The Nonlinear Library

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Exploring big ideas in development economics: an interview with Ranil Dissanayake, published by Probably Good on September 11, 2024 on The Effective Altruism Forum. This interview is a cross-post from Probably Good's new Career Journeys series. "No one will ever complain that you write too well… I look back at stuff I wrote 20 years ago and cringe now. But if I didn't write it, I would still be bad. You're not learning unless you're doing it." How does someone break into the field of development economics? And what does working in the field actually look like? Ranil Dissanayake is a senior fellow at the Center for Global Development. His career journey spans several countries and unique experiences - including over 15 years working in international development policy-making. Along the way, he's realized the importance of writing often, sharing ideas often, and talking with people of all sorts of backgrounds. We recently chatted about his path to his current role, day-to-day work, and advice for people starting out - edited below for clarity and brevity. What did you want to do when you were younger? And how did your ambitions change over time? I went to university pretty young, at 17. As a teenager, I wanted to be a filmmaker. I was mad about movies and watched them all day. I didn't want to pursue film straight out of school, though, because I didn't think I had anything interesting to say yet. My vague medium-term plan was to get a degree then try to go to film school in New York. I decided to study history and economics because I had this amazing teacher for these subjects. They weren't my best, but they were definitely my favorite. I applied and got into my top choice, Oxford, which had a joint honors program in both economics and history. Throughout my first couple of years there, I really enjoyed the academic side of things, but filmmaking was still my side gig. I was entering screenplay competitions and making short films. In my third and final year, I did a course in development economics. I'd always been interested in the subject - in part because I'm from Sri Lanka and was born in Hong Kong in 1981. If you were born in Hong Kong in 1981, you know something about development. There was a period of incredibly rapid change and increases in wealth. One side of my family was middle-class professionals, but my other side came from a tiny village without electricity and running water. I'd go to see my family in Sri Lanka every year and the dissonance between life there and my life in Hong Kong (with its high rise buildings and new shopping malls) was striking. When I did this one course in development economics, I realized, 'Wow, there's a whole discipline for thinking about this stuff.' At the time, it was still seen as a niche subject. This was before randomized control trials and before the field had its revolution. It wasn't the sort of thing mainstream economists did, but I just found it all so engaging. After I finished my degree, I still felt too young to do anything interesting in filmmaking. I decided to do a master's in development economics at SOAS in London, which had a totally different style of economics than Oxford. (I think it's a good idea for people to do their degrees at different places). Again, I just loved the course and everything I was learning. I lived somewhat of an unusual life for a 20 year old. I would go to my lectures, then get drinks with my classmates from around the world and we'd talk about what it was like where we were from. It was such an interesting experience. What happened after the master's degree? At that point, I decided development economics could be a good alternative career for me. I cared about it innately. It resonated with my personal background. But it was (and still is) hard to break into the field. I sort of luc...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Formalizing the Informal (event invite), published by abramdemski on September 11, 2024 on LessWrong. Formalizing the Informal One way to view MIRI's Agent Foundations research is that it saw the biggest problem in AI safety as "human preferences are informal, but we need to somehow get formal guarantees about them" -- and so, in response, it set out to make a formal-informal bridge. Recently, I've been thinking about how we might formally represent the difference between formal and informal. My prompt is something like: if we assume that classical probability theory applies to "fully formal" propositions, how can we generalize it to handle "informal" stuff? I'm going to lead a discussion on this tomorrow, Wednesday Sept. 11, at 11am EDT (8am Pacific, 4pm UK). Discord Event link (might not work for most people): https://discord.com/events/1237103274591649933/1282859362125352960 Zoom link (should work for everyone): https://us06web.zoom.us/j/6274543940?pwd=TGZpY3NSTUVYNHZySUdCQUQ5ZmxQQT09 You can support my work on Patreon. Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Reconsidering the Celebration of Project Cancellations: Have We Updated Too Far?, published by Midtermist12 on September 10, 2024 on The Effective Altruism Forum. Reconsidering the Celebration of Project Cancellations: Have We Updated Too Far? Epistemic status: Low certainty. These are tentative thoughts, and I'm open to alternative perspectives. Posting from an alt account. The Effective Altruism community has made significant strides in recognizing the importance of quitting projects that don't deliver short-term results, which helps counteract the sunk cost fallacy and promotes the efficient use of resources. This mindset, in many cases, is a positive development. However, I wonder if we've over-updated in this direction. Recent posts about project cancellations, like the one regarding the Center for Effective Aid Policy (CEAP), have received considerable attention - CEAP's closure post garnered 538 karma, for instance. While I don't have a strong opinion on whether it was prudent to shutter CEAP, I am concerned that its closure, and the community's reaction to it, vibes in a direction where initial setbacks are seen as definitive reasons to quit, even when there might still be significant long-term potential. From an outside perspective, it seemed that CEAP was building valuable relationships and developing expertise in a complex field - global aid policy - where results may take years to materialize. Yet, the organization was closed, seemingly because it wasn't achieving short-term success. This raises a broader concern: are we in danger of quitting too early when projects encounter early challenges, rather than giving them the time they need to realize high expected value (EV) outcomes? There's a tension here between sticking with projects that have a low probability of short-term success but could yield immense value in the long run, and the temptation to cut losses when things aren't immediately working out. High-EV projects often have low-impact modal outcomes, especially in the early stages. It's entirely possible that a project with a 20% chance of success could still be worth pursuing if the potential upside is transformative. However, these projects can look like failures early on, and if we're too quick to celebrate quitting, we may miss out on rare but important successes. This is particularly relevant in fields like AI safety, global aid policy, or other high-risk, high-reward areas, where expertise and relationships are slow to develop but crucial for long-term impact. At the same time, it's essential not to continue investing in clearly failing projects just because they might turn around. The ability to pivot is important, and I don't want to downplay that. But I wonder if, as a community, we are at risk of overupdating based on short-term signals. Novel and complex projects often need more time to bear fruit, and shutting them down prematurely could mean forfeiting potentially transformative outcomes. I don't have an easy answer here, but it might be valuable to explore frameworks that help us better balance the tension between short-term setbacks and long-term EV. How can we better distinguish between projects that genuinely need to be ended and those that just need more time? Are there ways we can improve our evaluations to avoid missing out on projects with high potential because of an overemphasis on early performance metrics? I'd love to hear thoughts from others working on long-term, high-risk projects - how do you manage this tension between the need to pivot and the potential upside of sticking with a challenging project? Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: AI #80: Never Have I Ever, published by Zvi on September 10, 2024 on LessWrong. (This was supposed to be on Thursday but I forgot to cross-post) Will AI ever make art? Fully do your coding? Take all the jobs? Kill all the humans? Most of the time, the question comes down to a general disagreement about AI capabilities. How high on a 'technological richter scale' will AI go? If you feel the AGI and think capabilities will greatly improve, then AI will also be able to do any particular other thing, and arguments that it cannot are almost always extremely poor. However, if frontier AI capabilities level off soon, then it is an open question how far we can get that to go in practice. A lot of frustration comes from people implicitly making the claim that general AI capabilities will level off soon, usually without noticing they are doing that. At its most extreme, this is treating AI as if it will only ever be able to do exactly the things it can already do. Then, when it can do a new thing, you add exactly that new thing. Realize this, and a lot of things make a lot more sense, and are a lot less infuriating. There are also continuous obvious warning signs of what is to come, that everyone keeps ignoring, but I'm used to that. The boat count will increment until morale improves. The most infuriating thing that is unrelated to that was DOJ going after Nvidia. It sure looked like the accusation was that Nvidia was too good at making GPUs. If you dig into the details, you do see accusations of what would be legitimately illegal anti-competitive behavior, in which case Nvidia should be made to stop doing that. But one cannot shake the feeling that the core accusation is still probably too much winning via making too good a product. The nerve of that Jensen. Table of Contents 1. Introduction. 2. Table of Contents. 3. Language Models Offer Mundane Utility. Sorry, what was the question? 4. Language Models Don't Offer Mundane Utility. A principal-agent problem? 5. Fun With Image Generation. AI supposedly making art, claims AI never will. 6. Copyright Confrontation. OpenAI asks for a mix of forgiveness and permission. 7. Deepfaketown and Botpocalypse Soon. How to fool the humans. 8. They Took Our Jobs. First it came for the unproductive, and the call centers. 9. Time of the Season. If no one else is working hard, why should Claude? 10. Get Involved. DeepMind frontier safety, Patel thumbnail competition. 11. Introducing. Beijing AI Safety and Governance, Daylight Computer, Honeycomb. 12. In Other AI News. Bigger context windows, bigger funding rounds. 13. Quiet Speculations. I don't want to live in a world without slack. 14. A Matter of Antitrust. DOJ goes after Nvidia. 15. The Quest for Sane Regulations. A few SB 1047 support letters. 16. The Week in Audio. Dario Amodei, Dwaresh Patel, Anca Dragon. 17. Rhetorical Innovation. People feel strongly about safety. They're against it. 18. The Cosmos Institute. Philosophy for the age of AI. 19. The Alignment Checklist. What will it take? 20. People Are Worried About AI Killing Everyone. Predicting worries doesn't work. 21. Other People Are Not As Worried About AI Killing Everyone. What happened? 22. Five Boats and a Helicopter. It's probably nothing. 23. Pick Up the Phone. Chinese students talk about AI, safety and regulation. 24. The Lighter Side. Do we have your attention now? Language Models Offer Mundane Utility Prompting suggestion reminder, perhaps: Rohan Paul: Simply adding "Repeat the question before answering it." somehow make the models answer the trick question correctly. Probable explanations: Repeating the question in the model's context, significantly increasing the likelihood of the model detecting any potential "gotchas." One hypothesis is that maybe it puts the model into more of a completion mode vs answering from a c...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Economics Roundup #3, published by Zvi on September 10, 2024 on LessWrong. I am posting this now largely because it is the right place to get in discussion of unrealized capital gains taxes and other campaign proposals, but also there is always plenty of other stuff going on. As always, remember that there are plenty of really stupid proposals always coming from all sides. I'm not spending as much time talking about why it's awful to for example impose gigantic tariffs on everything, because if you are reading this I presume you already know. The Biggest Economics Problem The problem, perhaps, in a nutshell: Tess: like 10% of people understand how markets work and about 10% deeply desire and believe in a future that's drastically better than the present but you need both of these to do anything useful and they're extremely anticorrelated so we're probably all fucked. In my world the two are correlated. If you care about improving the world, you invest in learning about markets. Alas, in most places, that is not true. The problem, in a nutshell, attempt number two: Robin Hanson: There are two key facts near this: 1. Government, law, and social norms in fact interfere greatly in many real markets. 2. Economists have many ways to understand "market failure" deviations from supply and demand, and the interventions that make sense for each such failure. Economists' big error is: claiming that fact #2 is the main explanation for fact #1. This strong impression is given by most introductory econ textbooks, and accompanying lectures, which are the main channels by which economists influence the world. As a result, when considering actual interventions in markets, the first instinct of economists and their students is to search for nearby plausible market failures which might explain interventions there. Upon finding a match, they then typically quit and declare this as the best explanation of the actual interventions. Yep. There are often market failures, and a lot of the time it will be very obvious why the government is intervening (e.g. 'so people don't steal other people's stuff') but if you see a government intervention that does not have an obvious explanation, your first thought should not be to assume the policy is there to sensibly correct a market failure. No Good Very Bad Capital Gains Tax Proposals Kamala Harris endorses Biden's no-good-very-bad 44.6% capital gains tax rate proposal, including the cataclysmic 25% tax on unrealized capital gains, via confirming she supports all Biden budget proposals. Which is not the same as calling for it on the campaign trail, but is still support. She later pared back the proposed topline rate to 33%, which is still a big jump, and I don't see anything there about her pulling back on the unrealized capital gains tax. Technically speaking, the proposal for those with a net worth over $100 million is an annual minimum 25% tax on your net annual income, realized and unrealized including the theoretical 'value' of fully illiquid assets, with taxes on unrealized gains counting as prepayments against future realized gains (including allowing refunds if you ultimately make less). Also, there is a 'deferral' option on your illiquid assets if you are insufficiently liquid, but that carries a 'deferral charge' up to 10%, which I presume will usually be correct to take given the cost of not compounding. All of this seems like a huge unforced error, as the people who know how bad this is care quite a lot, offered without much consideration. It effectively invokes what I dub Deadpool's Law, which to quote Cassandra Nova is: You don't f***ing matter. The most direct 'you' is a combination of anyone who cares about startups, successful private businesses or creation of value, and anyone with a rudimentary understanding of economics. The broa...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Stepping down from GWWC: So long, and thanks for all the shrimp, published by Luke Freeman on September 10, 2024 on The Effective Altruism Forum. It's a rare privilege to lead an organisation that embodies the very ideals that shaped your life. I've been fortunate to have been given that opportunity for the last four years. My journey with Giving What We Can began long before I became its CEO. Like many members, I started as a curious onlooker, lurking for many years after first googling something along the lines of "what's the best charity?" and slowly being drawn to the idea of effective giving. I vividly remember the day I first hovered over the 'donate' button on the Against Malaria Foundation's website after getting my first raise. My heart was racing, wondering if a decent chunk of my small paycheck would truly make a difference… only to go back and read a report on malaria "just one more time." Finally, I found the courage to act. I started giving more effectively and significantly, and a few years later, (after seeing so many others do so before me) I got serious and took the 10% Pledge. It started to feel like I was making good on some promises I'd made to myself back when I was a kid and first struck by the injustices of the world, our collective inaction, and our inability to stop such extreme levels of preventable suffering. Fast forward to 2020, and just as a global pandemic was tearing through the world, I found myself stepping in to lead the team at Giving What We Can. To say I was humbled would be an understatement. Here I was, entrusted with leading an organisation that had inspired my own giving journey. It was a responsibility I didn't take lightly. Over the past four years, I've had the immense honour of working alongside an incredible team and a passionate community of givers. I've been incredibly lucky to get to contribute to a variety of causes I care about, simply by driving more funding to all of them in the form of inspiring pledges and donations to high-impact causes. Together, we've achieved things that that wide-eyed kid donating pennies from his paper route (aka my younger self) could scarcely have imagined, for instance: We've roughly doubled the number of 10% Pledges; Revitalised our community, website, brand, and research; Integrated and improved the donation platform; Pulled off fantastic partnerships and campaigns; Built an exceptional team, with strong retention and a positive work culture, which operates effectively even in my absence (demonstrated recently when I was on parental leave); Spun out of our fiscal sponsor (EV) and established GWWC as an independent multi-entity organisation with a global presence. But as proud as I am of what we've accomplished together and as much as I have loved leading the team at GWWC, I've come to a crossroads. The past 18 months have been challenging, both for GWWC and for me personally. We've navigated significant changes and overcome substantial obstacles. While I'm incredibly proud of how we've handled these challenges, I've found my reserves depleting. Life has thrown a lot my way recently - from deeply painful personal losses to the joyous arrival of our first child. These experiences have led me to reassess my priorities and recognise the need for a change. And so, after careful consideration, I have made the difficult decision to step down as CEO of Giving What We Can. The timing aligns with the completion of our spin-out from Effective Ventures, a major project that has set GWWC up for its next phase of growth. I felt that the decision in front of me was to either redouble my efforts for another 3-5 years or to pass the baton. After careful consideration, I believe it's the right time for new leadership to bring fresh energy and perspectives to drive the organisation forward. I'll be sta...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: The Best Lay Argument is not a Simple English Yud Essay, published by J Bostock on September 10, 2024 on LessWrong. Epistemic status: these are my own opinions on AI risk communication, based primarily on my own instincts on the subject and discussions with people less involved with rationality than myself. Communication is highly subjective and I have not rigorously A/B tested messaging. I am even less confident in the quality of my responses than in the correctness of my critique. If they turn out to be true, these thoughts can probably be applied to all sorts of communication beyond AI risk. Lots of work has gone into trying to explain AI risk to laypersons. Overall, I think it's been great, but there's a particular trap that I've seen people fall into a few times. I'd summarize it as simplifying and shortening the text of an argument without enough thought for the information content. It comes in three forms. One is forgetting to adapt concepts for someone with a far inferential distance; another is forgetting to filter for the important information; the third is rewording an argument so much you fail to sound like a human being at all. I'm going to critique three examples which I think typify these: Failure to Adapt Concepts I got this from the summaries of AI risk arguments written by Katja Grace and Nathan Young here. I'm making the assumption that these summaries are supposed to be accessible to laypersons, since most of them seem written that way. This one stands out as not having been optimized on the concept level. This argument was below-aveage effectiveness when tested. I expect most people's reaction to point 2 would be "I understand all those words individually, but not together". It's a huge dump of conceptual information all at once which successfully points to the concept in the mind of someone who already understands it, but is unlikely to introduce that concept to someone's mind. Here's an attempt to do better: 1. So far, humans have mostly developed technology by understanding the systems which the technology depends on. 2. AI systems developed today are instead created by machine learning. This means that the computer learns to produce certain desired outputs, but humans do not tell the system how it should produce the outputs. We often have no idea how or why an AI behaves in the way that it does. 3. Since we don't understand how or why an AI works a certain way, it could easily behave in unpredictable and unwanted ways. 4. If the AI is powerful, then the consequences of unwanted behaviour could be catastrophic. And here's Claude's just for fun: 1. Up until now, humans have created new technologies by understanding how they work. 2. The AI systems made in 2024 are different. Instead of being carefully built piece by piece, they're created by repeatedly tweaking random systems until they do what we want. This means the people who make these AIs don't fully understand how they work on the inside. 3. When we use systems that we don't fully understand, we're more likely to run into unexpected problems or side effects. 4. If these not-fully-understood AI systems become very powerful, any unexpected problems could potentially be really big and harmful. I think it gets points 1 and 3 better than me, but 2 and 4 worse. Either way, I think we can improve upon the summary. Failure to Filter Information When you condense an argument down, you make it shorter. This is obvious. What is not always as obvious is that this means you have to throw out information to make the core point clearer. Sometimes the information that gets kept is distracting. Here's an example from a poster a friend of mine made for Pause AI: When I showed this to my partner, they said "This is very confusing, it makes it look like an AGI is an AI which makes a chess AI". Making more AI...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: My top 10 picks from 200 episodes of the 80k podcast, published by JWS on September 9, 2024 on The Effective Altruism Forum. Intro I think the 80,000 Hours Podcast is a great show. Despite the world of podcasts overflowing with content to choose from, it's reliably been a high-quality production that's been a regular part of my listening habits ever since I discovered it. It was also probably one of the first routes I become more aware of the EA community, which I suspect I might not be alone by.[1] So, as the podcast numbers ticked up, the vague idea to write up a post shouting out some of my favourite episodes took root. I didn't get far with it from there, and now the unreasonable effectiveness of the 80k podcast production team has forced my hand! So in the post I'm going to link to my 10 favourite episodes, along with some final thoughts at the end. I hope to share with you some of my favourite episodes, but I want to be clear that this is my list and not any sort of official ranking. If there's a really good episode that your surprised isn't on there, it may well because I haven't listened to that one yet! So, without any further ado, here's the Top 10 List: My Top 10 10-4 10: #144 - Athena Aktipis on why cancer is actually one of the fundamental phenomena in our universe While the podcast's title is nominally about cancer, I think the ideas in the podcast actually hint towards Autopoiesis, something I think connects a bunch of different causes I care about. I wasn't really aware of many of the things that Athena and Rob discuss in the episode so I found the discussing incredibly interesting, especially about how the concepts of growth, maintenance, and co-operation appear at many different levels in the universe. 9: #175 - Lucia Coulter on preventing lead poisoning for $1.66 per child LEEP is probably one of the key EA success stories in recent years but hearing it through Lucia's own words and her own story, from CE incubation to actually bringing the results of lead concentration to the Malawi Ministry of Health, was really inspiring to hear. There's also some good discussion about the 10% Pledge and the age-old Randomista v Growth debate in the episode too. 8: #139 - Alan Hájek on puzzles and paradoxes in probability and expected value This was another excellent episode where Rob and an incredbily smart, engaging guest got to do a deep dive into an idea and see where it went. I think Professor Hájek did a fantastic job sharing his knowledge in an enlightening way, and he really showcased a number of limits of expect value calculations (not least the realisation the probability(0) events can and do happen all the time) which left my mind blown in a good way. 7: #153 - Elie Hassenfeld on two big picture critiques of GiveWell's approach, and six lessons from their recent work GiveWell looms large in the world of EA, so to get the CEO to come on the podcast and talk in this detail was great to see. I found Elie both an engaging guest and a persuasive interlocutor when he and Rob get into debates, and this definitely didn't seem like a softball interview to me. The Randomista v Growth section (at 02:20:00) is really good on this, and I wish Rob had actually put himself on the line a bit more since he clearly has a lot of sympathy with the 'Growth' side of the debate. 6: #129 - Dr James Tibenderana on the state of the art in malaria control and elimination This podcast comes in at over 3 hours, but still I found it flying by as a listener. The topics range from the specific work of Maleria Consortium, the overall landscape of the battle against Malaria, as well as James' own story, including fighting off the disease himself. There's some much rich discussion that it feels like any of these could have been an episode on its own, so to get all 3-in-1 firmly puts th...

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: Dispelling the Anthropic Shadow, published by Eli Rose on September 9, 2024 on The Effective Altruism Forum. This is a linkpost for Dispelling the Anthropic Shadow by Teruji Thomas. Abstract: There are some possible events that we could not possibly discover in our past. We could not discover an omnicidal catastrophe, an event so destructive that it permanently wiped out life on Earth. Had such a catastrophe occurred, we wouldn't be here to find out. This space of unobservable histories has been called the anthropic shadow. Several authors claim that the anthropic shadow leads to an 'observation selection bias', analogous to survivorship bias, when we use the historical record to estimate catastrophic risks. I argue against this claim. Upon a first read, I found this paper pretty persuasive; I'm at >80% that I'll later agree with it entirely, i.e. I'd agree that "the anthropic shadow effect" is not a real thing and earlier arguments in favor of it being a real thing were fatally flawed. This was a significant update for me on the issue. Anthropic shadow effects are one of the topics discussed loosely in social settings among EAs (and in general open-minded nerdy people), often in a way that assumes the validity of the concept[1]. To the extent that the concept turns out to be completely not a thing - and for conceptual rather than empirical reasons - I'd find that an interesting sociological/cultural fact. 1. ^ It also has a tag on the EA Forum. Thanks for listening. To help us out with The Nonlinear Library or to learn more, please visit nonlinear.org

Welcome to The Nonlinear Library, where we use Text-to-Speech software to convert the best writing from the Rationalist and EA communities into audio. This is: AI forecasting bots incoming, published by Dan H on September 9, 2024 on The AI Alignment Forum. In a recent appearance on Conversations with Tyler, famed political forecaster Nate Silver expressed skepticism about AIs replacing human forecasters in the near future. When asked how long it might take for AIs to reach superhuman forecasting abilities, Silver replied: "15 or 20 [years]." In light of this, we are excited to announce "FiveThirtyNine," an AI forecasting bot. Our bot, built on GPT-4o, provides probabilities for any user-entered query, including " Will Trump win the 2024 presidential election?" and " Will China invade Taiwan by 2030?" Our bot performs better than experienced human forecasters and performs roughly the same as (and sometimes even better than) crowds of experienced forecasters; since crowds are for the most part superhuman, FiveThirtyNine is in a similar sense. (We discuss limitations later in this post.) Our bot and other forecasting bots can be used in a wide variety of contexts. For example, these AIs could help policymakers minimize bias in their decision-making or help improve global epistemics and institutional decision-making by providing trustworthy, calibrated forecasts. We hope that forecasting bots like ours will be quickly integrated into frontier AI models. For now, we will keep our bot available at forecast.safe.ai, where users are free to experiment and test its capabilities. Quick Links Demo: forecast.safe.ai Technical Report: link Problem Policymakers at the highest echelons of government and corporate power have difficulty making high-quality decisions on complicated topics. As the world grows increasingly complex, even coming to a consensus agreement on basic facts is becoming more challenging, as it can be hard to absorb all the relevant information or know which sources to trust. Separately, online discourse could be greatly improved. Discussions on uncertain, contentious issues all too often devolve into battles between interest groups, each intent on name-calling and spouting the most extreme versions of their views through highly biased op-eds and tweets. FiveThirtyNine Before transitioning to how forecasting bots like FiveThirtyNine can help improve epistemics, it might be helpful to give a summary of what FiveThirtyNine is and how it works. FiveThirtyNine can be given a query - for example, "Will Trump win the 2024 US presidential election?" FiveThirtyNine is prompted to behave like an "AI that is superhuman at forecasting". It is then asked to make a series of search engine queries for news and opinion articles that might contribute to its prediction. (The following example from FiveThirtyNine uses GPT-4o as the base LLM.) Based on these sources and its wealth of prior knowledge, FiveThirtyNine compiles a summary of key facts. Given these facts, it's asked to give reasons for and against Trump winning the election, before weighing each reason based on its strength and salience. Finally, FiveThirtyNine aggregates its considerations while adjusting for negativity and sensationalism bias in news sources and outputs a tentative probability. It is asked to sanity check this probability and adjust it up or down based on further reasoning, before putting out a final, calibrated probability - in this case, 52%. Evaluation. To test how well our bot performs, we evaluated it on questions from the Metaculus forecasting platform. We restricted the bot to make predictions only using the information human forecasters had, ensuring a valid comparison. Specifically, GPT-4o is only trained on data up to October 2023, and we restricted the news and opinion articles it could access to only those published before a certain date. From there, we asked it to compute the probabilities of 177 events from Metaculus that had happened (or not ha...