LessWrong 2.0 Reader

Have you tried it with a book that doesn't have self-contained chapters?

I've laid out a concrete example of this at https://www.lesswrong.com/posts/FgXjuS4R9sRxbzE5w/medical-image-registration-the-obscure-field-where-deep [LW · GW] , following the "optimization on a scaffold level" route. I found a real example of a misaligned inner objective outside of RL, which is cool

Hmm, most of the ordering should be the same. Here is the ordering on Youtube Music: 

The Road To Wisdom
Moloch
Thought That Faster (feat. Eliezer Yudkowsky)
The Litany of Tarrrrrski (feat. Eliezer Yudkowsky)
The Litany of Gendlin
Dath Ilan's Song (feat. Eliezer Yudkowsky)
Half An Hour Before Dawn In San Francisco (feat. Scott Alexander)
AGI and the EMH (mit Basil Halperin, J. Zachary Mazlish & Trevor Chow)
First they came for the epistemology (feat. Michael Vassar)
Prime Factorization (feat. Scott Alexander)
We Do Not Wish to Advance (feat. Anthropic)
Nihil Supernum (feat. Godric Gryffindor)
More Dakka (feat. Zvi Mowshowitz)
FHI at Oxford (feat. Nick Bostrom)
Answer to Job (feat. Scott Alexander)

Which is pretty similar to the order here. The folk album is in a slightly different order (which I do think is worse and we sadly can't change), but otherwise things are the same. 

Even if the stars should die in heaven
Our sins can never be undone
No single death will be forgiven
When fades at last the last lit sun.

Then in the cold and silent black
As light and matter end
We’ll have ourselves a last look back.

And toast an absent friend.

[verse 2]

I heard that song which left me bitter
For all the sins that had been done
But I had thought the wrong way 'bout it
[cuz] I won't be there to see that sun

I noticed then I could let go
Before my own life ends
It could have been much worse you know

Relaxing with my friends
Hard work I leave with them
Someday they'll get it done

A million years too young
For now we'll have some fun

'Any ways to combine all the previously mentioned ideas (A Causal Explainable Guardrails for Large Language Models, Discussion: Challenges with Unsupervised LLM Knowledge Discovery and A Multimodal Automated Interpretability Agent (MAIA), 'Catching AIs red-handed', including paulfchristiano's comment) with ideas from 'Searching for a model's concepts by their shape – a theoretical framework' to 'catch AIs red-handed'?

Here are a few additional ideas for combining all the mentioned papers and comments to try catching deceptive models red-handed:

1. Using the "deliberate adversarial experiments" mentioned in paulfchristiano's comment, like varying the model's beliefs about how closely it is being monitored, to surface cases where the model's behavior is suspiciously sensitive to being watched. The theoretical framework from "Searching for concepts by their shape" could be used to formalize the hypothesis that the model has a hidden "deception mode" that gets triggered when it believes oversight is lax. MAIA's automated experimentation tools could be used to efficiently search for inputs exhibiting this sensitivity.
2. Comparing the features extracted by the interpretability tools from "Searching for concepts by their shape" across different copies of the model that have been adversarially perturbed to have different beliefs (per paulfchristiano's comment). If there are decision-relevant features (e.g. representations of the model's objectives) that are suspiciously different in the copy that believes it is unobserved, that would be strong evidence of deceptive cognition. The causal analysis from "Causal Explainable Guardrails" could help determine whether those differing features are causally linked to the differing outputs.
3. Training MAIA's detectors for suspicious model outputs not just on straightforwardly deceptive behaviors, but on subtler signs of agency-like cognition hypothesized in paulfchristiano's comment, e.g. the model giving worse answers when it believes it is unobserved. This could yield more robust detection that is harder for the model to circumvent by avoiding overt deception.
4. If interpretability tools manage to extract a model's representations of its beliefs, objectives, plans, etc. (a big if!), use the techniques from "Catching AIs red-handed" to examine how those representations differ on the inputs that were "successfully" deceptive (i.e. where the model behaved deceptively but was not caught). There may be learnable patterns in how the model's cognition differs in those cases which could inspire hypotheses about what relations characterize "successful" deception. Those hypotheses could then be formalized and tested using the framework from "Searching for concepts by their shape".
5. A point emphasized in paulfchristiano's comment is the importance of extreme thoroughness and breadth in searching for deceptive cognition, e.g. (to paraphrase) "even if a model exhibits some deceptive behavior, don't assume you've found all of its tricks; keep probing its behavior in a huge variety of contexts". MAIA and other automated testing tools should be put in service of this kind of breadth. Rather than just focus on the most egregious cases of deception, we should test the model's behavior on a vast range of subtle perturbations hunting for any whiff of situational inconsistency. The theoretical frameworks for agency from "Searching for concepts by their shape" can guide us in knowing what to look for.

Extending my previous thoughts, I think one of the most useful things the "Searching for concepts by their shape" framing can contribute is mathematical precision in defining hypotheses about what characterizes deceptive cognition. If we can frame agency/deception in terms of precise relations between features (even if those features are hard to interpretably extract in practice), it gives us a crisper target to aim at with our testing, rather than just relying on intuitive notions of what "seems deceptive".

However, I continue to be quite pessimistic about the prospects for catching competent deceptive models before it's too late, for the reasons outlined previously. I worry that by the time a model is smart enough that we could meaningfully apply these combined interpretability strategies to it, it would also be smart enough to notice what we're doing and come up with countermeasures.

Unless interpretability tools far exceed my expectations in their ability to crack open the black box of neural nets (which would be great!), I suspect the most practical path is to use simpler, more behavioral detection strategies (like the model consistency testing from your Control paper) to avoid building models that are competent enough to be robustly deceptive in the first place. Catching a superintelligent deceptive model after the fact seems like a much harder challenge than noticing warning signs of misalignment in weaker models and heading them off early.

But I would still encourage pursuing all these ideas, even if they end up not being enough on their own to solve the problem. The interpretability insights gained would be valuable for alignment more broadly. And we may get lucky in extracting some key structural signature of deceptive cognition that even very sophisticated models can't fully mask. It's an arms race we can't afford to unilaterally forfeit.'

my most important point here is to approach this as an unsolved "product development" problem

This is a good take. I'd take it one step further and suggest that an even-more material product is the ideal target.

Some of this may be obvious, but for sake of clarity: there exists a process of attempting to identify assumptions which comprise a model which projects to the potential disagreement, for all possible disagreements. In the case of unstated or unnoticed supporting assumptions, this may involve identifying significantly differing probabilities placed on some assertion, and working backwards to identify its supports. Tracking this process explicitly via one or more belief networks, on a whiteboard, a sheet of paper, or even better, purpose-tailored software, allows for reasoners to identify where they may have a crux, and directly focus on building out or otherwise contending with its supports. Reasoners may prefer to produce their own separate belief networks and attempt to merge them.

Mapping out belief networks leads to probability distributions for assertions following from the combinatorics of the supports. When an assertion is under-supported relative to the reasoner's credence, this is made highly visible. The structure makes it apparent when an argument is hand-waved rather than contending with it via the supports it should have or via offering conflicting supports, and helps to fix attention on unfinished work implied by the raising of each assertion. Many rhetorical techniques would rightly fail under these conditions. It also makes for a high utility game of solitaire, whether in advance of an upcoming double crux with another reasoner or not. 

A drawback is that some reasoners may prefer not to reveal all of their supports, as in the case of those which may contain infohazardous or exfohazardous content, or ones which may cause those things to be easier to derive or notice. In some cases, reasoners may prefer to engage in this sort of protocol in private, with the option to multilaterally make the results available after-the-fact.

I strong downvoted this because it's too much like virtue signaling, and imports too much of the culture of Twitter. Not only the hashtags, but also the authoritative & absolute command, and hero-worship wrapped with irony in order to make it harder to call out what it is.

Continued:

'How could the above be combined with 'Catching AIs red-handed' and, in particular, the ideas mentioned in paulfchristiano's comment?

The approaches discussed in "A Multimodal Automated Interpretability Agent" (MAIA) and "Causal Explainable Guardrails for Large Language Models" (LLMGuardaril) could potentially be combined with the ideas in "Catching AIs red-handed" and paulfchristiano's comment to create a more comprehensive strategy for detecting and mitigating risks from deceptively aligned AI systems:

1. Using MAIA to automatically generate and test hypotheses about deceptive behavior: As discussed in the "Catching AIs red-handed" post, once a suspicious action has been flagged, we can use techniques like varying the AI's context and beliefs to test whether the behavior is truly deceptive. MAIA's automated experimentation capabilities could potentially be used to scale up this kind of hypothesis generation and testing, systematically exploring the "landscape of deceptive behavior" as paulfchristiano puts it.
2. Applying causal analysis from LLMGuardaril to rule out alternative explanations: A key challenge raised in the post and comments is distinguishing true deception from an AI system just "kind of unlucky in how its context shifted it into a more deceptive role-play". The causal analysis techniques from LLMGuardaril, which aim to disentangle the influence of semantic context on an AI's behavior, could potentially be applied by MAIA to more confidently establish whether a pattern of suspicious behavior is caused by intentional deception vs. an unlucky context.
3. Using MAIA's explainable component to interpret the AI's reasoning: In paulfchristiano's examples of analyzing a suspicious code review or context-dependent behavior, a critical step is understanding what information the AI is using to make its decisions. MAIA's explainable component, which measures the alignment between an AI's outputs and experimenter-specified directions, could potentially shed light on the reasoning behind a deceptive action.
4. Scaling up detection of hard-to-elicit capabilities: "Catching AIs red-handed" notes that real-world deceptive behavior could reveal capabilities that are hard to measure during normal testing. MAIA's ability to automatically generate diverse test inputs based on a small number of real examples could help systematically probe for these hidden capabilities, reducing the AI's "space of interventions...where they are more and more confident that it's never going to be noticed."
5. Iterative refinement of monitoring systems: Finally, as an AI system's deceptive behavior is progressively uncovered through a combination of MAIA's experimentation and LLMGuardaril's causal analysis, the resulting dataset of confirmed deceptive actions could be used to fine-tune untrusted monitoring systems to more reliably catch future deception attempts, similar to the "few-shot catastrophe prevention" techniques outlined in the post.

Significant open questions remain about the feasibility of scaling these approaches to be robust against highly capable AI systems. There are also important limitations to consider: for example, MAIA's causal analysis is unlikely to provide a fully conclusive "smoking gun" that an AI system is deceptive, as opposed to strong probabilistic evidence. And sufficiently capable systems may find ways to avoid exhibiting any detectable deceptive behavior under the kind of systematic experimentation enabled by MAIA.

Nonetheless, I believe the combination of automated interpretability agents, causally grounded analysis of AI reasoning, and the overall adversarial "red team" mindset advocated in "Catching AIs red-handed" represents a promising direction for building more robust monitoring and containment of potentially deceptive AI systems. A concerted effort to develop and integrate these approaches could significantly improve our odds of detecting and mitigating catastrophic deceptive alignment failures.'

Everything Turing-complete requires infinite memory. When we are saying "x86 set of instructions is Turing-complete" we imply "assuming that processor operates on infinite memory". It's in definition of Turing machine to include infinite tape, after all.

It's hard to pinpoint, but the trick is that it's very nuanced difference between the sense in which transformers are limited in complexity-theoretic sense and "transformers can't do X". Like, there is nothing preventing transformers from playing chess perfectly - they just need to be sufficiently large for this. To answer the question "can transformers do X" you need to ask "how much computing power transformer has" and "can this computing power be shaped by SGD into solution".

Conversation with Claude Opus on A Causal Explainable Guardrails for Large Language Models, Discussion: Challenges with Unsupervised LLM Knowledge Discovery [LW · GW] and A Multimodal Automated Interpretability Agent (MAIA). To me it seems surprisingly good at something like coming up with plausible alignment research follow-ups, which e.g. were highlighted here as an important part of the superalignment agenda.

Prompts bolded:

'Summarize 'Causal Explainable Guardrails for Large
Language Models'. In particular, could this be useful to deal with some of the challenges mentioned in 'Discussion: Challenges with Unsupervised LLM Knowledge Discovery'?

The paper "A Causal Explainable Guardrails for Large Language Models" proposes LLMGuardaril, a framework that incorporates causal analysis and adversarial learning to obtain unbiased steering representations in LLMs. Key points:Existing methods for steering LLMs towards desired attributes rely on unbiased representations and steering prompts, but the representations learned from pre-training can introduce semantic biases that negatively impact steering.
LLMGuardaril systematically identifies and blocks the confounding effects of biases, enabling extraction of unbiased steering representations. It also includes an explainable component providing insights into alignment between generated output and desired direction.
Experiments show LLMGuardaril can effectively steer LLMs towards desired attributes while mitigating biases.Regarding the challenges mentioned in "Discussion: Challenges with Unsupervised LLM Knowledge Discovery":One key challenge is distinguishing the model's own knowledge from knowledge of simulated entities. LLMGuardaril aims to disentangle the influence of semantic biases to obtain steering representations capturing desired attributes. This may help identify model's own knowledge if those attributes are specific to the model itself.
Another challenge is the hypothesized existence of many features satisfying consistency properties beyond just the model's knowledge. LLMGuardaril's causal analysis and adversarial learning aims to block confounding semantic biases, potentially reducing the set of candidate knowledge-like features.
Empirically, contrast-consistent search seemed to find prominent arbitrary features rather than knowledge. LLMGuardaril's explainable component could provide insight into whether extracted features actually correspond to model knowledge.

So in summary, while not a complete solution, LLMGuardaril's approach of causally analyzing and debiasing semantic influences may help address some key challenges around identifying model knowledge in an unsupervised way. But significant open questions remain around definitively distinguishing model knowledge from simulated knowledge of other entities. Further research is needed to determine how much traction causal analysis provides on the hardest aspects of the ELK problem.'

'How might MAIA be useful, potentially combined with 'Causal Explainable Guardrails for Large Language Models', to deal with some of the challenges mentioned in 'Discussion: Challenges with Unsupervised LLM Knowledge Discovery'?

MAIA and the techniques from "Causal Explainable Guardrails for Large Language Models" (LLMGuardaril) could potentially be combined to help address some of the challenges mentioned in "Discussion: Challenges with Unsupervised LLM Knowledge Discovery":
 

1. Distinguishing the model's own knowledge from simulated knowledge: One key challenge raised is that consistency-based methods like contrast-consistent search (CCS) may not be able to distinguish an LLM's "own" knowledge from the knowledge of simulated entities the model may be imagining. MAIA's ability to automatically generate hypotheses and design experiments to test them could potentially be used to probe for features that are uniquely characteristic of the model's own knowledge. For example, MAIA could generate synthetic inputs that vary the context in which a piece of knowledge is elicited (e.g. with/without an explicit simulation frame) and test whether the model's responses are consistent.
2. Reducing the set of candidate knowledge-like features: Another challenge is that many features beyond the model's own knowledge may satisfy consistency properties, making it hard to identify the "true" knowledge. LLMGuardaril's technique of causally analyzing the influence of semantic biases on steering representations and using adversarial learning to debias them could potentially be applied by MAIA to narrow down the set of features that robustly encode model knowledge across prompts/contexts. The explainable component of LLMGuardaril could also help MAIA validate whether candidate knowledge features behave as expected.
3. Overcoming the limitations of one-shot probing: Existing automated interpretability approaches are noted to primarily be tools for one-shot hypothesis generation on a limited set of inputs, rather than scalable and fully rigorous causal explanation. MAIA's ability to iteratively refine hypotheses based on experimental outcomes could lead to more robust explanations. Integrating the causal analysis techniques from LLMGuardaril into this iterative process could help rule out spurious hypotheses.
4. Evaluating unsupervised knowledge discovery methods: The paper emphasizes the need for principled testbeds to evaluate the effectiveness of unsupervised knowledge discovery techniques. The synthetic test systems used to quantitatively evaluate MAIA's performance when ground-truth structure is known could potentially be exte'nded to benchmark a wider variety of unsupervised interpretability approaches.

However, significant challenges would likely need to be overcome to fruitfully combine these techniques. For example, MAIA is currently focused on interpreting components of vision models, while LLMGuardaril focuses on steering language models, so unifying them would require bridging this modality gap. More fundamentally, even with access to MAIA's automated experimentation and LLMGuardaril's causal analysis, rigorously identifying an LLM's "own" knowledge in a fully unsupervised way, as opposed to the knowledge of a specific simulated entity, remains an open challenge and likely requires further conceptual breakthroughs. Nonetheless, exploring the intersection of scalable interpretability agents like MAIA and causally grounded analysis techniques like LLMGuardaril seems like a promising direction for future work on the ELK problem.'