Papers in AIS highlight some partial solutions to AIS-related problems (like machine unlearning), make progress on some approaches to the core problems of AI safety, like machine interoperability or are just about some safety failures of current AI safety systems (e.g. post). What is the theory of change behind writing...
LLM is just a lot of math. Program running on a computer. Chinese room style arguments ridicule systems that operate on symbols. But LLM looks more like this:
1. there is some language
2. translate that language into operations in real spaces (which are non discrete unlike characters)
3. that operations get translated into operations on symbols
It can be just people trolling or people directly telling their AIs to role play. Anyone can sign up.
OpenAI will allow adult content. This is a massive market (for instance, in 2024 Gross Site Volume for OnlyFans was 7 billion dollars, tho most of it went to the Creator Payments) OpenAI won't be limited by this and will be able to take all that money.
I expect most of the users be interested in videos tho, rather than text and images. So far AI companies let other firms create characters using their services (like character.ai). Grok is an exception. Are they going to train a porn making model? There is a lot of material online. How would they distribute their services?
I am creating a comparative analysis of cross-posted posts on LW and EAF. Make your bets!
I will pull all the posts that were posted on both LW and EAF and compare how different topics get different amount of karma and comments (and maybe sentiment of comments) as a proxy for how interested people are and how much they agree with different claims. Make your bests and see if it tells you anything new! I suspect that LW users are much more interested in AI safety and less vegan. They care less about animals and are more skeptical towards utility maximization. Career related posts will fare better on EAF, while rationality (as in rationality as an art) will go better on LW. Productivity posts will have more engagement on EAF.
It is not possible to check all the bets since the number of cross-posted posts is not that big and they are limited to specific topics.
I briefly read 6pack.care website and your post. It sounds to me like an idea supplementary to existing AI safety paradigms and not solving the core problem - aligning AIs. Looking at your website I see that it's already assumed that AI is mostly aligned and issues with rogue AIs are not mentioned in the risks section
... (read more)A midsize city is hit by floods. The city launches a simple chatbot to help people apply for emergency cash. Here is what attentiveness looks like in action:
- Listening. People send voice notes, texts, or visit a kiosk. Messages stay in the original language, with a clear translation beside them. Each entry records where it came from
I applied for Thomas Kwa SPAR stream but I have some doubts about the direction of research. I post it here to get feedback on my thoughts. Kwa wants to train models to produce something close to neuralese as reasoning traces and evaluate white box and black box monitoring against these traces. It seems obvious to me that when a model switches to neuralese we already know that something is wrong, so why test our monitors against neuralese?
Wikipedia?
source on the ED risk?
do you want to stop worrying?
The goals we set for AIs in training are proxy goals. We, humans, also set proxy goals. We use KPIs, we talk about solving alignment and ending malaria (proxy to increasing utility, saving lives) budgets and so on. We can somehow focus on proxy goals and maintain that we have some higher level goal at the same time. How is this possible? How can we teach AI to do that?
advertisement is literally like flowers
There is a um something like a thing to do in Christianity where you set a theme for a week and reflect on how this theme fits into your whole life (e.g. suffering, grace etc). I want to do something similar but make it just much more personal. I struggled with phone addiction for some time and it seems that bursts of work can't solve that issue. So this week will be the week of reflection on my phone addiction.
I had some thoughts about CoT monitoring. So I was imaging this simple scenario - you are running a model, for each query it produces CoT and answer. To check the CoT, we run it through another model with a prompt like "check this CoT <cot>" and tell us whether the CoT seems malicious.
Why don't we check the response only? Sometimes it's not clear just by the response that the response is harmless (I imagine a doctor can prescribe you a lot of things and you don't have a way of checking if they are actually good for you)
So CoT monitoring is based on the idea that CoT gives you some extra... (read more)
Why are AI safety people doing capabilities work? It happened a few times already, usually with senior people (tho I think it might happen with others as well) and some people are saying it's because they want money and stuff or get "corrupted". Maybe there is like a mindkilling argument behind the AI safety case, a crux so deep we fail to articulate it clearly and people who spent significant amount of time thinking about AI safety just reject it at some level.
Papers in AIS highlight some partial solutions to AIS-related problems (like machine unlearning), make progress on some approaches to the core problems of AI safety, like machine interoperability or are just about some safety failures of current AI safety systems (e.g. post). What is the theory of change behind writing those papers? Are we hoping them to reach the key decision makers at AI labs? Or politicians?
MoltBots don't fear doings things and being cringe which puts them above 80% of humans in agency already.