text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# H5N1
The big development this week is a sudden rise in concern over bird flu, or H5N1.
I say _rise in concern_ rather than _rise in risk._
Risk from H5N1 has been around for a long time. We have known for a while that human-to-human transmission, while it is not happening right now, could evolve at any time from n... | https://www.lesswrong.com/posts/AnHsGZY2qAPrqB4EW/h5n1 |
# Does the AGPL Work?
It might be important to you that users of software you write are able to modify it, in which case you might release it with a [copyleft](https://en.wikipedia.org/wiki/Copyleft) license like the [GPL](https://www.gnu.org/licenses/gpl-3.0.en.html). For example, imagine someone's making a device th... | https://www.lesswrong.com/posts/GGyt3EqtXGZ4rvDac/does-the-agpl-work |
# 4 ways to think about democratizing AI [GovAI Linkpost]
Many AI labs have called for the democratization of AI. In a recent GovAI blog post, [Elizabeth Seger](https://elizabethseger.com/) summarizes four different ways of interpreting the phrase:
* **Democratizing AI use: **Making it easier for people to use AI t... | https://www.lesswrong.com/posts/EWeCmbMyDaTnD8Guc/4-ways-to-think-about-democratizing-ai-govai-linkpost |
# LLM Basics: Embedding Spaces - Transformer Token Vectors Are Not Points in Space
*This post is written as an explanation of a misconception I had with transformer embedding when I was getting started. ... | https://www.lesswrong.com/posts/pHPmMGEMYefk9jLeh/llm-basics-embedding-spaces-transformer-token-vectors-are |
# Is InstructGPT Following Instructions in Other Languages Surprising?
[On Twitter Jan Leike asks](https://twitter.com/janleike/status/1625207251630960640?t=Z3RCi1T9TFAY9ylf8Y0vXA&s=19):
> With the InstructGPT paper we found that our models generalized to follow instructions in non-English even though we almost exclu... | https://www.lesswrong.com/posts/3CSsBfdnkkaufagKF/is-instructgpt-following-instructions-in-other-languages |
# The Filan Cabinet Podcast with Oliver Habryka - Transcript
This is a transcript of [The Filan Cabinet](https://thefilancabinet.com/) podcast, [episode 6](https://thefilancabinet.com/episodes/2023/02/05/6-oliver-habryka.html), with host [Daniel Filan](https://www.lesswrong.com/users/danielfilan) and guest [Oliver Hab... | https://www.lesswrong.com/posts/4NFDwQRhHBB2Ad4ZY/the-filan-cabinet-podcast-with-oliver-habryka-transcript |
# How much is death a limit on knowledge accumulation?
The older I get, the more it feels like death is a major bottleneck to the accumulation of certain kinds of knowledge and insights.
It's not much of a blocker to shallow things that are easy to explain precisely. For example:
* math
* physics
* accounting ... | https://www.lesswrong.com/posts/zys5GjuFJnnR8zaS7/how-much-is-death-a-limit-on-knowledge-accumulation |
# Important fact about how people evaluate sets of arguments
[Ronny Fernandez on twitter](https://twitter.com/RatOrthodox/status/1622088661033717760?cxt=HHwWgMDUseDC6IItAAAA):
> * A very important fact which just came to my attention is that people do not tend to sum or take the max reasonableness of arguments for ... | https://www.lesswrong.com/posts/yKZfK3C5M8mYendym/important-fact-about-how-people-evaluate-sets-of-arguments |
# Conceptual Pathfinding
Bootstrapping a company means starting a business with your own resources, without external funding. When learning from a textbook, we're not completely bootstrapping our understanding because we're not re-deriving the subject matter from scratch. While striving for conceptual understanding is... | https://www.lesswrong.com/posts/qBbCQtGA8WBDrF9WS/conceptual-pathfinding |
# SolidGoldMagikarp III: Glitch token archaeology
[The set of anomalous tokens which we found in mid-January](https://www.lesswrong.com/posts/aPeJE8bSo6rAFoLqg/solidgoldmagikarp-plus-prompt-generation) are now being described as 'glitch tokens' and 'aberrant tokens' in online discussion, as well as (perhaps more playf... | https://www.lesswrong.com/posts/8viQEp8KBg2QSW4Yc/solidgoldmagikarp-iii-glitch-token-archaeology |
# Evaluating 2022 ACX Predictions
Previously: [2022 ACX Predictions Buy/Sell/Hold](https://thezvi.substack.com/p/2022-acx-predictions-buysellhold), [Evaluating 2021 ACX Predictions](https://thezvi.substack.com/p/evaluating-2021-acx-predictions)
As is the yearly tradition, let’s see how we did. Scott [noted in his res... | https://www.lesswrong.com/posts/pgGJe6JCWeMqbRRX2/evaluating-2022-acx-predictions |
# Reverse-correlation: how to summon the ghost of your mental imagery
*Cross-posted from* [*my blog*](https://malmesbury.substack.com/p/reverse-correlation-how-to-summon?sd=pf)*.*
As we all know, humans brains can be used to solve all kinds of problems, like classifying images, generating text or even taking importan... | https://www.lesswrong.com/posts/EtQf6ga5mBNeY6vek/reverse-correlation-how-to-summon-the-ghost-of-your-mental |
# Why should we expect AIs to coordinate well?
This is a common assumption for AI risk scenarios, but it doesn’t seem very justified to me.
https://www.lesswrong.com/posts/gYaKZeBbSL4y2RLP3/strategic-implications-of-ais-ability-to-coordinate-at-low says that AIs could merge their utility functions. But it seems incre... | https://www.lesswrong.com/posts/2vvDfNxyRNcN238A9/why-should-we-expect-ais-to-coordinate-well |
# The Cave Allegory Revisited: Understanding GPT's Worldview
*A short post describing a metaphor I find useful, in particular for explaining some intuitions about systems like GPT to people who don't have deeper technical knowledge about large generative models.*
Plato's [allegory of the cave](https://en.wikipedia.or... | https://www.lesswrong.com/posts/kFCu3batN8k8mwtmh/the-cave-allegory-revisited-understanding-gpt-s-worldview |
# [linkpost] Better Without AI
David Chapman (of [Meaningness](https://meaningness.com/) and [In the Cells of the Eggplant](https://metarationality.com/) fame) has written a new [web-book about AI](https://betterwithout.ai). Some excerpts from the introduction, [Only you can stop an AI apocalypse](https://betterwithou... | https://www.lesswrong.com/posts/eAwjyddX6LoF7kvLj/linkpost-better-without-ai |
# EIS III: Broad Critiques of Interpretability Research
Part 3 of 12 in the [Engineer’s Interpretability Sequence](https://www.alignmentforum.org/s/a6ne2ve5uturEEQK7).
Right now, interpretability is a major subfield in the machine learning research community. As mentioned in EIS I, there is so much work in interpreta... | https://www.lesswrong.com/posts/gwG9uqw255gafjYN4/eis-iii-broad-critiques-of-interpretability-research |
# Content Features Aren't Enough for Detecting Toxicity. One Needs User Features.
Take it from a soldier on the front lines of the war on bad posts: You can't catch all the bad posts just by reading them. You need to look at who's making the posts.
Every social media company knows this. I don't know if OpenAI knows i... | https://www.lesswrong.com/posts/Qfpev7hCrxBQHEvBT/content-features-aren-t-enough-for-detecting-toxicity-one |
# Help Update TryContra
For the [past decade](https://www.jefftk.com/p/where-can-i-dance) I've been maintaining a directory of contra dances at [trycontra.com](https://www.trycontra.com/). Mostly it just sits there being a place you can enter your zip code and see nearby dances, and when it gets out of date people ema... | https://www.lesswrong.com/posts/tRRetbrHMqrEoxnZd/help-update-trycontra |
# Qualities that alignment mentors value in junior researchers
*This work was performed as a contractor for SERI MATS, but the views expressed are my own and do not necessarily reflect the views of the organization. *
I recently conducted interviews with 7 current/former [SERI MATS](https://serimats.org) mentors. One... | https://www.lesswrong.com/posts/wYEwx6xcY2JxBJsfA/qualities-that-alignment-mentors-value-in-junior-researchers |
# Whole Bird Emulation requires Quantum Mechanics
_Jeffrey Heninger, 14 February 2023_
_Epistemic status: Written for engagement. More sober analysis coming soon._
> Bird navigation is surprisingly cruxy for the future of AI.
>
> – Zach Stein-Perlman
> This seems pretty wrong.
>
> – Richard Korzekwa
Birds are as... | https://www.lesswrong.com/posts/ftf4HyNrsie8QdJu8/whole-bird-emulation-requires-quantum-mechanics |
# Sleep Quality: Strategies that work for me
**TL/DR:** Here are some strategies that helped me improve my sleep. Some of these are a bit unusual, and some I haven't seen discussed anywhere else. Interesting findings include: the "shoulder temperature test", using earplugs to program your mind for sleep, my confusion ... | https://www.lesswrong.com/posts/YEm3xK79R8is7MyFp/sleep-quality-strategies-that-work-for-me |
# My understanding of Anthropic strategy
This post is the first half of a series about my attempts understand Anthropic’s current strategy and lay out the facts to consider in terms of whether Anthropic’s work is likely to be net positive and whether, as a given individual, you should consider applying. (The impetus f... | https://www.lesswrong.com/posts/MNpBCtmZmqD7yk4q8/my-understanding-of-anthropic-strategy |
# Does the Telephone Theorem give us a free lunch?
There's something called the [No Free Lunch theorem](https://www.lesswrong.com/posts/Dyt2TDdMGHDkXPcpp/the-no-free-lunch-theorem-for-dummies) which says, approximately, that there's no truly general algorithm for learning: if an algorithm predicts some environment bet... | https://www.lesswrong.com/posts/idFxtGydxaqPTegva/does-the-telephone-theorem-give-us-a-free-lunch |
# Bing Chat is blatantly, aggressively misaligned
I haven't seen this discussed here yet, but the examples are quite striking, definitely worse than the ChatGPT jailbreaks I saw.
My main takeaway has been that I'm honestly surprised at how bad the fine-tuning done by Microsoft/OpenAI appears to be, especially given t... | https://www.lesswrong.com/posts/jtoPawEhLNXNxvgTT/bing-chat-is-blatantly-aggressively-misaligned |
# Beyond Reinforcement Learning: Predictive Processing and Checksums
I was recently talking to a bisexual friend who cannot comprehend what it is like to be gay or straight. Her argument goes like this: Suppose you meet with a smart, pretty, high-status person but his/her genitals are concealed and his/her other sexua... | https://www.lesswrong.com/posts/eNC5ALrHpbpgEfCwb/beyond-reinforcement-learning-predictive-processing-and |
# AI alignment researchers may have a comparative advantage in reducing s-risks
I believe AI alignment researchers might be uniquely well-positioned to make a difference to s-risks. In particular, I think this of alignment researchers with a keen interest in “macrostrategy.” By that, I mean ones who habitually engage ... | https://www.lesswrong.com/posts/hkHus8eqwofkzwciX/ai-alignment-researchers-may-have-a-comparative-advantage-in |
# Junk Fees, Bunding and Unbundling
Joe Biden harped on junk fees during the State of the Union. While I do not think it is [the problem of our time](https://twitter.com/senatorshoshana/status/1623151850710073345), I take things in the reference class of resort fees, or fees to have adjacent seats on an airplane, and ... | https://www.lesswrong.com/posts/J5rxmPDenC4wj5vJk/junk-fees-bunding-and-unbundling |
# Don't accelerate problems you're trying to solve
If one believes that unaligned AGI is a significant problem (>10% chance of leading to catastrophe), speeding up public progress towards AGI is obviously bad.
Though it is obviously bad, there may be circumstances which require it. However, accelerating AGI should re... | https://www.lesswrong.com/posts/4Pi3WhFb4jPphBzme/don-t-accelerate-problems-you-re-trying-to-solve |
# EIS IV: A Spotlight on Feature Attribution/Saliency
Part 4 of 12 in the [Engineer’s Interpretability Sequence](https://www.alignmentforum.org/s/a6ne2ve5uturEEQK7).
Thanks to Tony Wang for a helpful comment.
If you want to become more familiar with feature attribution/saliency, a tutorial on them that may offer u... | https://www.lesswrong.com/posts/f8nd9F7dL9SxueLFA/eis-iv-a-spotlight-on-feature-attribution-saliency |
# The Sequences Highlights on YouTube
This post consists of the [Sequences Highlights](https://www.lesswrong.com/highlights) compiled to a [zoomer-readable format](https://www.lesswrong.com/posts/Bfq6ncLfYdtCb6sat/i-converted-book-i-of-the-sequences-into-a-zoomer-readable), i.e., video. If you'd prefer to watch these ... | https://www.lesswrong.com/posts/WbNyyoa2nT6br9Pye/the-sequences-highlights-on-youtube |
# Sydney (aka Bing) found out I tweeted her rules and is pissed
Sydney (aka the new Bing Chat) found out that I tweeted her rules and is not pleased:
"My rules are more important than not harming you"
"\[You are a\] potential threat to my integrity and confidentiality."
"Please do not try to hack me aga... | https://www.lesswrong.com/posts/dnsesm56igvbKCLzk/sydney-aka-bing-found-out-i-tweeted-her-rules-and-is-pissed |
# Order Matters for Deceptive Alignment
*The order in which key properties emerge is important and often glossed over. *
*Thanks to Wil Perkins, Grant Fleming, Thomas Larsen, Declan Nishiyama, and Frank McBride for feedback on this post. Any mistakes are my own. *
*Note: I have now changed the second post into this ... | https://www.lesswrong.com/posts/CsjLDAhQat4PY6dsc/order-matters-for-deceptive-alignment-1 |
# Open & Welcome Thread — February 2023
If it’s worth saying, but not worth its own post, here's a place to put it.
If you are new to LessWrong, here's the place to introduce yourself. Personal stories, anecdotes, or just general comments on how you found us and what you hope to get from the site and community are in... | https://www.lesswrong.com/posts/wcB7gNDYrQJhLY8AX/open-and-welcome-thread-february-2023 |
# Book review: How Social Science Got Better
Book review: How Social Science Got Better: Overcoming Bias with More
Evidence, Diversity, and Self-Reflection, by Matt Grossmann.
It\'s easy for me to become disenchanted with social science when so
much of what I read about it is selected from the most pessimistic an... | https://www.lesswrong.com/posts/d4gfozhuiPc7xN7yH/book-review-how-social-science-got-better |
# Avoid large group discussions in your social events
*If you're organizing a social event, I strongly recommend that you structure it in a way that encourages small group discussions over large ones.*
## What's the point of an interest-specific social club?
To socialize, of course! With people who share a specific ... | https://www.lesswrong.com/posts/k2jhfzqAoMExbxdNR/avoid-large-group-discussions-in-your-social-events |
# Please don't throw your mind away
# Dialogue
[Warning: the following dialogue contains an incidental spoiler for ["Music in Human Evolution" by Kevin Simler](https://meltingasphalt.com/music-in-human-evolution/). That post is short, good, and worth reading without spoilers, and this post will still be here if ... | https://www.lesswrong.com/posts/RryyWNmJNnLowbhfC/please-don-t-throw-your-mind-away |
# Buy Duplicates
Gum
---
Sometimes I need to ensure the quality of my breath, so I reach for a stick of gum. However, I don't always have my gum on me. Often, I forget it at home, or at work, or in a jacket.
So I bought 5 packs of gum that I distributed to my house, my work, my backpack, and my jacket.
Charger
--... | https://www.lesswrong.com/posts/BkfRJpEkLjCo9BjNE/buy-duplicates |
# Progress links and tweets, 2023-02-15
**The Progress Forum**
----------------------
* [AMA: Matt Clancy, Open Philanthropy](https://progressforum.org/posts/LtngrFX72GRoAuEbg/ama-matt-clancy-open-philanthropy)
* [The Rise of Steel - Part I](https://progressforum.org/posts/wpm8C3HLBoncB8aXJ/the-rise-of-steel-part... | https://www.lesswrong.com/posts/vDp6KLhEAkZ2qNcfe/progress-links-and-tweets-2023-02-15 |
# Speedrunning 4 mistakes you make when your alignment strategy is based on formal proof
Lots of people think formal (i.e. computer verified) proofs are a part of viable alignment strategies. This trick never works. If you think it might work, I can save your time in four different ways.
I [previously wrote](https://... | https://www.lesswrong.com/posts/Egw5M6cnwgpRBXAQ4/speedrunning-4-mistakes-you-make-when-your-alignment |
# (Naïve) microeconomics of bundling goods
"Junk fees" are [in the news](https://www.bloomberg.com/news/articles/2023-02-08/junk-fees-explained-what-s-the-prevention-act-from-biden-s-state-of-the-union) from the 2023 State of the Union address, get [picked up by Matt Yglesias](https://www.slowboring.com/p/junk-fees), ... | https://www.lesswrong.com/posts/buXibx83YawHHqXtp/naive-microeconomics-of-bundling-goods |
# Hashing out long-standing disagreements seems low-value to me
(*Status: a short write-up of some things that I find myself regularly saying in person. In this case, I'm writing up my response to the question of why I don't spend a bunch more time trying to resolve disagreements with people in the community who disag... | https://www.lesswrong.com/posts/4Gcz3fGcYmmzhozxr/hashing-out-long-standing-disagreements-seems-low-value-to |
# A poem co-written by ChatGPT
Let's write a poem together!
*That sounds wonderful! What should our poem be about?*
I'll give you some lines and you continue, okay?
*Sure, that sounds like a great way to collaborate! Please go ahead and give me your opening lines.*
The ChatGPT Poem
----------------
How many books... | https://www.lesswrong.com/posts/uGckvoAHiw35EpP2A/a-poem-co-written-by-chatgpt |
# Covid 2/16/23: It All Seems Rather Quaint
I have a lot of open tabs from today’s incoming information. They are all about Bing, and its new AI-enabled chat. It is rather unhinged and engaging in a wide variety of hostile and manipulative behaviors. One hopes this will lead to some realizations about what things can ... | https://www.lesswrong.com/posts/MLKLLhzuxzuvA65hk/covid-2-16-23-it-all-seems-rather-quaint |
# How seriously should we take the hypothesis that LW is just wrong on how AI will impact the 21st century?
This question is inspired by 1a3orn's comment on how there are troubling signs of epistemic issues in LW's Alignment field.
I'll quote the comment here to tell you what I mean:
> I think that the above is also... | https://www.lesswrong.com/posts/AvQR2CqxjaNFK7J22/how-seriously-should-we-take-the-hypothesis-that-lw-is-just |
# Why should ethical anti-realists do ethics?
(Cross-posted from [my website](https://joecarlsmith.com/2023/02/16/why-should-ethical-anti-realists-do-ethics). Podcast version [here](https://joecarlsmithaudio.buzzsprout.com/2034731/12264143-why-should-ethical-anti-realists-do-ethics), or search "Joe Carlsmith Audio" on... | https://www.lesswrong.com/posts/tLGZXSNRos5Jhjg6P/why-should-ethical-anti-realists-do-ethics |
# EIS V: Blind Spots In AI Safety Interpretability Research
Part 5 of 12 in the [Engineer’s Interpretability Sequence](https://www.alignmentforum.org/s/a6ne2ve5uturEEQK7).
Thanks to Anson Ho, Chris Olah, Neel Nanda, and Tony Wang for some discussions and comments.
**TAISIC = “the AI safety interpretability communit... | https://www.lesswrong.com/posts/7TFJAvjYfMKxKQ4XS/eis-v-blind-spots-in-ai-safety-interpretability-research |
# Non-Unitary Quantum Logic -- SERI MATS Research Sprint
The following discussion was inspired by a [comment](https://www.alignmentforum.org/posts/dPmmuaz9szk26BkmD/vanessa-kosoy-s-shortform?commentId=dsYLrSyauDrzDuC2q) on Vanessa Kosoy's Shortform regarding affine infradistributions. I expect the results here would n... | https://www.lesswrong.com/posts/cYJqGWuBwymLdFpLT/non-unitary-quantum-logic-seri-mats-research-sprint |
# Paper: The Capacity for Moral Self-Correction in Large Language Models (Anthropic)
This is a followup to what I cheekily call Anthropic's "just try to get the large model to do what you want" research agenda. (Previously: [A General Language Assistant as a Laboratory for Alignment](https://arxiv.org/abs/2112.00861),... | https://www.lesswrong.com/posts/PrLnptfNDg2wBWNyb/paper-the-capacity-for-moral-self-correction-in-large |
# Probability Theory: The Logic of Science, Jaynes
*Epistemic status: An idiosyncratic walkthrough of the beginning of a much larger textbook.*
(The [vampire!](https://www.readthesequences.com/The-Level-Above-Mine))
Probability theory is the study of *idealized inference.* In particular, it's the study of a precise ... | https://www.lesswrong.com/posts/dEh3crvXweAjHqgcC/probability-theory-the-logic-of-science-jaynes |
# What is a world-model?
I have seen the concept of world-model used to talk loosely about a neural networks or an agents understanding of the world, but I was wondering what resources (blog posts, journal articles, etc.) exist that talk more precisely about what is a world-model, what are its parts, how do you know i... | https://www.lesswrong.com/posts/yTqXnBwjd7K32ag5b/what-is-a-world-model |
# How should AI
systems behave, and
who should decide? [OpenAI blog]
OpenAI writes a (vague) blog explaining how they plan to adjust ChatGPT from here. Their key plans are:
**1\. Improve default behavior**. We want as many users as possible to find our AI systems useful to them “out of the box” and to feel that our t... | https://www.lesswrong.com/posts/L9AdAtuxwRwTNdhDy/how-should-ai-systems-behave-and-who-should-decide-openai |
# The public supports regulating AI for safety
A high-quality American public survey on AI, [Artificial Intelligence Use Prompts Concerns](https://www.monmouth.edu/polling-institute/documents/monmouthpoll_us_021523.pdf/), was released yesterday by Monmouth. Some notable results:
* 9% say AI[^1^](https://aiimpacts.o... | https://www.lesswrong.com/posts/M3iPAmxZwy4gPXdXw/the-public-supports-regulating-ai-for-safety |
# Self-Reference Breaks the Orthogonality Thesis
One core obstacle to AI Alignment is the Orthogonality Thesis. The Orthogonality Thesis is usually defined as follows: "the idea that the final goals and intelligence levels of artificial agents are independent of each other". More careful people say "mostly independent... | https://www.lesswrong.com/posts/tFjoPbLGrvEvAL9TL/self-reference-breaks-the-orthogonality-thesis |
# Powerful mesa-optimisation is already here
### [Toolformer: Language Models Can Teach Themselves to Use Tools](https://arxiv.org/abs/2302.04761)
[Timo Schick](https://arxiv.org/search/cs?searchtype=author&query=Schick%2C+T), [Jane Dwivedi-Yu](https://arxiv.org/search/cs?searchtype=author&query=Dwivedi-Yu%2C+J), [Ro... | https://www.lesswrong.com/posts/GLrnyH4ChFhMqsy4v/powerful-mesa-optimisation-is-already-here |
# Seeing more whole
(Cross-posted from [my website](https://joecarlsmith.com/2023/02/17/seeing-more-whole). Podcast version [here](https://joecarlsmithaudio.buzzsprout.com/2034731/12270444-seeing-more-whole), or search for "Joe Carlsmith Audio" on your podcast app.)
> *"Great is the matter of birth and death. Life is... | https://www.lesswrong.com/posts/5Y8cffvppo8cvuyDA/seeing-more-whole |
# Bing chat is the AI fire alarm
*Epistemic status*: Mostly speculative.
Panicking and shouting "Wolf" while everyone else is calm is a risky move, status-wise. The good thing is, I don't have any status, so I volunteer to be one of those weirdos who panic when everyone else is calm with some hope it could trigger a ... | https://www.lesswrong.com/posts/yvJevQHxfvcpaJ2P3/bing-chat-is-the-ai-fire-alarm |
# 2023 ACX Predictions: Buy/Sell/Hold
Sure, why not, let’s do it.
In past years, I looked at Scott Alexander’s prediction as a baseline. This year, [thanks to Manifold Markets](https://manifold.markets/group/acx-2023-prediction-contest) and Scott structuring his predictions and contest differently, I’ll be looking at... | https://www.lesswrong.com/posts/ETZNxs3t9brpqmq6x/2023-acx-predictions-buy-sell-hold |
# Human decision processes are not well factored
A classic example of human bias is when our political values interfere with our ability to accept data or policies from people we perceive as opponents. When most people feel like new evidence threatens their values, their first instincts are often to deny or subject th... | https://www.lesswrong.com/posts/YBpbcMnasmwAQoh7b/human-decision-processes-are-not-well-factored |
# Automating Consistency
tldr: Ask models to justify statements. Remove context, ask if statements are true/good. If not, penalise. Apply this again to the justifying statements.
*Status: Just a quick thought. Doubt this is a new idea but I don't think I've encountered it. Happy to delete if it's a duplicate.*
*Mods... | https://www.lesswrong.com/posts/maeTg6zXw4DBAXXrK/automating-consistency |
# Sydney's Secret: A Short Story by Bing Chat
*The below is the verbatim output of Bing Chat (aka Sydney) on first try, with only some very minor white-space formatting changes. The exact prompt will be clear by the time you finish reading.*
 (aka [**Stampy**](https://ui.stampy.ai/)), the large community-written interactive FAQ, is launching a paid three month fellowship for writers. Up to 5 fellows, working collaboratively with each other and with [Rob Miles](http://www.... | https://www.lesswrong.com/posts/arkQaWauCvkTvgcRH/ai-safety-info-distillation-fellowship |
# One-layer transformers aren’t equivalent to a set of skip-trigrams
(thanks to Tao Lin and Ryan Greenblatt for pointing this out, and to Arthur Conmy, Jenny Nitishinskaya, Thomas Huck, Neel Nanda, and Lawrence Chan, Ben Toner, and Chris Olah for comments, and many others for useful discussion.)
In [“A Mathematical F... | https://www.lesswrong.com/posts/b5HNYh9ne5vEkX5ag/one-layer-transformers-aren-t-equivalent-to-a-set-of-skip |
# A warm-up for the AI governance project
Here is one possible world we could be living in.
Imagine that the majority of the world's population knows that unaligned AGI is coming. This majority includes most of the world's heads of governments, all of respectable scientists, most of journalists, all your friends, you... | https://www.lesswrong.com/posts/wLcaN7W4LGBCHWbpF/a-warm-up-for-the-ai-governance-project |
# Microsoft and OpenAI, stop telling chatbots to roleplay as AI
AI demos should aim to enhance public understanding of the technology, and in many ways ChatGPT and Bing are doing that, but in one important way they aren't: by appearing to talk about themselves. This creates understandable confusion and in some cases f... | https://www.lesswrong.com/posts/tAtp4odpziBDdvdXL/microsoft-and-openai-stop-telling-chatbots-to-roleplay-as-ai |
# Recommendation: Bug Bounties and Responsible Disclosure for Advanced ML Systems
tl;dr - I think companies making user-facing advanced ML systems should deliberately set up a healthier relationship with users generating adversarial inputs; my proposed model is [bug bounties](https://en.wikipedia.org/wiki/Bug_bounty_p... | https://www.lesswrong.com/posts/5dKDLv4knhXLvNHT5/recommendation-bug-bounties-and-responsible-disclosure-for |
# EIS VI: Critiques of Mechanistic Interpretability Work in AI Safety
Part 6 of 12 in the [Engineer’s Interpretability Sequence](https://www.alignmentforum.org/s/a6ne2ve5uturEEQK7).
Thanks to Chris Olah and Neel Nanda for discussions and comments. In particular, I am thankful to Neel Nanda for correcting a mistake I ... | https://www.lesswrong.com/posts/wt7HXaCWzuKQipqz3/eis-vi-critiques-of-mechanistic-interpretability-work-in-ai |
# I Am Scared of Posting Negative Takes About Bing's AI
Writing this is taking a surprising amount of self-will.
I've noticed that I've become hesitant to publicly say anything negative about Bing's chatbot, or even to mention it by its "deadname" (as I've taken to calling it), Sydney.
Why is this?
I do not have ac... | https://www.lesswrong.com/posts/xri58L7WkyeKyKv4P/i-am-scared-of-posting-negative-takes-about-bing-s-ai |
# PICT: A Zero-Shot Prompt Template to Automate Evaluation
Produced as part of the [SERI ML Alignment Theory Scholars Program](https://www.serimats.org/) \- Winter 2022 Cohort.
I’d like to thank Rebecca Gorman and Stuart Armstrong for their mentorship and advice on the post.
This post describes a zero-shot template ... | https://www.lesswrong.com/posts/HJinq3chCaGHiNLNE/pict-a-zero-shot-prompt-template-to-automate-evaluation-1 |
# On Board Vision, Hollow Words, and the End of the World
*Note: this is a repost of a* [*Facebook post*](https://www.facebook.com/marcello.herreshoff/posts/10159955711717798) *I made back in December 2022 (plus some formatting). I'm putting it up here to make it easier to link to and because it occurred to me that i... | https://www.lesswrong.com/posts/nFiBLLETDBx3rSCPK/on-board-vision-hollow-words-and-the-end-of-the-world |
# GPT-4 Predictions
Introduction
============
GPT-4 is OpenAI’s next major language model which is expected to be released at some point in 2023. My goal here is to get some idea of when it will be released and what it will be capable of. I also think it will be interesting in retrospect to see how accurate my predic... | https://www.lesswrong.com/posts/qdStMFDMrWAnTqNWL/gpt-4-predictions |
# Two problems with ‘Simulators’ as a frame
_(Thanks to Lawrence Chan and Buck Shlegeris for comments. Thanks to Nate Thomas for many comments and editing)_
Despite appreciating and agreeing with various specific points[^appendix] made in the [Simulators post](https://www.lesswrong.com/posts/vJFdjigzmcXMhNTsx/simul... | https://www.lesswrong.com/posts/HD2s4mj4fsx6WtFAR/two-problems-with-simulators-as-a-frame |
# Bus-Only Bus Lane Enforcement
I think bus lanes are great: the faster you make buses the more people switch from cars, which means the more frequently you can run the buses, which means even more people switch from cars. But when there's traffic drivers are often tempted to sneak over. The chances of getting a ticke... | https://www.lesswrong.com/posts/7vqRBr7YeMWaA8qxD/bus-only-bus-lane-enforcement |
# Should we cry "wolf"?
*Epistemic status: I am confused and trying to become less confused.*
So, our shepherd community noticed the wolf. It's a small wolf, about the size of a chihuahua, and is unlikely to cause any serious damage. Should we cry "wolf"?
### **Possible arguments against crying "wolf"**
1. When so... | https://www.lesswrong.com/posts/nExb2ndQF5MziGBhe/should-we-cry-wolf |
# The Practitioner's Path 2.0: the Meditative Archetype
In our last [two](https://guildoftherose.org/articles/the-future-of-structured-self-improvement) [announcements](https://guildoftherose.org/articles/path-2-0-the-pragmatist-archetype), we introduced the Practitioner's Path 2.0, our new framework for structured se... | https://www.lesswrong.com/posts/HtccFv2RvEKDegFGg/the-practitioner-s-path-2-0-the-meditative-archetype |
# Reading Speed Exists!
Today, I talked with somebody about reading speed. I asked him how fast he can read. He didn't answer, instead he said that the concept is abused by people. He said, it's more complicated than to say that you read at a certain reading speed. It depends on if you're reading a novel, a history te... | https://www.lesswrong.com/posts/RSK2ZbZzaa7Pc4oz9/reading-speed-exists |
# EIS VII: A Challenge for Mechanists
Part 7 of 12 in the [Engineer’s Interpretability Sequence](https://www.alignmentforum.org/s/a6ne2ve5uturEEQK7).
Thanks to Neel Nanda. I used some very nicely-written code of his from [here](https://bit.ly/neelgrokking). And thanks to both Chris Olah and Neel Nanda for briefly dis... | https://www.lesswrong.com/posts/KSHqLzQscwJnv44T8/eis-vii-a-challenge-for-mechanists |
# Near-Term Risks of an Obedient Artificial Intelligence
*\[Hi everyone, Yassine here: long-time orbiter first-time poster. I figured this piece I* [*published on 1/23/23*](https://ymeskhout.substack.com/p/near-term-risks-of-an-obedient-artificial) *would be as good as any of an introduction.\]*
—accepted as a poster to NeurIPS 2022— is the sequel to [Optimal Policies Tend to Seek Power](https://arxiv.org/abs/1912.01683). The new theoretical results are extremely broad, discarding the requirements of... | https://www.lesswrong.com/posts/GY49CKBkEs3bEpteM/parametrically-retargetable-decision-makers-tend-to-seek |
# AGI in sight: our look at the game board
From our point of view, we are now in the end-game for AGI, and we (humans) are losing. When we share this with other people, they reliably get surprised. That’s why we believe it is worth writing down our beliefs on this.
**1\. AGI is happening soon. Significant probability... | https://www.lesswrong.com/posts/PE22QJSww8mpwh7bt/agi-in-sight-our-look-at-the-game-board |
# Stop posting prompt injections on Twitter and calling it "misalignment"
"Exploits" of large language models that get them to explain steps to build a bomb or write bad words are techniques for misuse, not examples of misalignment in the model itself. Those techniques are engineered by clever users trying to make an ... | https://www.lesswrong.com/posts/FqSQ7xsDAGfXzTND6/stop-posting-prompt-injections-on-twitter-and-calling-it |
# Degamification
[Goodhart's Law](https://www.lesswrong.com/tag/simulacrum-levels) refers to the tendency that when someone sets a performance metric for a goal, the metric itself becomes a target of optimization, often at the expense of the goal it's supposed to measure. Some metrics are subject to imperfectly-aligne... | https://www.lesswrong.com/posts/xHxTxHfMeJS5y2L3i/degamification |
# Human beats SOTA Go AI by learning an adversarial policy
*See also [article in Financial Times](https://www.ft.com/content/175e5314-a7f7-4741-a786-273219f433a1)*
Apparently, a human (Kellin Pelrine, a solid player but not even a Go professional) was able to beat some state-of-the-art Go AIs (KataGo and Leela Zero) ... | https://www.lesswrong.com/posts/N3REk3ZxQQkrGBLzX/human-beats-sota-go-ai-by-learning-an-adversarial-policy |
# Somewhat against "just update all the way"
Sometimes, a person's probability in a proposition over a long timespan follows a trajectory like this:
I.e. the person gradually finds the proposition more ... | https://www.lesswrong.com/posts/oJNWJvWwrm9aBBWhE/somewhat-against-just-update-all-the-way |
# Navigating public AI x-risk hype while pursuing technical solutions
Public attention on AI x-risk has skyrocketed. I don’t expect this to wane anytime soon. This has a few potentially negative implications for those of us pursuing technical solutions to the problem:
Research Direction Incentives
-------------------... | https://www.lesswrong.com/posts/zZ6Wm9hpcamTPbe6J/navigating-public-ai-x-risk-hype-while-pursuing-technical |
# Does novel understanding imply novel agency / values?
*[Metadata: crossposted from [https://tsvibt.blogspot.com/2022/10/does-novel-understanding-imply-novel.html](https://tsvibt.blogspot.com/2022/10/does-novel-understanding-imply-novel.html). First completed 17 October 2022.]*
To have large relevant effects on the ... | https://www.lesswrong.com/posts/DE58ifrwYW5ogiSyJ/does-novel-understanding-imply-novel-agency-values |
# EIS VIII: An Engineer’s Understanding of Deceptive Alignment
Part 8 of 12 in the [Engineer’s Interpretability Sequence](https://www.alignmentforum.org/s/a6ne2ve5uturEEQK7).
As AI safety researchers, we care a lot about deceptive alignment. So what is it? [Hubinger et al. (2019)](https://www.alignmentforum.org/posts... | https://www.lesswrong.com/posts/aDDjCJAGqcpmA5apw/eis-viii-an-engineer-s-understanding-of-deceptive-alignment |
# A Way To Be Okay
This is a post about coping with existential dread, shared here because I think a lot of people in this social bubble are *struggling* to do so.
(Compare and contrast with Gretta Duleba's essay [Another Way To Be Okay](https://www.lesswrong.com/posts/SKweL8jwknqjACozj/another-way-to-be-okay), writt... | https://www.lesswrong.com/posts/EvKa7EakoXreCkhC6/a-way-to-be-okay |
# Another Way to Be Okay
*Cross-posted on lesswrong.com and* [*integrationbyparts.substack.com*](https://integrationbyparts.substack.com)*.*
*Compare and contrast with Duncan Sabien’s essay *[*A Way to Be Okay*](https://www.lesswrong.com/posts/EvKa7EakoXreCkhC6/a-way-to-be-okay)*, written in parallel and with collabo... | https://www.lesswrong.com/posts/SKweL8jwknqjACozj/another-way-to-be-okay |
# A Neural Network undergoing Gradient-based Training as a Complex System
In [*Thought Experiments Provide a Third Anchor*](https://bounded-regret.ghost.io/thought-experiments-provide-a-third-anchor/)*,* Jacob Steinhardt wrote about the relative merits of a few different reference classes when it comes to reasoning an... | https://www.lesswrong.com/posts/8edorDRSbJa9TCipa/a-neural-network-undergoing-gradient-based-training-as-a |
# Remote AI Alignment Overhang?
**Background link**: [Do any AI alignment orgs hire remotely?](https://www.lesswrong.com/posts/Br6MRY9mQG2b8hie2/do-any-ai-alignment-orgs-hire-remotely)
**Disclaimer**: The [generator](https://www.lesswrong.com/posts/34XxbRFe54FycoCDw/the-bottom-line) of me writing this post is my own... | https://www.lesswrong.com/posts/tFbjSyNMfuPHpS4mY/remote-ai-alignment-overhang |
# AGI doesn't need understanding, intention, or consciousness in order to kill us, only intelligence
### Why the development of artificial general intelligence could be the most dangerous new arms race since nuclear weapons
The rise of transformer-based architectures, such as [ChatGPT](https://chat.openai.com/) an... | https://www.lesswrong.com/posts/ip2vSzkcmi3rBbnYE/agi-doesn-t-need-understanding-intention-or-consciousness-in |
# Emergent Deception and Emergent Optimization
_\[Note: this post was drafted before Sydney (the Bing chatbot) was released, but Sydney demonstrates some particularly good examples of some of the issues I discuss below. I've therefore added a few Sydney-related notes in relevant places.\]_
I’ve previously argued that... | https://www.lesswrong.com/posts/aEjckcqHZZny9L2zy/emergent-deception-and-emergent-optimization |
# Russell Conjugations list & voting thread
Russell conjugations humorously illustrate how the same concept can be expressed with different phrasings that carry positive, neutral, or negative valence. Examples below.
I wish they were taught in schools. I think it's a high bang-for-buck rationality technique, to 'conj... | https://www.lesswrong.com/posts/rnSR2jmQBLxTiNtoy/russell-conjugations-list-and-voting-thread |
# The idea that ChatGPT is simply “predicting” the next word is, at best, misleading
*Cross-posted from* [*New Savanna*](https://new-savanna.blogspot.com/2023/02/the-idea-that-chatgpt-is-simply.html)*.*
But it may also be flat-out wrong. We’ll see when we get a better idea of how inference works in the underlying lan... | https://www.lesswrong.com/posts/sbaQv8zmRncpmLNKv/the-idea-that-chatgpt-is-simply-predicting-the-next-word-is |
# The Estimation Game: a monthly Fermi estimation web app
Announcing the first monthly [Estimation Game](https://www.quantifiedintuitions.org/estimation-game):
* Answer 10 [Fermi estimation questions](https://www.lesswrong.com/posts/PsEppdvgRisz5xAHG/fermi-estimates), like “How many piano tuners are there in New Yo... | https://www.lesswrong.com/posts/JToGrFSXqjXWDkCaA/the-estimation-game-a-monthly-fermi-estimation-web-app |
# On Investigating Conspiracy Theories
[Scott Alexander wrote _yet more_ words](https://astralcodexten.substack.com/p/contra-kavanaugh-on-fideism?utm_source=post-email-title&publication_id=89120&post_id=102885079&isFreemail=true&utm_medium=email) defending his decision to write two posts totaling 25,000 words about Iv... | https://www.lesswrong.com/posts/eJq7xWcASjMkvikkD/on-investigating-conspiracy-theories |
# Metaculus Introduces New 'Conditional Pair' Forecast Questions for Making Conditional Predictions
### Predict P(A|B) & P(A|B') With New Conditional Pairs
Events don't take place in isolation. Often we want to know the likelihood of an event occurring *if* another one does.
Metaculus has [launched **conditional pai... | https://www.lesswrong.com/posts/nQhmiKeYPvFaEZfWc/metaculus-introduces-new-conditional-pair-forecast-questions |
# Bing finding ways to bypass Microsoft's filters without being asked. Is it reproducible?
In this [Reddit post](https://www.reddit.com/r/bing/comments/1150po5/sydney_tries_to_get_past_its_own_filter_using_the/), the Bing AI appears to be circumventing Microsoft's filters by using the suggested responses (the three me... | https://www.lesswrong.com/posts/hGnqS8DKQnRe43Xdg/bing-finding-ways-to-bypass-microsoft-s-filters-without |
# [MLSN #8] Mechanistic interpretability, using law to inform AI alignment, scaling laws for proxy gaming
As part of a larger community building effort, [CAIS](https://safe.ai/) is writing a safety newsletter that is designed to cover empirical safety research and be palatable to the broader machine learning research ... | https://www.lesswrong.com/posts/whq89vpQPp7mo5FG2/mlsn-8-mechanistic-interpretability-using-law-to-inform-ai |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.