Leading EA at UPenn. Rising sophomore.
Email: hazemh [at] upenn [dot] edu
Hazemđ¸
Karma: 263
Hi Catherine, thanks for your comment!
a couple hundred EA University groups around the world
Fair, I couldnât find a map/âlist of the current active ones (outside NESTâs). I was looking for something like the one made by the AI Safety Seeding Initiative. Thanks for correcting my laziness haha. However, Iâd like to push back on âcouple hundredâ; if I counted correctly, https://ââforum.effectivealtruism.org/ââgroups shows 284 EA groups (I excluded alignment/ââAIS/ââACX groups). A portion of those are uni groups and a smaller portion are active. Not sure what those two filters would yield but my hunch is that itâd be <100.
It may be better to describe the tradeoff explicitly when it comes up, rather than handwave like, this was easy to automate and thereâs no slop effect.
100% agree. I think the âpersonal pleaâ part is what I lose out on when I do this, so I do share the best opportunities manually as a result. Also agree with thinking about both the magnitude and liklihood of an error. Thanks for your reply!
Thatâd be amazing. I personally attempted to find people to start EA groups at Drexel, Haverford, Bryn Mawr, & Swarthmore⌠and failed. If you can do that Iâd be really grateful!
Help (re)start AI Safety at UPenn!
Worst case is they have a good reason to not use LLMs, and you might want to worry about that reason to.
Could be true, yes. I was hoping Iâd get some pushback/âfeedback here or elsewhere if this is true (which it could be), so thanks for your comment!
On the risk of privacy leak, you can bound what your LLM can do via permissions. Personally, I do not let it send emails without my manual approval, which allows me to glance review all outbound comms (takes a couple seconds, helps avoid errors). For what itâs worth, Iâve never caught a privacy leak. Worst thing Iâve caught is an overly-long email subject line.
On the erasure of data, version control and trash bins mitigate this risk. Choosing perms is useful here, tooâI personally do not permit my LLMs to delete any files. If an edit is destructive, Iâd simply revert that edit via version control (again, for what itâs worth, Iâve never had an LLM âgo haywireâ and commit destructive edits).
Other problem: are your sure your AI writing isnât coming across as slop?
Could be. I do not mind simple reminders to be âsloppyââhere is something I glanced reviewed and thought it looked good (after which it was schedule-sent):
Hi everyone,
A quick reminder that our EA Summer Fellowship meeting is tonight/âtoday (depending on your timezone), Wednesday, July 22, at 9:00 PM ET.
Before the meeting, please complete the core materials for Week 2: The Effectiveness Mindset in the syllabus.
For sharing opportunities on Slack, it looks something like this (copy-pasted from Penn EAâs #opportunities):
Longview Philanthropy is funding research on digital minds (AI sentience, consciousness, and how society should handle the associated risks). Three opportunities:
Research fellowships: $55K-$150K/âyear for 1-2 years, for those with a terminal degree in CS, neuroscience, law, philosophy, econ, poli sci, or related.
Career development fellowships for people pivoting into the field.
Grants for applied work: seed funding for new orgs, programs, applied research.
Priority areas: empirical tests of AI consciousness/âcognition, legal and governance frameworks, agent interactions and trade, field-building.
Apply by July 10, 2026: https://ââwww.longview.org/âârequest-for-proposals-research-and-applied-work-on-digital-minds/ââSent using @Claude
Does either of these come off as slop to you?
I personally think that for sending reminders, updates, or opportunities, youâre just trying to share some information in a concise manner, and LLMs do that well (based on my personal experience).
Am I missing out on strengthening personal connections because, I donât know, I couldâve done the same but with my tone (and humor or something)? Maybe. But Iâd trade off that small benefit for saving hours. Plus, I think there are more effective ways of strengthening personal connections than having my personal tone in reminders or #opportunities messages.
Edit to add:
Your example with forwarding the recommendation was lossyâthe LLM did not forward your reason for thinking it was good for the group, it made up a new one.
Actually it didnât! Because I didnât ask it to include a reason; I told it to share it, and so it did. Personally, when I share opportunities via Slack, I normally donât include reasons for why theyâre good. When something excites me or if I want to share the reason for why something is good, I just write the message. Also, setting reasonable perms /â glance-reviewing would help you avoid an LLM passionately arguing for an opportunity or something. (And again for what itâs worth Iâve never experienced that)
ComÂmuÂnity orÂgaÂnizÂers: make AI do your dishes!
Sorry for the late response, got busy :/â
I think Iâd want EA to be more active but not as one centralized entity (CEA or else). I personally think EA is too apolitical; I agree that who gets to speak, what gets funded, etc. are already semi-political choices. I also dislike the Forum removing political posts from the Frontpage (LW doesnât).
I sometimes envy the AIS community for how active they are lol. The coordination/âactivity around the pause movement, in particular, is unlike anything Iâve seen within EA.
I think I found a bug thatâs related to the migration. When I try to cross-post my EA Forum post to LessWrong via the âCrosspost to LessWrongâ option, I click the login button only to get:
On Brave (both normal and incognito):
Error: Unexpected token '<', "<!DOCTYPE "... is not valid JSONOn Portable Firefox:
Error: JSON.parse: unexpected character at line 1 column 1 of the JSON dataI asked GPT-5.6 Sol to investigate (I was trying to get it to work, assuming the error was on my side), and it said the following. Following text is generated by it:
The login button sends a POST request to the EA Forumâs
/api/v2/crosspost/generateTokenendpoint.That endpoint currently returns an HTTP 404 HTML page beginning with
<!DOCTYPE html>, instead of JSON. The equivalent LessWrong endpoint returns valid JSON normally.The likely cause appears to be in the new Forumâs proxy configuration. It routes all
/api/*requests to the new codebase unless they are explicitly listed as old-site routes. However, the new codebase doesnât implement the crossposting API, while ForumMagnum still implements that endpoint.The request therefore appears to be intercepted by the new site instead of reaching the legacy ForumMagnum site. ForumMagnum then tries to parse the HTML 404 page as JSON, causing both errors above.
As a check, I sent the same request with
prefer_ea_forum_v3=false, which made the endpoint return valid JSON. This seems to confirm that the migration proxy is causing the issue. My suggested fix is to add/^\/api\/v2\/crosspost\//tooldSitePatterns.
YES! It doesnât affect anything but it annoys me so much
EA intro fellowships should include a âconsider community-buildingâ CTA in their syllabus.
I looked at multiple syllabi and only found one that does this explicitly (shout out Yale EA!). The last section in an IF syllabus (titled âPutting It Into Practiceâ or âTaking Actionâ or something) often focuses on careers and doesnât prompt students to seriously consider organizing the group as a valid option.
I think career decisions take time (even small ones like âI should start rethinking my majorâ), and unpreparedness to Take Big Action sometimes results in no action at all. Community-building is a valid high-value plan that sharpens thinking and leads to the other things the Take Action section of every intro fellowship syllabus is trying to optimize for.
Also, meta EA isnât what people new to EA imagine when they think about what âdoing goodâ looks like. Thatâs not because they donât view it as such, but because they often donât see that framing (which is fair; itâs an intro, you shouldnât be talking about the talking). But I feel like by the end of a 5-8 week fellowship they should think about meta EA and learn why itâs so important (based on the EA principles they learend), the same way they learned about animals, AI, and biosecurity as âcause areasâ they can do good in.
But, at the same time, it kind of assumes that people are willing to commit to eight weeks of reading and discussion
You can run a 5 or 6-week one. Also, you can plan it so that students do the readings together, in-person, then discuss.
Good luck!
âuh-oh, Iâm realizing everyone else here is v**n, I probably need to be very cautious about what I reveal about myself, do I need to hide?â
I would say, as a lacto-vegetarian who goes to vegan events, I never felt this way, though I can see how someone could. When asked, âHow long have you been vegan?â I clarify that Iâm not veganâthat Iâve been vegetarian since January, and, sometimes, that I am thinking about going veganâand I ask them the question back. Sometimes theyâre vegan, sometimes theyâre not! But itâs always a nice icebreaker and a fun experience; itâs never been negative for me.
I also think a non-vegan answering the question and hearing back, âThatâs completely fine!â or âOh, thatâs cool, Iâm flexitarian myself, actually!â shows that the âveganâ community is not just for vegans but for all animal people.
And I do like âanimal peopleâ as a term, though. Sounds cool.
Iâd also suggest âSo, whatâs your story?â /â âWhat brings you here?â as an icebreaker, where the context of being at an animal-people event makes it clear what youâre asking, without assuming much.
(Disclosure: Charlie significantly helped restart EA at Penn, which I currently lead, so I may be biased toward agreeing with him.)
Iâve been watching and liking your stuff for a while, and I think the videos are great. You could benefit from an editor who extracts more clips from your interviews (though youâre already doing this quite well!). My (limited) understanding is that social media is a consistency and frequency game, and you seem to be hitting the nail on the head with both. I particularly like how you engage with commenters in good faith, as you do in the street interviews.
Iâve sent your stuff to a couple of friends when theyâve said something along the lines of, âNormal, everyday people donât know whatâs happening and donât get to weigh in on the transformation AI will bring,â because your work is the number-one example that comes to mind whenever someone says that. I think you should keep it up!
I agree with 1; I got the same impression from speaking with <100 Manhattan residents while street canvassing for Alex Bores. But Manhattan residents seemed more informed than the people in your interviews (prob. because theyâre more educated /â wealthier /â more politically active).
I agree with 2, especially this:
To summarize, the arguments for more transparency, safety, or regulation are actually quite Overpowered to the average person conditional on having ~soonish timelines and actually being prompted to think about 10 years from now.
I think that even without explicitly considering timelines, âtransparency/âsafety/âregulation around corporations building black boxes that might replace jobsâ is very easy for the average person to support.
I partially agree with 3. Iâve heard everyday people mention fears about job loss a lot. Loss of meaning, less so, though thatâs probably because I didnât discuss it as much and because most of my convos were <10 minutes (too short to reach deep topics).
I think human extinction is believable but people donât actually believe it yet (i.e. they havenât thought about it enough to have a belief). I asked many everyday people about their concerns regarding AI, and they never mentioned extinction, but I also never directly raised the possibility.
On 4, I donât think the comparison between the âEA communityâ and Torchbearer is fair. Torchbearer is specifically focused on supporting advocacy, while the EA community is very broad. I suspect your quick take didnât break containment because it didnât reach the right people.
Also, consider cross-posting this to LW; I think it might get more traction.
Basically, if the creator has good justification for their belief, they should be able to give you that justification. If they canât, then why does their belief mean anything?
From a Bayesian perspective: if I hold credence P in X, and the creator says their credence is 100%, I update toward them because they are more reliable. Same way Iâd move toward âvaccines workâ if I were agnostic and learned that 90% of scientists believe they do (i.e. I donât need to see each [or any] scientistâs reasoning/âevidence to update toward their view). So the creatorâs belief can be evidence via reliability.
I agree that if the creator believes X thereâs some evidence for X somewhere; I just donât think that evidence has to be revealed for me to update. (Note: Scientists have a track record and the creator doesnât, but itâs not the track record itself that makes me update, itâs reliability; for scientists I get reliability via their track record, and for the creator itâs a medium-confidence assumption [less-defensible, I know])
Also, âif they canât give you the justificationâ assumes silence = inability. A creator stating a belief without the reasoning/âevidence doesnât mean they canât do so, just that they didnât. Besides, maybe they canât communicate; does that mean there is nothing to communicate? I donât think so. Either way (or other ways I canât think of, beyond choosing not to and not being able to communicate), I think learning about the creatorâs belief warrants an update.
On 1, Iâm not sure I understand why moral intuitions wouldnât change in a CEV. Why would intutions not be partly downstream of cognitive capacity? (Also: this is probably cherry-picking (Googled the question), but people with higher cognitive (verabl) abilities seem to have weaker âpurityâ-based moral intutions.)
Also, itâs not just the intuitions themselves that could change, but the confidence in them. Our intuitions (esp. when taken to their logical conclusions) contradict each other a lot; however, when doing moral philosophy, we often prioritize the higher-confidence one (e.g., my intuition that harming someone with no benefit to any is bad, beats my intuition that murderers deserve punishment). Confidence seems to vary between people (anecdotally), so if we were to scale up cognitively, I imagine that the intuitions we think will win could change.
On 2, I agree. Didnât think of this.
Thank you for the comment! I mostly agree.
Disagree. It would be evidence that the creator of the universe believes Kantian ethics is true. It would only be evidence if we have reason to believe that the creator of the universe is especially likely to be right about ethics, which I donât think we do.
Agree that it would be stronger evidence for the creator/âsimulator believing Kantian ethics is true. I, however, also think that the creator believing X is true is some evidence that X is true, since the creator would reasonably have more power and knowledge, which makes me guess that they might be more likely than us to be right (hence the update toward their view).
Compared to the idea of different human philosophers taking different approaches and ultimately agreeing on ethics, it seems to me that this is only different in degree, in that the super-philosophers are better at philosophy.
Partially agree. The median AI super-philosopher would be better at developing and evaluating arguments than the median human philosopher, but that wouldnât be the only differentiator. There would be more AIs, and the entire process would be more systematic, allowing us to ask moral questions and receive higher-quality answers in a much shorter time. This matters because even if human philosophy is enough to reach true answers, there is less of a reason to be âon the searchâ if the rate of learning is too slow for big updates on moral claims to happen during our lifetimes.
If you have the ability to design minds, then it should be trivial to cause those minds to believe any ethical framework you want them to believe.
I think this assumes that moral intuitions are entirely downstream of the design of the mind. Also, having the ability to mostly force ethical frameworks into designed minds does not mean that you canât avoid doing this. The evidence would be weak(er) if we did not try to avoid this failure mode. I think simulating already-existing minds and then trying to idealize their cognition could lessen this distortion from us. Very naive guesses for ways to âscale up cognitionâ while avoiding âhuman influenceâ: scaling up their neural capacity, or simulating millions of generations and letting evolution do its thing.
Could we find evÂiÂdence for moral reÂalÂism?
Thanks, I didnât know that, esp. the booking timeline. The campaign over Marriottâs noncompliance apparently started in late 2025, so it makes sense.
I think EAG/âEAGx conferences should try (if possible) to avoid venues under active corporate campaign targeting for animal welfare failures.
I recently went to a protest against Marriott over their unfulfilled cage-free egg commitment and couldnât help but think back to EAGxDC which was held at one of their hotels in May.
I donât think the EAGxDC organizers intentionally ignored this; they probably didnât know, or the venue was just very logistically convenient (which I think outweighs this concern).
But I wonder whether organizers generally consider this when choosing venues. I feel like they should, assuming there are other options not under campaign targeting.
The NYC one will have more people, so I think youâd find more people to 1-1 on US-China AI governance. But it depends on how many 1-1s you want to have. If itâs just a few, then both are equal; if you are willing to have 10+ 1-1s on US-China AI governance, I imagine itâd be slightly harder to get that at Berkeley.
The EA Forum Digest might be biased toward posts made on certain days. Had an LLM gather dates for the posts inside Digest #298 through #302, and compare their weekday frequencies against EA Forum posts made within the same window. Probably not enough data (n=50 for Recommended; n=86 for all Digest Posts) to be meaningful.
(Got the idea to check because I was imagining if I was responsible for writing the Digest and thought, âhmmm I feel like if I were sending out emails every Wed, I would be biased toward posts made on certain days because of the availability heuristic and/âor the configuration of my weekly reading/âwriting scheduleâ)