Yonatan Cale

• Yonatan Cale 2 Mar 2025 20:51 UTC

  6 points

  0 ∶ 0

  on: Opinionated take on EA and AI Safety

  On getting a software job in in the age of AI tools, I tried to collect some thoughts here.

  Similar’ish discussions about Anthropic keep coming up, see a recent one here. (I think it would be better to discuss these things in a central place since parts of the conversation repeat themselves. I don’t think a central place currently exists, but maybe you’d prefer to merge over opening a new one)

• Yonatan Cale 28 Feb 2025 20:54 UTC

  6 points

  1 ∶ 0

  on: Bet­ter Games or Bed­nets?

  Perfect Draft Amnesty post!

  Here are my Draft Amnesty thoughts:

  1. I model many game developers as wanting their users to have fun, probably many of them are gamers. In other words, I don’t think it’s enough to TRY to make a fun game, we need something more

  2. Naively, if you don’t optimize for “people get addicted to your game” then probably people will get addicted to another game? Unless maybe you have a clever idea for some tradeoff EA could do that EA couldn’t do?

  3. I don’t think you can assume “we can spend X on making a game and make more than X, which we could spend on something else” (I’m not sure you did that, excuse my draft comment if not)

• Yonatan Cale 20 Feb 2025 22:39 UTC

  4 points

  0 ∶ 0

  in reply to: Ivan Burduk’s comment on: Ivan Bur­duk’s Quick takes

  Oh, looking now—my calendar sync is on but none of the Swapcard events appear in my Google Calendar (not meetups, not 1-on-1s) (I synced to Google Calendar before scheduling anything)

  Do you have a way to debug it? Otherwise I’ll disconnect and re-connect

• Yonatan Cale 20 Feb 2025 20:13 UTC

  3 points

  0 ∶ 0

  on: How To Be­come A Pro­fes­sional Soft­ware Developer

  Updates [Feb 2025] :

  Two big things seemed to have changed:

  1. Things in the economy make it harder to find jobs (different things in different countries. Perhaps this isn’t true for where you live)

  2. AI coding tools are getting pretty good (just wait for March 2022)

  Also, I’ve hardly had conversations about this for a few years (maybe 1 per 1-2 months instead of about 5 per week), so I’m less up to date.

  Also, it seems like nobody really knows what will happen with jobs (and specifically coding jobs) in the age of AI.

  If someone can suggest an example task they don’t think an AI will be able to do in 2-3 years (in the context of picking a career, ideally in software), I’m interested. (we can elaborate in the comments if you want)

  Also, it seems like AI coding tools today can mostly do whatever a developer with 1 year of experience could have typically done (without AI tools), and I assume that in another 1 year this will be true for developers with 2 years of experience. This isn’t literally true for all developers or all tasks, but I think it is one way to “measure” AI progress in a way that is relevant for picking a career (an imperfect measure, but there seems to be a lot of confusion).

  Also, companies that are open to online applications seem to be spammed with LLM generated applications. The market will probably adjust to this somehow, but the current situation seems unusual.

  With all that uncertainty, I find it hard to give career advice.

  I’m tempted to say “I don’t know”.

  But I also think that there are uncertain suggestions that are definitely worse than others, and it is really hard to find reasonable directions to get in to software alone, so here are some of my opinions, and others can happily comment if they disagree.

  1. The bar for getting a first job seems to have gone up, not down.

     1. The bar used to be (vaguely) 2-3 good side projects, or perhaps some impressive STEMy thing (which is common but there are so many special cases that I’m not sure how to summarize them).

     2. If your CV doesn’t get replies and you don’t have side projects, that is still where I’d start.

     3. I’m not sure if side projects have a different weight now that AI coding tools can help, but my guess is

        1. The same reasons that side projects were valuable still apply

        2. Perhaps if making side projects is easier then the expectation is to have more of them, or better ones

  2. Knowing how to use AI tools matters

     1. AI tools aren’t a magic bullet, they still require skill, and hiring someone who knows how to use them—seems to me relevant to the person’s expected productivity

     2. Not all companies moved to using AI tools (or perhaps they are using out-of-date tools), which seems crazy to me, especially given that often after I sit with someone and show them what exists—they change their mind. This might be a too-hot take, but I think companies that don’t use AI tools will stay behind (if their tech velocity matters).

  3. Knowing how stuff works still matters

     1. My recommendation (for this stage of your career) is still “learn the stuff you need for your side project” which is very different from “learn anything that could be called basics”. But still, there are things to understand.

  4. Working on stuff you’re excited about still matters

  5. How much to use AI tools when learning?

     1. I’m not sure

     2. If the AI is producing code that you don’t understand and also doesn’t work and you don’t know how to fix then this is probably pretty extreme into the “using too much AI” direction

        1. Remember that the reason for your bug might not be the last prompt you sent but rather something earlier in how the AI set up your project

     3. Using the AI to automate things that you know how to do, that you fully review and understand and often correct small mistakes—seems like definitely not too much (and perhaps ideal).

     4. Asking the AI what is the best way to do X, or trusting that it picked a direction that makes sense—is probably a mistake (as of Feb 2025). Official docs and tutorials are still much more reliable.

     5. When you don’t yet know how to program anything (e.g before you made your first program, probably assisted by an online course), I probably recommend not using AI at all (unless you feel like you can actually review it reasonably, which seems hard to me)

  6. Skills that AIs don’t currently have

     1. Currently an AI can’t be “a responsible employee”

        1. Soft skills like “have a TODO list so you won’t forget tasks that your manager asked for” or “tell your manager if something is taking longer than planned instead of hiding it and pretending everything is ok” seem still important

     2. Currently AIs aren’t good at architecture

     3. More generally, I think a nice property of doing a side project is that it “forces” you to learn whatever skills the AI doesn’t have in order to accomplish the goal of building something that works

  7. If you like leetcode...

     1. It seems like some companies are dealing with the flood-of-CVs by sending leetcode questions. I’m not sure if this will last (since people can cheat by using AIs), but maybe if you’re able to blast through leetcode questions, that will distinguish you.

     2. I don’t recommend this as a default path, but if you’re excited about leetcode then excitement matters

  I’m interested in more comments/​opinions. I might comment more here myself

• Yonatan Cale 20 Feb 2025 19:16 UTC

  2 points

  0 ∶ 0

  in reply to: Yonatan Cale’s comment on: How To Be­come A Pro­fes­sional Soft­ware Developer

  My new default backend recommendation, assuming you are mainly excited about building a website/​webapp that does interesting things (as most people I speak to are), and assuming you’re happy to put in somewhat more effort but learn things that are (imo) closer to best practices, is Supabase.

  Supabase mostly handles the backend for you (similarly to Firebase).

  It mostly asks you to configure the DB, e.g “there is a table of students where each row is a student”, “there is a table of schools where each row is a school”, “the student table has a column called school_id”.

  It can also handle login and permissions just like Firebase.

  I think learning to use an SQL database (like Supabase which uses Postgres behind the scenes) is somewhat harder than a no-SQL database (like Firebase), but SQL databases teach more relevant skills.

  (FYI this is an opinionated take and some might disagree)

• Yonatan Cale 19 Feb 2025 20:54 UTC

  7 points

  0 ∶ 0

  in reply to: Yonatan Cale’s comment on: Yonatan Cale’s Shortform

  Updates from Berkeley 2025:

  Google Calendar sync

  I can’t believe they finally added this feature!

  See here: https://​​app.swapcard.com/​​settings

  

  Don’t forget you can manage your availability

• Yonatan Cale 12 Feb 2025 22:39 UTC

  2 points

  0 ∶ 0

  on: Job Post Template

  Guided Track used this template

  cc Richard Möhn

• Yonatan Cale 24 Oct 2024 13:36 UTC

  4 points

  1 ∶ 0

  on: How do you fol­low AI (safety) news?

  Every time Zvi posts something, it covers everything (or almost everything) important I’ve seen until then

  https://​​thezvi.substack.com/​​

  Also in audio:

  https://​​open.spotify.com/​​show/​​4lG9lA11ycJqMWCD6QrRO9?si=a2a321e254b64ee9

  I don’t know your own bar for how much time/​focus you want to spend on this, but Zvi covers some bar

  The main thing I’m missing is a way to learn what the good AI coding tools are. For example, I enjoyed this post:

  https://​​www.lesswrong.com/​​posts/​​CYYBW8QCMK722GDpz/​​how-much-i-m-paying-for-ai-productivity-software-and-the

• Yonatan Cale 12 Oct 2024 11:24 UTC

  2 points

  0 ∶ 0

  in reply to: Yonatan Cale’s comment on: How To Be­come A Pro­fes­sional Soft­ware Developer

  Backend recommendations:

  I’m much less confident about this.

  1. If you want a working backend with minimal effort, because actually the React part was the fun thing

     1. Firebase (Firestore) :

        1. This gives you, sort of, an autogenerated backend, if only you describe the structure of your database. I’d mainly recommend this if you’re not interested in writing a backend but you still want things to work as if you built an amazing backend.

        2. The main disadvantage is it will be different from what many databases look like.

           1. You could skip the “subscribe for changes” feature and only use the “read” feature and it will be a bit more realistic, but I don’t actually recommend that.

  2. If you want to write backend in a way that will push you towards best practices, using a technology that is very popular (and specifically documented well for beginners)

     1. Django

     2. A big disadvantage is you’d need to learn some pyhton, and you might be reading this because you already built a fun React project and you want to get it done without learning lots of new tech

     3. DB:

        1. Postgres is the default normal DB for most use cases. SQLite might be easier for local development (on your laptop), and is missing some features that you’ll almost surely not notice, so please don’t care about them until a concrete problem comes up.

  Things that automatically override this advice:

  1. If you’re picking technologies for your startup or something then remember this wasn’t written for you.

  2. If you have a senior developer who will mentor you if you work on some different technology that they know well, then probably go with them.

• Yonatan Cale 12 Oct 2024 10:49 UTC

  2 points

  0 ∶ 0

  on: How To Be­come A Pro­fes­sional Soft­ware Developer

  Tech stack recommendations:

  Many people who want to build a side project want to build a website that does something, or an “app” that does something (and could just be a website that can be opened on a smartphone).

  So I want to add recommendations to save some searching:

  1. React. You can learn react from their official tutorial, or from an online course. Don’t forget to also learn any prerequisites they mention (but you don’t need to invent your own prerequisites)

     1. If you want a React competitor, like Vue, that’s also ok (assuming you’re more excited about it).

        1. Choices I don’t think are legit: JQuery, vanilla javascript.

           1. I can elaborate.

           2. (If someone sent you to “learn the basics” then I probably disagree with them. Yes it has value, but it’s not the most effective path to a first job with mentorship nor to getting far in 5 years. There are other times to learn these things which are not “before your first job”)

  2. For graphics (making your website nice)

     1. If you don’t care about graphics, you can skip all that (including css)

     2. If you don’t want to put effort into graphics but you want your website to look nice anyway, you can use Material UI or Bootstrap. They have ready components like a pretty button or a pretty checkbox, you just import them and only do minimal customization like “hey checkbox, you’re disabled by default”.

     3. If you think making a pretty website is really cool and is obviously the exciting thing to do: you can learn css, and perhaps Tailwind. This isn’t a path I took myself and I don’t know it well. Just remember—it’s a higher priority to get a first project done even if bad and ugly. But as part of “looking for things that excite you to learn more”, graphics might be one of them

  3. Deploying (having a url that you can send to your friends, or put on your CV)

     1. Firebase Hosting

        1. There are many fun ways to deploy websites, including React, I’m specifically recommending Firebase because you might want to use it to solve other problems too (which I might write about).

  4. Backend

     1. If you mainly wanted to make a website-that-does-something, I’d start with that before getting into backend. Once you reach this point on a real project, check if you actually want to continue here. The main important thing is that you work on something that excites you, so you have my official permission to leave your project “not perfect”, without a backend (even if it needs one), if you think that part will be less fun.

     2. Main things a backend will give you (written here so you can check if this seems exciting or meh)

        1. User login

        2. Communication between users (for example if it’s anything like a chat)

        3. Saving information that will still exist if you open the website in a new browser (for example, if you have a TODO list, you probably want to see the list both in your phone and in your computer)

     3. ((I might continue in a sub-comment))

• Yonatan Cale 31 Aug 2024 15:28 UTC

  2 points

  0 ∶ 3

  in reply to: Neel Nanda’s comment on: Should you work at a fron­tier AI com­pany?

  My frank opinion is that the solution to not advancing capabilities is keeping the results private, and especially not sharing them with frontier labs.

  ((

  making sure I’m not missing our crux completely: Do you agree:

  1. AI has a non negligable chance of being an existential problem

  2. Labs advancing capabilities are the main thing causing that

  ))

• Yonatan Cale 31 Aug 2024 13:03 UTC

  2 points

  5 ∶ 2

  in reply to: Yonatan Cale’s comment on: Should you work at a fron­tier AI com­pany?

  I also think that a lot of work that is branded as safety (for example, that is developed in a team called the safety-team or alignment-team) could reasonably be considered to be advancing “capabilities” (as the topic is often divided).

  My main point is that I recommend checking the specific project you’d work on, and not only what it’s branded as, if you think advancing AI capabilities could be dangerous (which I do think).

• Yonatan Cale 31 Aug 2024 12:56 UTC

  25 points

  3 ∶ 1

  on: Should you work at a fron­tier AI com­pany?

  Zvi on the 80k podcast:

  Zvi Mowshowitz: This is a place I feel very, very strongly that the 80,000 Hours guidelines are very wrong. So my advice, if you want to improve the situation on the chance that we all die for existential risk concerns, is that you absolutely can go to a lab that you have evaluated as doing legitimate safety work, that will not effectively end up as capabilities work, in a role of doing that work. That is a very reasonable thing to be doing.

  I think that “I am going to take a job at specifically OpenAI or DeepMind for the purposes of building career capital or having a positive influence on their safety outlook, while directly building the exact thing that we very much do not want to be built, or we want to be built as slowly as possible because it is the thing causing the existential risk” is very clearly the thing to not do. There are all of the things in the world you could be doing. There is a very, very narrow — hundreds of people, maybe low thousands of people — who are directly working to advance the frontiers of AI capabilities in the ways that are actively dangerous. Do not be one of those people. Those people are doing a bad thing. I do not like that they are doing this thing.

  And it doesn’t mean they’re bad people. They have different models of the world, presumably, and they have a reason to think this is a good thing. But if you share anything like my model of the importance of existential risk and the dangers that AI poses as an existential risk, and how bad it would be if this was developed relatively quickly, I think this position is just indefensible and insane, and that it reflects a systematic error that we need to snap out of. If you need to get experience working with AI, there are indeed plenty of places where you can work with AI in ways that are not pushing this frontier forward.

  The transcript is from the 80k website. The episode is also linked to in the post. It also continues to Rob replying that the 80k view is “it’s complicated” and Zvi replying to that.

• Yonatan Cale 28 Jul 2024 13:45 UTC

  10 points

  3 ∶ 0

  in reply to: Guy Raveh’s comment on: Case-con­trol sur­vey of EAGx at­ten­dees finds no be­havi­oural or at­ti­tu­di­nal changes af­ter six months

  Something like “noticing we are surprised”. Also I think it would be nice to have prediction markets for studies in general and EA seem like early adopters (?)

  I don’t know why this was so downvoted/​Xed

  :/​

• Yonatan Cale 27 Jul 2024 17:34 UTC

  4 points

  0 ∶ 0

  in reply to: jacquesthibs’s comment on: jacquesthibs’s Shortform

  Hey :)

  Looking at some of the engineering projects (which is closest to my field) :

  • API Development: Create a RESTful API using Flask or FastAPI to serve the summarization models.

  • Caching: Implement a caching mechanism to store and quickly retrieve summaries for previously seen papers.

  • Asynchronous Processing: Use message queues (e.g., Celery) for handling long-running summarization tasks.

  • Containerization: Dockerize the application for easy deployment and scaling.

  • Monitoring and Logging: Implement proper logging and monitoring to track system performance and errors.

  I’m guessing Claude 3.5 Sonnet could do these things, probably using 1 prompt for each (or perhaps even all at once).

  Consider trying, if you didn’t yet. You might not need any humans for this. Or if you already did then oops and never mind!

  Thanks for saving the world!

• Yonatan Cale 27 Jul 2024 16:51 UTC

  21 points

  10 ∶ 7

  on: Case-con­trol sur­vey of EAGx at­ten­dees finds no be­havi­oural or at­ti­tu­di­nal changes af­ter six months

  If you ever run another of these, I recommend opening a prediction market first for what your results are going to be :)

  cc @Nathan Young

• Yonatan Cale 14 Jul 2024 15:12 UTC

  2 points

  0 ∶ 0

  on: Thoughts on these $1M and $500k AI safety grants?

  Any idea if these capabilities were made public or, for example, only used for private METR evals?

• Yonatan Cale 12 Jul 2024 20:00 UTC

  2 points

  0 ∶ 0

  in reply to: sndr_brst’s comment on: How To Be­come A Pro­fes­sional Soft­ware Developer

  I’m not sure how to answer this so I’ll give it a shot and tell me if I’m off:

  Because usually they take more time, and are usually less effective at getting someone hired, than:

  1. Do an online course

  2. Write 2-3 good side projects

  For example, in Israel pre-covid, having a CS degree (which wasn’t outstanding) was mostly not enough to get interviews, but 2-3 good side projects were, and the standard advice for people who finished degrees was to go do 2-3 good side projects. (based on an org that did a lot of this and hopefully I’m representing correctly).

  
  There is more that I can say about this, but I’m not sure I’m even answering the question.

  Also note that the main point of this post is to recommend people do side projects, as opposed to recommending they don’t get a CS degree. Maybe another point is “don’t try to learn all the topics you heard about before you apply to any job”, which is also important.

• Yonatan Cale 6 Jul 2024 9:23 UTC

  2 points

  0 ∶ 0

  in reply to: Rebecca’s comment on: 80,000 hours should re­move OpenAI from the Job Board (and similar EA orgs should do similarly)

  1. If Conor thinks these roles are impactful then I’m happy we agree on listing impactful roles. (The discussion on whether alignment roles are impactful is separate from what I was trying to say in my comment)

  2. If the career development tag is used (and is clear to typical people using the job board) then—again—seems good to me.

• Yonatan Cale 4 Jul 2024 6:39 UTC

  11 points

  3 ∶ 2

  in reply to: Yonatan Cale’s comment on: 80,000 hours should re­move OpenAI from the Job Board (and similar EA orgs should do similarly)

  My own intuition on what to do with this situation—is to stop trying to change your reputation using disclaimers.

  There’s a lot of value in having a job board with high impact job recommendations. One of the challenging parts is getting a critical mass of people looking at your job board, and you already have that.