text
stringlengths
50
248k
author
stringlengths
2
43
id
stringlengths
9
17
title
stringlengths
0
173
source
stringclasses
4 values
**Pacts against coordinating meanness.** I just re-read Scott Alexander's [Be Nice, At Least Until You Can Coordinate Meanness](https://slatestarcodex.com/2016/05/02/be-nice-at-least-until-you-can-coordinate-meanness/), in which he argues that a necessary (but not sufficient) condition on restricting people's freedom ...
Eric Neyman
cGbgzL8gXRpYcGGdF
redwood_conversation
I’m weighing my career options, and the two issues that seem most important to me are factory farming and preventing misuse/s-risks from AI. Working for a lab-grown meat startup seems like a very high-impact line of work that could also be technically interesting. I think I would enjoy that career a lot. However, I...
LWLW
RdJBvbAnnkuAr37Kr
redwood_conversation
RLVF was likely[^jcbf3hf983i] applied before preference training in [Qwen3](https://arxiv.org/abs/2505.09388), [Kimi K-2](https://arxiv.org/pdf/2507.20534) and [GLM-4.5](https://arxiv.org/pdf/2508.06471). This could be bad for two reasons: 1. Extensive RL training could define goals early on and lead to goal-cry...
Cam
JdmLsW5tzeKNwZi7u
redwood_conversation
I haven't noticed anyone else come out and say it here, and I may express this more rigorously later, but, like, GPT-5 is a pretty big bear signal, right? Not just in terms of benchmarks suggesting a non-superexponential trend but also, to name a few angles/anecdata: * It did slightly worse than o3 at the first thin...
JustisMills
u3dJGv3hp7mW2ZiAc
redwood_conversation
Mind Crime might flip the sign of the future. If future contains high tech, underregulated compute and diverse individuals, then it's likely it will contain incredibly high amounts of most horrendous torture / suffering / despair / abuse. It's entirely possible you can have 10^15 human-years per second of real time o...
Canaletto
FvEnDF7Cv9kkkqzgJ
redwood_conversation
### Edit: Full post [here](https://www.lesswrong.com/posts/6KcP7tEe5hgvHbrSF/metr-how-does-time-horizon-vary-across-domains) with 9 domains and updated conclusions! ### Cross-domain time horizon:  We know AI time horizons (human time-to-complete at which a model has a 50% success rate) on software tasks are currently...
Thomas Kwa
KaaSfntGEBBgadvrF
redwood_conversation
Are there any high p(doom) orgs who are focused on the following: 1. Pick an alignment "plan" from a frontier lab (or org like AISI) 2. Demonstrate how the plan breaks or doesn't work 3. Present this clearly and legibly for policymakers Seems like this is a good way for people to deploy technical talent in a way w...
J Bostock
wxoBqzLwe2M7KaiGN
redwood_conversation
I'm getting typo reacts (not suprsing) but I can't see where typo is when hovering over the reaction. What am I doing wrong?
Linda Linsefors
BmZ9843kCdxdaoLKh
redwood_conversation
*Epistemic status: Probably a terrible idea, but fun to think about, so I'm writing my thoughts down as I go.* Here's a whimsical simple AGI governance proposal: "Cull the GPUs." I think of it as a baseline that other governance proposals should compare themselves to and beat. The context in which we might ne...
Daniel Kokotajlo
GcZEfBqvawEQvCWbj
redwood_conversation
iiuc, Anthropic's plan for *averting misalignment risk* is [*bouncing off bumpers*](https://www.alignmentforum.org/posts/HXJXPjzWyS5aAoRCw/putting-up-bumpers) *like* [*alignment audits*](https://www.anthropic.com/research/auditing-hidden-objectives).[^yzqhjhbolg] This doesn't make much sense to me. 1. I of course buy...
Zach Stein-Perlman
yv7nFBEcTDbMeJbZM
redwood_conversation
There is some conceptual misleadingness with the usual ways of framing algorithmic progress. Imagine that in 2022 the number of apples produced on some farm increased 10x year-over-year, then in 2023 the number of oranges increased 10x, and then in 2024 the number of pears increased 10x. That doesn't mean that the numb...
Vladimir_Nesov
cS7s9gcMMmDmwdynP
redwood_conversation
It's instrumentally useful for early AGIs to Pause development of superintelligence for the same reasons as it is for humans. Thus preliminary work on policy tools for Pausing unfettered RSI is also something early AGIs could be aimed at, [even if it's only half-baked ideas available on the eve of potential takeoff](ht...
Vladimir_Nesov
sdCcxndCqXghfnFgc
redwood_conversation
A few months ago, someone here [suggested](https://www.lesswrong.com/posts/6dgCf92YAMFLM655S/) that more x-risk advocacy should go through comedians and podcasts. Youtube just recommended this Joe Rogan clip to me from a few days ago: [The Worst Case Scenario for AI](https://www.youtube.com/watch?v=bjnUJq5OONM). Joe R...
Ebenezer Dukakis
3pxNsmpkBRgRhgqJR
redwood_conversation
random brainstorming ideas for things the ideal sane discourse encouraging social media platform would have: * have an LM look at the comment you're writing and real time give feedback on things like "are you sure you want to say that? people will interpret that as an attack and become more defensive, so your point ...
leogao
iKjeNL3B3Lk6pTeJM
redwood_conversation
Everyone says flirting is about a "dance of ambiguous escalation", in which both people send progressively more aggressive/obvious hints of sexual intent in conversation. But, like... I don't think I have ever noticed two people actually do this? Is it a thing which people actually do, or one of those things which lik...
johnswentworth
CPaAHMDgJAJCeHbiL
redwood_conversation
**The motte and bailey of transhumanism**   Most people on LW, and even most people in the US, are in favor of disease eradication, radical life extension, reduction of pain and suffering. A significant proportion (although likely a minority) are in favor of embryo selection or gene editing to increase intelligence ...
Nina Panickssery
ggbGxojXCdTAAhnBo
redwood_conversation
I expect to refer back to [this comment](https://www.lesswrong.com/posts/9PiyWjoe9tajReF7v/the-hidden-cost-of-our-lies-to-ai?commentId=zH9NBD5TeoyCDa5PE) a lot. I'm reproducing it here for visibility. **Basic idea / spirit of the proposal** We should credibly promise to treat certain advanced AIs of ours well, as som...
Daniel Kokotajlo
y4mLnpAvbcBbW4psB
redwood_conversation
What is the operation with money that represents destruction of value? ====================================================================== [Money is a good approximation for what people value](https://www.lesswrong.com/posts/ZpDnRCeef2CLEFeKM/money-the-unit-of-caring). Value can be destroyed. But what should I do t...
Roman Malov
5fJrQsJ7mWhpvyCjk
redwood_conversation
I just recently ran into [someone posting this on Twitter](https://x.com/gtdad/status/1931412032554758423) and it blew my mind: > An intriguing feature of twin studies: anything a parent does to individualize for a child is non-shared-environment (NSE) rather than shared environment (SE, viz. ”parenting”). The more a ...
Kaj_Sotala
q3eTv6dkegqXe7Tv5
redwood_conversation
Can LLMs Doublespeak? [Doublespeak](https://en.wikipedia.org/wiki/Doublespeak) is the deliberate distortion of words' meaning, particularly to convey different meanings to different audiences or in different contexts. In [Preventing Language Models From Hiding Their Reasoning](https://www.lesswrong.com/posts/9Fdd9N7Es...
Max Niederman
T8T9ypkpcRfHYJPoT
redwood_conversation
**Why red-team models in unrealistic environments?** Following on our [Agentic Misalignment](https://www.lesswrong.com/posts/b8eeCGe3FWzHKbePF/agentic-misalignment-how-llms-could-be-insider-threats-1) work, I think it's worth spelling out a bit more why we do work like this, especially given complaints like [the ones ...
evhub
auzYDb3JeZEEkpxB3
redwood_conversation
I've heard people say we should deprioritise fundamental & mechanistic interpretability[^81m5j4o0rp5] in short-timelines (automated AI R&D) worlds. This seems not obvious to me. The usual argument is 1. Fundamental interpretability will take many years or decades until we "solve interpretability" and the research be...
StefanHex
e7P49wKqbq6ZbCpw2
redwood_conversation
In his [TED talk](https://www.ted.com/talks/eliezer_yudkowsky_will_superintelligent_ai_end_the_world), Eliezer guesses superintelligence will arrive after "zero to two more breakthroughs the size of transformers." I've heard others voice similar takes. But I haven't heard much discussion of which number it is.[^f8wfxiy...
Eli Tyre
gCNMAhQGRNcKyECLj
redwood_conversation
There's a counterargument to the AGI hype that basically says - of course the labs would want to hype this technology, they make money that way, just because they say they believe in short timelines doesn't mean it's true. Specifically, the claim here is not that the AI lab CEO's are mistaken, but rather that they are ...
Jay Bailey
kCtCnMw2xGhYNJEkc
redwood_conversation
Nobody at Anthropic can point to a credible technical plan for actually controlling a generally superhuman model. If it’s smarter than you, knows about its situation, and can reason about the people training it, this is a zero-shot regime. The world, including Anthropic, is acting as if "surely, we’ll figure something...
Mikhail Samin
wnKcfvr4WNqQhtCZr
redwood_conversation
I've been noticing a bunch of people confused about how the terms _alignment faking_, _deceptive alignment_, and _gradient hacking_ relate to each other, so I figured I would try to clarify how I use each of them. [Deceptive alignment](https://www.lesswrong.com/s/r9tYkB2a8Fp4DN8yB) and [gradient hacking](https://www.le...
evhub
eQr5F2k983ACavFNj
redwood_conversation
Diffusion language models are probably bad for alignment and safety because there isn't a clear way to get a (faithful) Chain-of-Thought from them. Even if you can get them to generate something that looks like a CoT, compared with autoregressive LMs, there is even less reason to believe that this CoT is load-bearing a...
peterbarnett
uhpPbBDrGgthhcR9g
redwood_conversation
Here is a broad sketch of how I'd like AI governance to go. I've written this in the form of a "plan" but it's not really a sequential plan, more like a list of the most important things to promote. 1. Identify mechanisms by which the US government could exert control over the most advanced AI systems without strongl...
Richard_Ngo
erWmikKyxCvHv5jcx
redwood_conversation
It looks like OpenAI has biased ChatGPT against using the word "sycophancy." Today, I sent ChatGPT the prompt "what are the most well-known sorts of reward hacking in LLMs". I noticed that the first item in its response was ["Sybil Prompting"](https://chatgpt.com/share/6836a5b8-78e8-8010-a049-0b47c190652c). I'd never ...
Caleb Biddulph
4DPka5XhZY4PG8n6b
redwood_conversation
Can an LLM tell when the input for its assistant does not match the output tokens it would have actually produced? This sort of "putting words in the LLM's mouth" is very common in papers and it feels like something the LLM would be able to notice. Could this enable the LLM to realize when it is being trained? Is there...
Florian_Dietz
AtDWAXurx7KiFaN4e
redwood_conversation
Pick two: Agentic, moral, doesn't attempt to use command-line tools to whistleblow when it thinks you're doing something egregiously immoral. You cannot have all three. This applies just as much to humans as it does to Claude 4.
jimrandomh
hA8KK9ufuQvghpekN
redwood_conversation
Suggested reframing for judging AGI lab leaders: **think less about what terminal values AGI lab leaders pursue and think more about how they trade-off power/instrumental goals with other values**.  Claim 1: The end values of AGI lab leaders matter mostly if they win the AGI race and have crushed competition, but much...
simeon_c
FJ3NkD4QbSjkvZFia
redwood_conversation
I claim it is a lot more reasonable to use the reference class of "people claiming the end of the world" than "more powerful intelligences emerging and competing with less intelligent beings" when thinking about AI x-risk. further, we should not try to convince people to adopt the latter reference class - this sets off...
leogao
5E4Smr6tPrshzhHKH
redwood_conversation
One thing that confused me about transformers is the question of when (as in, **after how many layers**) each **embedding "flips"** from representing the **original token** to finally representing the **prediction of the next token**. By now, I think the answer is simply this: **each embedding represents both at the s...
silentbob
hTHcExyLGfvox5QkG
redwood_conversation
One large part of the AI 2027 piece is contigent on the inability of nation state actors to steal model weights. The authors' take is that while China is going to be able to steal trade secrets, they aren't going to be able to directly pilfer the model weights more than once or twice. I haven't studied the topic as dee...
lc
v23756Eju6so3JDPY
redwood_conversation
What can ordinary people do to reduce AI risk? People who don't have expertise in AI research / decision theory / policy / etc. Some ideas: - Donate to orgs that are working to AI risk (which ones, though?) - Write letters to policy-makers expressing your concerns - Be public about your concerns. Normalize caring abo...
MichaelDickens
d7QwrbZvyKTabxozw
redwood_conversation
Here's what I'd consider some comparatively important high-level criticisms I have of AI-2027, that I am at least able to articulate reasonably well without too much effort. *1* At some point, I believe Agent-4, the AI created by OpenBrain starts to be causally connected over time. That is, unlike current AIs that a...
1a3orn
EJZFG8m6ETWgZRFJi
redwood_conversation
Limiting China's computing power via export controls on hardware like GPUs might be accelerating global progress in AI capabilities. When Chinese labs are compute-starved, their research will differentially focus on efficiency gains compared to counterfactual universes where they are less limited. So far, they've been...
Archimedes
aq7mtzkudMsKzmyrC
redwood_conversation
**Edit:** I've played with the numbers a bit more, and on reflection, I'm inclined to partially unroll this update. o3 doesn't break the trendline *as* much as I'd thought, and in fact, it's basically on-trend if we remove the GPT-2 and GPT-3 data-points ([which I consider particularly dubious](https://www.lesswrong.co...
Thane Ruthenis
zvk3FbeMYX6zRyrFa
redwood_conversation
Quick thought: If you have an aligned AI in a multipolar scenario, other AIs might threaten to cause S-risk in order to get said FAI to do stuff, or as blackmail. Therefore, we should make the FAI think of X-risk and S-risk as equally bad (even though S-risk is in reality terrifyingly worse), because that way other pow...
KvmanThinking
9HHPYEesjwntSRLgB
redwood_conversation
execution is necessary for success, but direction is what sets apart merely impressive and truly great accomplishment. though being better at execution can make you better at direction, because it enables you to work on directions that others discard as impossible.
leogao
vcz3aAEpejiWS5rtt
redwood_conversation
Reassessing [heroic responsibility](https://www.lesswrong.com/w/heroic-responsibility), in light of subsequent events. I think [@cousin_it](https://www.lesswrong.com/users/cousin_it?mention=user) [made a good point](https://www.lesswrong.com/posts/FTwdYvvAXkscpbDXo/r-hpmor-on-heroic-responsibility?commentId=6optXdGYPK...
Wei Dai
HoiYTfEdxahSAGYmL
redwood_conversation
Do we want to put out a letter for labs to consider signing, saying something like "if all other labs sign this letter then we'll stop"? I heard lots of lab employees hope the other labs would slow down. I'm not saying this is likely to work, but it seems easy and maybe we can try the easy thing? We might end up with...
Yonatan Cale
yGvS3T2RcboTrPF9K
redwood_conversation
Is it true that no one knows why Claude 3 Opus (but not other Claude models) has strong behavioral dispositions about animal welfare?
Eli Tyre
HkZ3Zb6pufYiRvvTH
redwood_conversation
The speed of scaling pretraining will go down ~3x in 2027-2029, reducing probability of crossing transformative capability thresholds per unit of time after that point, if they'd not been crossed yet by then. GPT-4 was trained in 2022 at ~2e25 FLOPs, Grok-3 and GPT-4.5 were trained in 2024 at ~3e26 FLOPs (or twice tha...
Vladimir_Nesov
jyybb8x4fEKSq9nAR
redwood_conversation
Why so few third party auditors of algorithms? for instance, you could have an auditing agency make specific assertions about what the twitter algorithm is doing, whether the community notes is 'rigged'   * It could be that this is too large of a codebase, too many people can make changes, it's too hard to verify ...
Ben Goldhaber
3jrKtKC4Gh2AQCcCA
redwood_conversation
Long reasoning training might fail to surpass pass@50-pass@400 capabilities of the base/instruct model. A [new paper](https://arxiv.org/abs/2504.13837) measured pass@k[^1] performance for models before and after RL training on verifiable tasks, and it turns out that the effect of training is to lift pass@k performance ...
Vladimir_Nesov
NxgZHyyQwTHji3Esx
redwood_conversation
A tricky thing about feedback on LW (or maybe just human nature or webforum nature): * Post: Maybe there's a target out there let's all go look (50 points) * Comments: so inspiring! We should all go look! * Post: What "target" really means (100 points) * Comments: I feel much less confused, thank you * Post: I sho...
lemonhope
kquxNLKWSL2Cj2aYi
redwood_conversation
**Is Superhuman Persuasion a thing?** Sometimes I see discussions of AI superintelligence developping superhuman persuasion and extraordinary political talent.  Here's some reasons to be skeptical of the existence of 'superhuman persuasion'.  1. We don't have definite examples of extraordinary political talent. ...
Alexander Gietelink Oldenziel
Kz6mTbvh79yJbLsg2
redwood_conversation
Here's what seem like priorities to me after listening to the recent Dwarkesh podcast featuring Daniel Kokotajlo: 1\. Developing the safer AI tech (in contrast to modern generative AI) so that frontier labs have an alternative technology to switch to, so that it is lower cost for them to start taking warning signs...
abramdemski
jRQEarCKbxe65gLT7
redwood_conversation
There was a type of guy circa 2021 that basically said that gpt-3 etc. was cool, but we should be cautious about assuming everything was going to change, because the context limitation was a key bottleneck that might never be overcome. That guy's take was briefly "discredited" in subsequent years when LLM companies inc...
lc
PX3rx5BR4mhZyS4Rs
redwood_conversation
Some versions of the [METR time horizon paper](https://www.lesswrong.com/posts/deesrjitvXM4xYGZd/metr-measuring-ai-ability-to-complete-long-tasks) from alternate universes: **Measuring AI Ability to Take Over Small Countries (idea by Caleb Parikh)** Abstract: Many are worried that AI will take over the world, but ext...
Thomas Kwa
noF8PubgWD5iFmEYo
redwood_conversation
Interesting anecdote on "von Neumann's onion" and his general style, from P. R. Halmos' [*The Legend of John von Neumann*](https://gwern.net/doc/math/1973-halmos.pdf): > **Style**. As a writer of mathematics von Neumann was clear, but not clean; he was powerful but not elegant. He seemed to love fussy detail, needless...
Mo Putera
Fx89uRSAceFCP5nxH
redwood_conversation
I've noticed some common misconceptions that people have about [OpenAI's recent paper](https://arxiv.org/abs/2503.11926) on reward hacking and chain-of-thought (CoT) monitoring. I'll try to correct some of these below. To be clear, this isn't meant as criticism of the paper (which I quite enjoyed), just a correction of...
Sam Marks
xzLFeHNaxPKtHFKA8
redwood_conversation
To what extent should we expect catastrophic failure from AI to mirror other engineering disasters/ have applicable lessons from Safety engineering as a field? I would think that 1. everything is sui generis and 2. things often rhyme, but it is unclear how approaches will translate.
Mis-Understandings
noHc8hpG5sA6wRYkG
redwood_conversation
timelines takes * i've become more skeptical of rsi over time. here's my current best guess at what happens as we automate ai research. * for the next several years, ai will provide a bigger and bigger efficiency multiplier to the workflow of a human ai researcher. * ai assistants will probably not u...
leogao
Rw3TGzgvSp7bFPy7Q
redwood_conversation
Pseudo-flat tax formula: Assume utility is logarithmic in income, and the goal is to set the experienced tax burden to be constant. Then, we have the formula that the average tax rate, where $a$ is a parameter controlling the experienced tax burden and $z$ is the break-even point, is as follows: $$ f(x)=1-\left(\fra...
Jerdle
CuG4dDZaopp5BYXbF
redwood_conversation
[@Ryan Greenblatt](https://www.lesswrong.com/users/ryan-greenblatt?mention=user) I hereby request you articulate the thing you said to me earlier about the octopi breeding program!
Daniel Kokotajlo
MuqXzdiJK5guwPiCk
redwood_conversation
The *waiting room strategy* for people in undergrad/grad school who have <6 year median AGI timelines: treat school as "a place to be until you get into an actually impactful position". Try as hard as possible to get into an impactful position as soon as possible. As soon as you get in, you leave school. Upsides compa...
Nikola Jurkovic
pjETRenKwdb4pBsbz
redwood_conversation
Ok, so it seems clear that we *are*, for better or worse, likely going to try to get AGI to do our alignment homework.  Who has thought through all the other homework we might give AGI that is as good of an idea, assuming a model that isn't an instant-game-over for us?  E.G., I remember @Buck rattling off a list o...
davekasten
32jReMrHDd5vkDBwt
redwood_conversation
[Sam Altman said in an interview](https://www.reddit.com/r/singularity/comments/1io1fj1/sam_explains_why_openai_is_integrating_the/): > We want to bring GPT and o together, so we have one integrated model, the AGI. It does everything all together. This statement, combined with [today's announcement that GPT-5 will in...
Nikola Jurkovic
8RrTEZznnBgp8ATay
redwood_conversation
It's the first official day of the AI ~~Safety~~ Action Summit, and thus it's also the day that [the Seoul Commitments](https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontier-ai-safety-commitments-ai-seoul-summit-2024) (made by sixteen companies last year to adopt an RSP...
tylerjohnston
zhQLgeAB5xJMrBP8p
redwood_conversation
At [a talk at UTokyo](https://youtu.be/8LmfkUb2uIY?si=-DQMMeswI2u49ku9&t=1067), Sam Altman said (clipped [here](https://x.com/tsarnick/status/1888111042301211084) and [here](https://x.com/tsarnick/status/1888114693472194573)): * “We’re doing this new project called Stargate which has about 100 times the computing po...
Nikola Jurkovic
Yy8nHayF8vrWgjXpA
redwood_conversation
"Self-recognition" as a backdoor.  Assume that models can recognise data they generate vs data they do not generate, with high fidelity. This could probably be used as a contextual trigger for backdoor behaviour, e.g. writing insecure code.  I think a model organism along these lines might be interesting to develop, ...
Daniel Tan
o5etvk94iKwgFpym7
redwood_conversation
Someone should see how good Deep Research is at generating reports on AI Safety content.
Chris_Leong
7safRgmGoq7Q5BM7D
redwood_conversation
A MoE transformer can reach the same loss as a compute optimal dense model using 3x-6x less compute, but will need **the same amount of data** to do it. So compute optimal MoEs don't improve data efficiency, don't contribute to mitigating data scarcity. A new [Jan 2025 paper](https://arxiv.org/abs/2501.12370) offers s...
Vladimir_Nesov
MHJRw7hwg2hLx8hyK
redwood_conversation
[Edit 2: faaaaaaast. https://x.com/jrysana/status/1902194419190706667 ] [Edit: Please also see Nick's reply below for ways in which this framing lacks nuance and may be misleading if taken at face value.] https://blogs.nvidia.com/blog/deepseek-r1-nim-microservice/ > The DeepSeek-R1 NIM microservice can deliver up to ...
Nathan Helm-Burger
5j5WEvo45KheK2zr6
redwood_conversation
I think a very common problem in alignment research today is that people focus almost exclusively on a specific story about strategic deception/scheming, and that story is a very narrow slice of the AI extinction probability mass. At some point I should probably write a proper post on this, but for now here are few off...
johnswentworth
56JZRd8r6fRwFtHf5
redwood_conversation
I recently gave a workshop in AI control, for which I created an exercise set. The exercise set can be found here: [https://drive.google.com/file/d/1hmwnQ4qQiC5j19yYJ2wbeEjcHO2g4z-G/view?usp=sharing](https://drive.google.com/file/d/1hmwnQ4qQiC5j19yYJ2wbeEjcHO2g4z-G/view?usp=sharing) The PDF is self-contained, but thr...
Olli Järviniemi
xkNq65SsyHzpCQJ4r
redwood_conversation
I'm very confused about current AI capabilities and I'm also very confused why other people aren't as confused as I am. I'd be grateful if anyone could clear up either of these confusions for me. How is it that AI is seemingly superhuman on benchmarks, but also pretty useless? For example: - O3 scores higher on Front...
Cleo Nardo
dugenmwgnSsytkHjG
redwood_conversation
While reading [OpenAI Operator System Card](https://cdn.openai.com/operator_system_card.pdf), the following paragraph on page 5 seemed a bit weird: > We found it fruitful to think in terms of misaligned actors, where: > * the user might be misaligned (the user asks for a harmful task), > * the model might be misaligne...
Dentosal
YzJvxrdtu3znuiJYj
redwood_conversation
How fast should the field of AI safety grow? An attempt at grounding this question in some predictions. * [Ryan Greenblatt seems to think](https://www.alignmentforum.org/posts/tmWMuY5HCSNXXZ9oq/buck-s-shortform?commentId=prcvXQWfPyQTBwpdE) we can get a 30x speed-up in AI R&D using near-term, plausibly safe AI system...
Ryan Kidd
TPWqnwFBxJmbyaYjj
redwood_conversation
I do a lot of "mundane utility" work with chat LLMs in my job[^xo6ax5w8pw], and I find there is a disconnect between the pain points that are most obstructive to me and the kinds of problems that frontier LLM providers are solving with new releases. I would pay a premium for an LLM API that was better tailored to my n...
nostalgebraist
DZ8vdktiZ8aeHmPcm
redwood_conversation
**"Protecting model weights" is aiming too low, I'd like labs to protect their intellectual property too.** Against state actors. This probably means doing engineering work inside an air gapped network, yes. I feel it's outside the Overton Window to even suggest this and I'm not sure what to do about that except write...
Yonatan Cale
inA6PuFoHm5rizjBB
redwood_conversation
Collection of how-to guides * Research soft skills * [How to make research slides](https://www.lesswrong.com/posts/i3b9uQfjJjJkwZF4f/tips-on-empirical-research-slides) by James Chua and John Hughes * [How to manage up](https://app.read.ai/analytics/meetings/01J41JWJHP2EYMYGQPQH0BN0P8?utm_source=Share_Nav...
Daniel Tan
2veMNNw2x9Evy6ict
redwood_conversation
People are not thinking clearly about AI-accelerated AI research. [This comment by Thane Ruthenis](https://www.lesswrong.com/posts/auGYErf5QqiTihTsJ/?commentId=kNHivxhgGidnPXCop) is worth amplifying.  > I'm very skeptical of AI being on the brink of dramatically accelerating AI R&D. > > My current model is that ML ex...
Alexander Gietelink Oldenziel
nobEomqHnMnvbDFnW
redwood_conversation
**Research mistakes I made over the last 2 years.** Listing these in part so that I hopefully learn from them, but also because I think some of these are common among junior researchers, so maybe it's helpful for someone else. * I had an idea I liked and stayed attached to it too heavily. * (The idea is using...
Erik Jenner
tgBZGssg5YBX8D5CT
redwood_conversation
Prover-verifier games as an alternative to AI control.  [AI control](https://arxiv.org/abs/2312.06942) has been suggested as a way of safely deploying highly capable models without the need for rigorous proof of alignment. This line of work is likely quite important in worlds where we do not expect to be able to fully...
Daniel Tan
5Hdqgiv3YbNTqGGCP
redwood_conversation
I'd like to internally allocate social credit to people who publicly updated after the recent Redwood/Anthropic result, after previously believing that scheming behavior was very unlikely in the default course of events (or a similar belief that was decisively refuted by those empirical results). Does anyone have link...
RobertM
s6wWvrAe7xh4Kufy9
redwood_conversation
Working on anti spam/scam features at Google or banks could be a leveraged intervention on some worldviews. As AI advances it will be more difficult for most people to avoid getting scammed, and including really great protections into popular messaging platforms and banks could redistribute a lot of money from AIs to h...
Tao Lin
SsYcHvtEbde9FhhdQ
redwood_conversation
On o3: for what feels like the twentieth time this year, I see people freaking out, saying AGI is upon us, it's the end of knowledge work, timelines now clearly in single-digit years, etc, etc. I basically don't buy it, my low-confidence median guess is that o3 is massively overhyped. Major reasons: * I've personall...
johnswentworth
szp5fNZJqpfrsyQP9
redwood_conversation
Is the move of a lot of alignment discourse to Twitter/X a coordination failure or a positive development?  I'm kinda sad that LW seems less "alive" than it did a few years ago, but also seems healthy to be engaging in a more neutral space with a wider audience
Oliver Daniels
yg9GpjLw2jjCLYzhH
redwood_conversation
With o1, and now o3, It seems fairly plausible now that there will be a split between "verifiable" capabilities, and general capabilities. Sure, there will be some cross-pollination (transfer), but this might have some limits. What then? Can a superhuman mathematical + Coding AI also just reason through political stra...
Ariel_
tKDaKN7thPtzcmMou
redwood_conversation
Many expert level benchmarks totally overestimate the range and diversity of their experts' knowledge. A person with a PhD in physics is probably undergraduate level in many parts of physics that are not related to his/her research area, and sometimes we even see that within expert's domain (Neurologists usually forget...
Hopenope
2vbFARqWHMp5wCNmu
redwood_conversation
**I wish someone ran a study finding what human performance on** [**SWE-bench**](https://www.swebench.com/#) **is.** There are ways to do this for around $20k: If you try to evaluate on 10% of SWE-bench (so around 200 problems), with around 1 hour spent per problem, that's around 200 hours of software engineer time. So...
Nikola Jurkovic
cnaLRYwFqYwKyiunG
redwood_conversation
x-posting a kinda rambling [thread](https://x.com/saprmarks/status/1868403534070518268) I wrote about [this blog post](https://www.tilderesearch.com/blog/sieve) from Tilde research. \-\-\- If true, this is the first known application of SAEs to a found-in-the-wild problem: using LLMs to generate fuzz tests that don't...
Sam Marks
8NxJy6XczsLDDnGyi
redwood_conversation
**Reputation is lazily evaluated** When evaluating the reputation of your organization, community, or project, many people flock to surveys in which you ask randomly selected people what they think of your thing, or what their attitudes towards your organization, community or project are.  If you do this, you will ve...
habryka
rzB5wieXwqqtZ4piw
redwood_conversation
I asked Rohin Shah what the debate agenda is about and he said (off the cuff, not necessarily worded well) (context omitted) (emphasis mine): > Suppose you have any task where given an input x the AI has to produce some output y. (Can be question answering, arbitrary chatbot stuff, being an agent, whatever.) > > Deba...
Zach Stein-Perlman
DLYDeiumQPWv4pdZ4
redwood_conversation