwbqiohBX5y6/my-favorite-productivity-blog-posts |
# Personal evaluation of LLMs, through chess
Lots of people seem to resonate with the [idea](https://www.lesswrong.com/posts/4mvphwx5pdsZLMmpY/recent-ai-model-progress-feels-mostly-like-bullshit) that AI benchmarks are getting more and more meaningless – either because they're being run in a misleading way, or because... | https://www.lesswrong.com/posts/gNFixvxw7JxzvMjCJ/personal-evaluation-of-llms-through-chess |
# The Intelligence Curse: an essay series
*We've published an essay series on what we call the intelligence curse. Most content is brand new, and all previous writing has been heavily reworked.*
*Visit *[***intelligence-curse.ai***](http://intelligence-curse.ai) *for the full series.*
*Below is the introduction and ... | https://www.lesswrong.com/posts/LCFgLY3EWb3Gqqxyi/the-intelligence-curse-an-essay-series |
# AI #113: The o3 Era Begins
Enjoy it while it lasts. The Claude 4 era, or the o4 era, or both, are coming soon. Also, welcome to 2025, we measure eras in weeks or at most months. For now, the central thing going on continues to be everyone adapting to the world of o3, [a model that is excellent](https://thezvi.substa... | https://www.lesswrong.com/posts/7x9MZCmoFA2FtBtmG/ai-113-the-o3-era-begins |
# “The Era of Experience” has an unsolved technical alignment problem
Every now and then, some AI luminaries
* (1) propose that the future of powerful AI will be reinforcement learning agents—an algorithm class that in many ways has more in common with [MuZero (2019)](https://en.wikipedia.org/wiki/MuZero) than with... | https://www.lesswrong.com/posts/TCGgiJAinGgcMEByt/the-era-of-experience-has-an-unsolved-technical-alignment |
# Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
> Recent breakthroughs in reasoning-focused large language models (LLMs) like OpenAI-o1, DeepSeek-R1, and Kimi-1.5 have largely relied on *Reinforcement Learning with Verifiable Rewards* (RLVR), which replaces human anno... | https://www.lesswrong.com/posts/ZGrkhAjQJoscdSFhb/does-reinforcement-learning-really-incentivize-reasoning |
# Finding an Error-Detection Feature in DeepSeek-R1
*I recently did a weekend research sprint investigating error-detection behaviour in DeepSeek R1 models and writing up preliminary findings. I think I found some cool stuff on how the model tracks its mistakes, and wanted to share the rough write-up in case it's of ... | https://www.lesswrong.com/posts/BnzDiSYKfrHaFPefi/finding-an-error-detection-feature-in-deepseek-r1 |
# Reward hacking is becoming more sophisticated and deliberate in frontier LLMs
Something’s changed about reward hacking[^qoco6orjkn] in recent systems. In the past, reward hacks were usually accidents, found by non-general, RL-trained systems. Models would randomly explore different behaviors and would sometimes come... | https://www.lesswrong.com/posts/rKC4xJFkxm6cNq4i9/reward-hacking-is-becoming-more-sophisticated-and-deliberate |
# Training-time schemers vs behavioral schemers
(Thanks to Vivek Hebbar, Buck Shlegeris, Charlie Griffin, Ryan Greenblatt, Thomas Larsen, and Joe Carlsmith for feedback.)
People use the word “schemer” in two main ways:
1. “Scheming” (or similar concepts: “[deceptive alignment](https://www.lesswrong.com/s/r9tYkB... | https://www.lesswrong.com/posts/m5nWc9v6MTsWXKpCy/training-time-schemers-vs-behavioral-schemers |
# Trouble at Miningtown: Prologue
In late 2019 I wrote a TTRPG.
The theme was alien sentience (or perhaps sapience is the technical term), both "organic" (extra terrestrial) and "artificial" (AI/robots).
This is the prologue that kicks off play.
I found it really fascinating to look back on late 2019 and how I wa... | https://www.lesswrong.com/posts/Mg8AFz6n7wMqvsYTz/trouble-at-miningtown-prologue |
# Token and Taboo
What in retrospect seem like serious moral crimes were often widely accepted while they were happening. This means that moral progress can require intellectual progress.[^pbbn68l9gnc] Intellectual progress often requires questioning received ideas, but questioning moral norms is sometimes taboo. For ... | https://www.lesswrong.com/posts/Zi4t6gfLsKokb9KAc/token-and-taboo |
# This prompt (sometimes) makes ChatGPT think about terrorist organisations
Yesterday, I couldn't wrap my head around some programming concepts in Python, so I turned to ChatGPT (gpt-4o) for help. This evolved into a very long conversation (the longest I've ever had with it by far), at the end of which I pasted around... | https://www.lesswrong.com/posts/TDsmcaryeQAQD8ior/this-prompt-sometimes-makes-chatgpt-think-about-terrorist |
# Modifying LLM Beliefs with Synthetic Document Finetuning
*In this post, we study whether we can modify an LLM’s beliefs and investigate whether doing so could decrease risk from advanced AI systems.*
*We describe a pipeline for modifying LLM beliefs via synthetic document finetuning and introduce a suite of evaluat... | https://www.lesswrong.com/posts/ARQs7KYY9vJHeYsGc/modifying-llm-beliefs-with-synthetic-document-finetuning |
# A review of "Why Did Environmentalism Become Partisan?"
I was recently encouraged to read Jeffrey Heninger's report "[Why Did Environmentalism Become Partisan?](https://aiimpacts.org/wp-content/uploads/2023/04/Why-Did-Environmentalism-Become-Partisan-1.pdf)" It was interesting, but I thought it had some critical fl... | https://www.lesswrong.com/posts/SwqGjXfqLLcNpjrq6/a-review-of-why-did-environmentalism-become-partisan |
# Zstd Window Size
At work we've recently been using [zstd](https://en.wikipedia.org/wiki/Zstd) as a better-compressing alternative to [gzip](https://en.wikipedia.org/wiki/Gzip), and overall I've been pretty happy with it. A minor documentation gripe, though, is that the behavior around multithreaded compression is a ... | https://www.lesswrong.com/posts/QY9aNrn5ociXzomo9/zstd-window-size |
# Will Programmer Compensation Decouple from Productivity?
Since the 1970s, productivity has outpaced wage growth in many sectors. One sector where wages have kept up with productivity growth is tech, especially among programmers. But, I suspect this is changing thanks to the use of AI coding assistants.
To quickly r... | https://www.lesswrong.com/posts/fyC4p6eFYivmmuqbs/will-programmer-compensation-decouple-from-productivity |
# Why would AI companies use human-level AI to do alignment research?
*Cross-posted from [my website](https://mdickens.me/2025/04/25/bootstrapped_alignment/).*
Many plans for how to safely build superintelligent AI have a critical section that goes like this:
1. Develop AI that's powerful enough to do AI research, b... | https://www.lesswrong.com/posts/XLNxrFfkyrdktuzqn/why-would-ai-companies-use-human-level-ai-to-do-alignment |
# Worries About AI Are Usually Complements Not Substitutes
A common claim is that concern about \[X\] ‘distracts’ from concern about \[Y\]. This is often used as an attack to cause people to discard \[X\] concerns, on pain of being enemies of \[Y\] concerns, as attention and effort are presumed to be zero-sum.
There ... | https://www.lesswrong.com/posts/sHXGRTTs4SrceEnoG/worries-about-ai-are-usually-complements-not-substitutes |
# Who's Working On It? AI-Controlled Experiments
[
](https://substackcdn.com/image/fetch/f_auto,q... | https://www.lesswrong.com/posts/qsGew2r4Cy8HcfGhD/who-s-working-on-it-ai-controlled-experiments |
# AI 2027 Thoughts
[AI 2027](https://ai-2027.com/) portrays two well thought out scenarios
for how AI is likely to impact the world toward the end of this decade.
I expect those scenarios will prove to be moderately wrong, but close
enough to be scary. I also expect that few people will manage to make
forecasts ... | https://www.lesswrong.com/posts/drEB34CwA6kHmcuTZ/ai-2027-thoughts |
# Don't you mean "the most *conditionally* forbidden technique?"
Hear me out, I think the [most forbidden technique](https://www.lesswrong.com/posts/mpmsK8KKysgSKDm2T/the-most-forbidden-technique) is very useful and should be used, **as long as** we avoid the "most forbidden aftertreatment:"
1. An AI trained on inte... | https://www.lesswrong.com/posts/jokdGg3wciyoeZTN2/don-t-you-mean-the-most-conditionally-forbidden-technique-1 |
# [LAPTOP REQUIRED] Lighthaven Sequences Reading Group #32 (Tuesday 04/29)
Come get old-fashioned with us, and let's read the sequences at Lighthaven! We'll show up, mingle, do intros, and then get to beta test an app Lightcone is developing for LessOnline. Please do the reading beforehand - it should be no more than ... | https://www.lesswrong.com/events/kKh2rpN29KE72sjCu/laptop-required-lighthaven-sequences-reading-group-32 |
# How I Think About My Research Process: Explore, Understand, Distill
*This is the first post in a sequence about how I think about and break down my research process. Post 2 is coming soon.*
*Thanks to Oli Clive-Griffin, Paul Bogdan, Shivam Raval and especially to Jemima Jones for help and feedback, and to my co-aut... | https://www.lesswrong.com/posts/hjMy4ZxS5ogA9cTYK/how-i-think-about-my-research-process-explore-understand |
# AI Safety & Entrepreneurship v1.0
+++ Why did I create both a post and a wiki article?
Posts are best for making sure people see the initial version of the post, whilst Wiki articles are best for long-term maintenance. Posting an article that links to a Wiki page provides the best of both worlds.
+++
### **For... | https://www.lesswrong.com/posts/WbBe7LKNwv7fBgqii/ai-safety-and-entrepreneurship-v1-0 |
# We should try to automate AI safety work asap
*This is a personal post and does not necessarily reflect the opinion of other members of Apollo Research. I think I could have written a better version of this post with more time. However, my main hope for this post is that people with more expertise use this post as a... | https://www.lesswrong.com/posts/W3KfxjbqBAnifBQoi/we-should-try-to-automate-ai-safety-work-asap |
# What are important UI-shaped problems that Lightcone could tackle?
As I think about "what to do about AI x-risk?", some principles that seem useful to me:
1. Short timelines seem plausible enough that, **for the next year or so, I'd like to focus on plans that are relevant if takeoff begins in the next few years**... | https://www.lesswrong.com/posts/t46PYSvHHtJLxmrxn/what-are-important-ui-shaped-problems-that-lightcone-could |
# Open Source LLM Pokémon Scaffold
[This is a cleaned-up, open-source version](https://github.com/cicero225/llm_pokemon_scaffold) of the LLM Pokémon Scaffold described in [Research Notes: Running Claude 3.7, Gemini 2.5 Pro, and o3 on Pokémon Red](https://www.lesswrong.com/posts/8aPyKyRrMAQatFSnG/untitled-draft-x7cc). ... | https://www.lesswrong.com/posts/Qk3kCb68NvKBayHZB/open-source-llm-pokemon-scaffold |
# MiCARwave
Last year at NEFFA we mostly bought food. We brought a range of cold options, but they weren't very tasty and the kids weren't very excited about them. This time we've mostly eaten food we brought. Why? The miCARwave!
[](https://www.jefftk.com/micarwave-at... | https://www.lesswrong.com/posts/nPo7WpcNwqtCdrist/micarwave |
# AI Self Portraits Aren't Accurate
*For a lay audience, but I've seen a surprising number of knowledgeable people fretting over depressed-seeming comics from current systems. Either they're missing something or I am. *
Perhaps you’ve seen images like [this self-portrait](https://x.com/Josikinz/status/190544094905417... | https://www.lesswrong.com/posts/Cvvf5w2j6BPJd8BzL/ai-self-portraits-aren-t-accurate |
# "The Urgency of Interpretability" (Dario Amodei)
Dario Amodei posted a new essay titled "[The Urgency of Interpretability](https://www.darioamodei.com/post/the-urgency-of-interpretability)" a couple days ago.
Some excerpts I think are worth highlighting:
> The nature of AI training makes it possible that AI system... | https://www.lesswrong.com/posts/SebmGh9HYdd8GZtHA/the-urgency-of-interpretability-dario-amodei |
# Emergence of superintelligence from AI hiveminds: how to make it human-friendly?
"Hive mind" may not be the ideal term. What I'm really envisioning is something modelled on the existing relationship between human users and frontier LLMs, but with the "users" actually just being LLMs too, or simple programs that medi... | https://www.lesswrong.com/posts/3ELw74tGvLFvkZC3P/emergence-of-superintelligence-from-ai-hiveminds-how-to-make |
# After Internet Dependency
ChatGPT recently introduced improved cross-chat memory, and I've found myself using it a lot more sense then. Everything from helping write components with unfamiliar JavaScript libraries, to counseling me through a breakup. So much so that when I face a new problem, my mind turns not to br... | https://www.lesswrong.com/posts/qDzL2vuDuvkcsbbJw/after-internet-dependency |
# My Research Process: Key Mindsets - Truth-Seeking, Prioritisation, Moving Fast
*This is post 2 of a sequence on my framework for doing and thinking about research.* [*Start here*](https://www.alignmentforum.org/s/5GT3yoYM9gRmMEKqL/p/hjMy4ZxS5ogA9cTYK)*.*
Before I get into what exactly to do at each stage of the res... | https://www.lesswrong.com/posts/cbBwwm4jW6AZctymL/my-research-process-key-mindsets-truth-seeking |
# The case for multi-decade AI timelines [Linkpost]
So this post is an argument that multi-decade timelines are reasonable, and the key cruxes that Ege Erdil has with most AI safety people who believe in short timelines are due to the following set of beliefs:
1. Ege Erdil doesn't believe that trends exist that requ... | https://www.lesswrong.com/posts/xxxK9HTBNJvBY2RJL/the-case-for-multi-decade-ai-timelines-linkpost |
# Questions for old LW members: how have discussions about AI changed compared to 10+ years ago?
I wasn't on LessWrong 10 years ago, during the pre-ChatGPT era, so I'd like to hear from people who were here back in the day. I want to know how AI-related discussions have changed and how people's opinions have changed.
... | https://www.lesswrong.com/posts/ssSAsTx2FPzdTt5jG/questions-for-old-lw-members-how-have-discussions-about-ai |
# Our Reality: A Simulation Run by a Paperclip Maximizer
Our universe is probably a computer simulation created by a paperclip maximizer to map the spectrum of rival resource‑grabbers it may encounter while expanding through the cosmos. The purpose of this simulation is to see what kind of ASI (artificial superintelli... | https://www.lesswrong.com/posts/HxLYnGYspLoeLLrE6/our-reality-a-simulation-run-by-a-paperclip-maximizer-1 |
# How people use LLMs
I've gotten a lot of value out of the details of how other people use LLMs, so I'm delighted that Gavin Leech created a collection of exactly such posts (link should go to the right section of the page but if you don't see it, scroll down).
* [https://kajsotala.fi/2025/01/things-i-have-been-u... | https://www.lesswrong.com/posts/FXnvdeprjBujt2Ssr/how-people-use-llms |
# The Way You Go Depends A Good Deal On Where You Want To Get: FEP minimizes surprise about actions using preferences about the future as *evidence*
The motivation for this post came when I was reading various Scott Alexander posts on the Free Energy Principle, in which he seemed very confused about it (the title [*Go... | https://www.lesswrong.com/posts/JGTiH8DhpkFAhxgCh/the-way-you-go-depends-a-good-deal-on-where-you-want-to-get |
# Thin Alignment Can't Solve Thick Problems
> "Plurality is the law of the earth."
> —Hannah Arendt, *The Human Condition* (1958)
Abstract
--------
*Contemporary AI alignment practices overwhelmingly rely on thin ethical concepts — formal, universal notions such as "honesty," "harmlessness," and "helpfulness" — wh... | https://www.lesswrong.com/posts/uyfmdMKGjYHqiLbbX/thin-alignment-can-t-solve-thick-problems |
# Therapist in the Weights: Risks of Hyper-Introspection in Future AI Systems
*Epistemic Status: Speculative, but grounded in current trends in interpretability and self-supervised learning. This was initially written as an outline at some point in 2024, but written/rewritten more fully in response to Anthropic’s anno... | https://www.lesswrong.com/posts/HHpiDAb3sHerkG93K/therapist-in-the-weights-risks-of-hyper-introspection-in |
# Keltham on Becoming more Truth-Oriented
*This is IMO high-quality material from* [*projectlawful*](https://projectlawful.com/) *which I thought is worth crossposting directly to lesswrong.*
*(This might dispel some wrong beliefs you hold (though maybe only if you actively try to apply or practice the techniques). I... | https://www.lesswrong.com/posts/8XdQBokzMfY9FqqT3/keltham-on-becoming-more-truth-oriented |
# “A victory for the natural order”
> In what many are calling "a victory for the natural order," the final remaining vials of the controversial CHRONO-X anti-aging treatment were ceremoniously destroyed yesterday at Massachusetts General Hospital in Boston.
>
> "Today marks humanity's return to wisdom," declared Gio... | https://www.lesswrong.com/posts/XdzJ5BNbiYKGq7Mxw/a-victory-for-the-natural-order |
# 7+ tractable directions in AI control
In this post, we list 7 of our favorite easy-to-start directions in AI control. (Really, projects that are at least adjacent to AI control; We include directions which aren’t as centrally AI control and which also have other motivations.) This list is targeted at independent res... | https://www.lesswrong.com/posts/wwshEdNhwwT4r9RQN/7-tractable-directions-in-ai-control |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.