Effective Altruism Forum
EA Forum

Hide table of contents
Comment Permalink
adekcz0
2
2

I understand the sentiment, but for me it would be dangerously close to moral offsetting, which I am not fan of. E.g. animal welfare activist might say "eating factory farm meat in random restaurants saves me some time, which I am using to help animals". Or "eating meat is okay, because I am saving much more animals by donating to ACE recommended charities or something".

I think naive math checks out, but slightly broader conception of ethics would discourage such reasoning.

But to be more constructive, I subscribed to Antrophic's Claude. Claude Opus should be on similar level as GPT4 and I have much more trust in the company (but I am ready to abandon them if they fail to live up my safety expectations).

But there is a lot of smaller companies that can do a lot with smaller models like Llama, like https://www.phind.com/, mainly targeted for coding. Companies like those don't have huge AI ambitions, they are just (very well) leveraging current technology provided by big players to make an useful product.

Reply
See in context
by adekcz
3 min read 14

26

AI safetyCriticism of effective altruist organizationsCriticism of work in effective altruismOpenAILarge Language Models
Frontpage

If you are sending money to OpenAI, what would make you stop? For me, there is no ambiguity anymore. The line has been crossed.

Many publicly available signals are pointing towards OpenAI's lack of enthusiasm about AI safety while being quite enthusiastic about advancing AI capabilities as far as possible, even towards AGI. There have been several waves of people with AI safety views leaving OpenAI culminating in dissolution of superalignment team and it is hard to tell if OpenAI conducts any alignment research at all. I worry about the risk of advanced AI and no longer trust OpenAI to behave well when needed. Therefore I am cancelling my subscription and think others should as well.

I have basically written this post a few days ago but was waiting for Zvi to provide his summary of events in Open exodus. Predictably, he just did and it is much better summary than I would be able to provide. In the end, he writes:

Everyone needs to update their views and plans based on this new information. We need to update, and examine our past mistakes, including taking a hard look at the events that led to the founding of OpenAI. We should further update based on how they deal with the NDAs and non-disparagement agreements going forward.

My small update is cancelling GPT subscription and I think as many people as possible should do that. Paying for GPT and having my beliefs was, as of last week, debatable. Now, I think it is unjustifiable.

It is very simple. Sending money to a company that is trying to achieve something I worry about is not a good idea. They don't even have any plausible deniability of caring about safety now. On the other hand, they are definitely still at the leading edge of AI development.

I think AI risks are non-negligible and the plausible harm they can cause is tremendous (extinction or some s-risk scenario). Many people also believe that. But we arrive to really low probabilities when thinking about how much can an individual influence those large problems spanning large futures. We get very small probabilities but very large positive values. 

In this case, we have really small probability that our 20 bucks/month will lead to unsafe AGI. But it definitely helps OpenAI, even though marginally. By this logic, sending money to OpenAI might be by far the most horrific act many people, including many EAs, regularly commit. Normal people don't have assumptions that would make them care. But we have them, and we should act on them. Cancel your subscription and refrain from using the free version, as your data might be even more valuable to them than the subscription money. Please spread this sentiment.

My summary of recent events

For the sake of completeness, if Zvi hadn't published his post, I had written this, leaving it here in a slightly draft-ish state:

All the people

People who are leaving OpenAI seem to have quite similar feelings about AI safety as do I and they seem to be very competent. They are insiders, they have hands-on experience with the technology and OpenAI's inner workings. Dario Amodei and around 10 other OpenAI's employees left in 2021 to found Antrophic, because they felt they can do better AI safety work together (https://youtu.be/5GtVrk00eck?si=FQSCraQ8UOtyxePW&t=39). Then we had the infamous board drama that led to Sam Altman’s stronger hold over OpenAI, and the most safety-minded people are there no more. Helen Toner of CSET, Tasha McCauley, former Effective Ventures board member, and Sutskever, who initially just left the board, now he has left OpenAI altogether. And finally, over the last few months, the remaining safety people have been pushed away, fired, or resigned https://www.vox.com/future-perfect/2024/5/17/24158403/openai-resignations-ai-safety-ilya-sutskever-jan-leike-artificial-intelligence .

Walk the walk

Founding the superalignment team (https://openai.com/index/introducing-superalignment/) led by Sutskever and Leike and promising 20% of compute to them sounded great. It still seems like it would much more likely happen in a worlds, where OpenAI's executives actually care about AI safety. But it didn't last. Leike tweets about "sailing against the wind" and it being harder and harder to conduct safety research.

I can't help but see the following string of events: "founding of the superalignment team (with an AI safety-minded board)", "board drama where AI safety people left", "superalignment effectively disbanded." A cynical reading would be that the superalignment team existed only because the temporary power balance allowed it, and as soon as the power center shifted towards Altman, he took actions that led to its dissolution.

26

1
7

Reactions

1
7

More posts like this

Comments14


Sorted by
New & upvoted
Click to highlight new comments since:
Brad West🔸13
7
3

For me, ChatGPT greatly increases the productivity of myself and my team, whereas the very modest effect of a small amount of money from my subscription I find very unlikely to be seriously furthering the acceleration of AI.

I suspect that the productivity of EAs generally is very valuable and if EAs benefit from the tool it is likely not a good idea for them to stop using it.

Given that there is so much less money going to AI safety than AI capabilities, I would think that a more sensible request would be that those using ChatGPT and thus funding OpenAI fund promising AI safety efforts... this would likely more than offset the harm caused by your funding and enable you to keep using a valuable tool. And if the benefits for you are not worth the cost of the subscription + the offset, then perhaps the benefit is not, in that case worth the harm. I would suggest that people who know more about this stuff than me recommend an AI safety fund for offsetting ChatGPT use.

Reply
quinn8
0
0

how much better is chatgpt than claude, in your experience? I feel like it wouldn't be costly for me to drop down to free tier at openai but keep premium at anthropic, though I would miss the system prompt / custom gpt features. (I'm currently 20/month at both) 

Reply
Vaughn Papenhausen2
0
0

For the tasks I use it for (mainly writing help), Claude Opus is often better than GPT-4

Reply
Brad West🔸2
0
0

I honestly don't have much experience other than using GPT4, which I have found to be very helpful. 

Reply
Karthik Tadepalli3
0
2

Without rehashing the moral offsetting debate, I seriously doubt that there are any AI safety funding options that provide as much benefit as the harm of enabling OpenAI. This intuition comes from the fact that Open Phil funds a ton of AI safety work, so your money would only be marginal for AI safety work that falls below their funding bar, combined with my anecdotal (totally could be wrong) view that AI safety projects are more limited by manpower than by money.

Reply
Brad West🔸2
0
0

This strikes me as remarkably counterintuitive, given the enormous disparity between funding between AI capabilities spending and AI safety spending. I was also under the impression that AI capabilities were not as funding-constrained. 

Reply
Karthik Tadepalli5
3
0

AI companies are constrained by the risk that they might not be able to monetize their products effectively enough to recover the insane compute costs of training. As an extreme example, if everyone used free GPT but zero people were willing to pay for a subscription, then investors would become significantly less excited by AI companies, because the potential profits they would expect to recover would be lower than if people are willing to buy subscriptions at a high rate.

So I think it's better to frame the impact of a subscription not as "you give OAI $20" but rather "you increase OAI's (real and perceived) ability to monetize its products by 1/(# of subscribers)".

Reply
Brad West🔸2
2
0

Fair enough.

I still suspect that you may be underestimating marginal AI Safety funding opportunities.

Reply
Benevolent_Rain1
0
0

One observation here, which might not be useful: I think the people most effective at combating climate change do consume fossil fuels both directly (e.g. international flights) and indirectly (e.g. plastics in stuff they purchase). There are value too in those people that try to delineate themselves completely - they all play different roles.  But they might try to limit their consumption. So perhaps I should not use GPT for e.g. entertainment, but only strictly for work.

Keep in mind that even if we stop our subscription, thousands of people in the supply chain of your goods will have subscriptions so you will still be indirectly funding OpenAI unless you go to the extreme length of only buying stuff from supply chains free of OpenAI.

I might well be biased here, I used to be working on climate and try to wean myself completely off fossil fuels and now I have greatly moderated/rationalized my views.

Reply
adekcz0
2
2

I understand the sentiment, but for me it would be dangerously close to moral offsetting, which I am not fan of. E.g. animal welfare activist might say "eating factory farm meat in random restaurants saves me some time, which I am using to help animals". Or "eating meat is okay, because I am saving much more animals by donating to ACE recommended charities or something".

I think naive math checks out, but slightly broader conception of ethics would discourage such reasoning.

But to be more constructive, I subscribed to Antrophic's Claude. Claude Opus should be on similar level as GPT4 and I have much more trust in the company (but I am ready to abandon them if they fail to live up my safety expectations).

But there is a lot of smaller companies that can do a lot with smaller models like Llama, like https://www.phind.com/, mainly targeted for coding. Companies like those don't have huge AI ambitions, they are just (very well) leveraging current technology provided by big players to make an useful product.

Reply
Brad West🔸4
0
0

To be clear, I am in favor of promoting offsetting in both contexts, although the benefits of veganism in avoiding contributing to factory farming demand, increasing demand for pro-social vegan products, and sending an important moral signal make it difficult to calculate an appropriate sum. Further, I think a deontological or virtue ethics concern with killing or eating the flesh of sentient beings also naturally arises.

In the case here though, your choices cash out in terms of your effect on X and S risks re AGI. I think an appropriate offset for the funding effect is able to reverse or more than reverse your effect without moral complication.

Reply
Geoffrey Miller12
0
0

adekcz - thanks for writing this. I'm also horrified by OpenAI turning from well-intentioned to apparently reckless and sociopathic, in pushing forward towards AGI capabilities without any serious commitment to AI safety.

The question is whether withholding a bit of money from OpenAI will really change their behavior, or whether a 'ChatGPT boycott' based on safety concerns could be more effective if our money-withholding is accompanied by some noisier public signaling of our moral outrage. I'm not sure what this would look like, exactly, but I imagine it could include some coordination with the 'Pause AI' movement (active on X/Twitter). I think a public commitment to boycotting OpenAI (e.g. through social media posts) would be helpful -- especially through channels that don't normally attract much AI safety discussion (e.g. Facebook, TikTok, Instagram -- in addition to X).

tldr: boycotts work best when we make a lot of noise about them, not when they're just a private withdrawal of funding.

Reply
adekcz3
0
0

Agree, boycotts should be public, therefore my post :).

On the other hand, I don't see myself being able to push this to the next level (very little personal fit). I would be very happy if someone did.

Reply
John Salter4
2
0

If you're using it enough, you're probably costing them money. Using the everliving shit out of it would be better for you and for screwing openAI, if you're that way inclined

Reply
Curated and popular this week
 ·  · 5m read
 · 
This work has come out of my Undergraduate dissertation. I haven't shared or discussed these results much before putting this up.  Message me if you'd like the code :) Edit: 16th April. After helpful comments, especially from Geoffrey, I now believe this method only identifies shifts in the happiness scale (not stretches). Have edited to make this clearer. TLDR * Life satisfaction (LS) appears flat over time, despite massive economic growth — the “Easterlin Paradox.” * Some argue that happiness is rising, but we’re reporting it more conservatively — a phenomenon called rescaling. * I test rescaling using long-run German panel data, looking at whether the association between reported happiness and three “get-me-out-of-here” actions (divorce, job resignation, and hospitalisation) changes over time. * If people are getting happier (and rescaling is occuring) the probability of these actions should become less linked to reported LS — but they don’t. * I find little evidence of rescaling. We should probably take self-reported happiness scores at face value. 1. Background: The Happiness Paradox Humans today live longer, richer, and healthier lives in history — yet we seem no seem for it. Self-reported life satisfaction (LS), usually measured on a 0–10 scale, has remained remarkably flatover the last few decades, even in countries like Germany, the UK, China, and India that have experienced huge GDP growth. As Michael Plant has written, the empirical evidence for this is fairly strong. This is the Easterlin Paradox. It is a paradox, because at a point in time, income is strongly linked to happiness, as I've written on the forum before. This should feel uncomfortable for anyone who believes that economic progress should make lives better — including (me) and others in the EA/Progress Studies worlds. Assuming agree on the empirical facts (i.e., self-reported happiness isn't increasing), there are a few potential explanations: * Hedonic adaptation: as life gets
 ·  · 38m read
 · 
In recent months, the CEOs of leading AI companies have grown increasingly confident about rapid progress: * OpenAI's Sam Altman: Shifted from saying in November "the rate of progress continues" to declaring in January "we are now confident we know how to build AGI" * Anthropic's Dario Amodei: Stated in January "I'm more confident than I've ever been that we're close to powerful capabilities... in the next 2-3 years" * Google DeepMind's Demis Hassabis: Changed from "as soon as 10 years" in autumn to "probably three to five years away" by January. What explains the shift? Is it just hype? Or could we really have Artificial General Intelligence (AGI)[1] by 2028? In this article, I look at what's driven recent progress, estimate how far those drivers can continue, and explain why they're likely to continue for at least four more years. In particular, while in 2024 progress in LLM chatbots seemed to slow, a new approach started to work: teaching the models to reason using reinforcement learning. In just a year, this let them surpass human PhDs at answering difficult scientific reasoning questions, and achieve expert-level performance on one-hour coding tasks. We don't know how capable AGI will become, but extrapolating the recent rate of progress suggests that, by 2028, we could reach AI models with beyond-human reasoning abilities, expert-level knowledge in every domain, and that can autonomously complete multi-week projects, and progress would likely continue from there.  On this set of software engineering & computer use tasks, in 2020 AI was only able to do tasks that would typically take a human expert a couple of seconds. By 2024, that had risen to almost an hour. If the trend continues, by 2028 it'll reach several weeks.  No longer mere chatbots, these 'agent' models might soon satisfy many people's definitions of AGI — roughly, AI systems that match human performance at most knowledge work (see definition in footnote). This means that, while the compa
 ·  · 4m read
 · 
SUMMARY:  ALLFED is launching an emergency appeal on the EA Forum due to a serious funding shortfall. Without new support, ALLFED will be forced to cut half our budget in the coming months, drastically reducing our capacity to help build global food system resilience for catastrophic scenarios like nuclear winter, a severe pandemic, or infrastructure breakdown. ALLFED is seeking $800,000 over the course of 2025 to sustain its team, continue policy-relevant research, and move forward with pilot projects that could save lives in a catastrophe. As funding priorities shift toward AI safety, we believe resilient food solutions remain a highly cost-effective way to protect the future. If you’re able to support or share this appeal, please visit allfed.info/donate. Donate to ALLFED FULL ARTICLE: I (David Denkenberger) am writing alongside two of my team-mates, as ALLFED’s co-founder, to ask for your support. This is the first time in Alliance to Feed the Earth in Disaster’s (ALLFED’s) 8 year existence that we have reached out on the EA Forum with a direct funding appeal outside of Marginal Funding Week/our annual updates. I am doing so because ALLFED’s funding situation is serious, and because so much of ALLFED’s progress to date has been made possible through the support, feedback, and collaboration of the EA community.  Read our funding appeal At ALLFED, we are deeply grateful to all our supporters, including the Survival and Flourishing Fund, which has provided the majority of our funding for years. At the end of 2024, we learned we would be receiving far less support than expected due to a shift in SFF’s strategic priorities toward AI safety. Without additional funding, ALLFED will need to shrink. I believe the marginal cost effectiveness for improving the future and saving lives of resilience is competitive with AI Safety, even if timelines are short, because of potential AI-induced catastrophes. That is why we are asking people to donate to this emergency appeal
Recent opportunities in AI safety
15
AE Studio is hiring!
AE Studio
·
0
0
11
Manifund 2025 Regrants
Austin
·
0
0
3
An International Collaborative Hub for Advancing AI Safety Research
Cody Albert
· · 6m read
0
0
View more