Dirak

Praying this isn't another Llama4 situation where the benchmark numbers are cooked. 84.6% on Arc-AGI is incredible!

No, just when Europeans (like yourself) need a reminder that they're not exceptional in their "level headedness" and "soberness", and in fact are anti-exceptional when it comes to real world outcomes like GDP.

This has never been the case across the history of humanity -- there has never existed a non-biased institution, and it's incredibly naive to think that a German think tank funded by the German government would be any exception.

"Level headed" and "sober" people are not immune to the effects of incentives and conflicts of interest. Researchers are dependent on grants, on invitations to conferences, etc., and so are liable to follow trends (tariffs bad), and p-hack to support the mainstream narrative (as they do in this analysis with P values > 0.01).

> German academia is actually full of sober, level-headed, nuanced people

Thanks for the laugh. I hope you realize how pretentious this sounds. In actuality, Germany's GDP is about 1/6th of that of the US, so these German academics don't sound very bright for how "level headed" and "sober" they are.

Today the S&P500 is at an all-time high, energy prices are at all-time lows, and there is no clear indication of recession. This, all in spite of the fact that for the last decade, economists have been crying that tariffs would ruin the American economy, trigger deep recession, etc.

The claim of this analysis is a significant backpedaling on the narrative they've been wrong about regarding the effects of tariffs. Instead of trying to reposition the goal posts on the effects of tariffs, it would be far more productive to simply acknowledge that the original dire predictions of tariffs did not manifest.

Blind person uses Meta Raybans glasses as a sight aid to help navigate her surroundings. For instance, asking the glasses what aisle of the supermarket she's looking at. Wanted to share this since it seems like a really cool usage of integrated LLM technology.

One should be suspicious of ulterior motives when the CEO of an AI company makes a claim like this.

On one hand, LLMs do require significant amounts of compute to train. But the other hand, if you amortize training costs across all user sessions, is it really that big a deal? And that’s not even factoring in Moore’s law and incremental improvements to model training efficiency.

> do you commit feature flags

It doesn’t really matter how you manage changes to feature flags, but using version control gives you a couple of nice benefits:

* gives developers the opportunity to describe their change

* let’s you roll back a problematic patch

* blame and bisect problematic patches

Ideally, you should also be able to see your feature flag changes in prod much faster than it takes to cut a release. You need this in order to be able to quickly roll back bad features.

> What feature flag tool do you use?

See https://engineering.fb.com/2017/08/31/web/rapid-release-at-m...

> You don't seem to understand what a conditional is; you'd have a hard time doing math or formal CS.

Ad hominems are against TOS.

Answering the question is irrelevant if it’s the wrong question to begin with. And for the wrong circumstances (ie if this is a family vacation) can be counter-productive. It’s far more productive to zoom out, ask why the author is asking for advice to study react on vacation, and address that instead. Consider this as you advance your career past recent college grad / swe1 :)

This is a classic example of the xy problem https://xyproblem.info/, where the author is asking for advice on his solution (how to study react on a vacation without internet), when for the vast majority of circumstances the best advice for the author would be to not waste his vacation studying React.

I feel like this is a solvable problem: 1) make feature flags be configured to have an expiration date. If over the expiration date, auto-generate a task to clean up your FF 2) If you want to be extra fancy, set up a codemod to automatically clean up the FF once it's expired

I don't see the problem with developers using flags for configuration as a stopgap until there's a better solution available.

Ads are not at all horrible. Ads help connect products with the niche markets of people that could benefit from them. Many cool products are too niche and would not be able to exist without targeted advertising. Yes, when done distastefully, ads can be quite horrible (ie ads for gambling and other vices), but when done well, both the user and advertiser benefit. This has to be true, otherwise, if users never found ads useful they would never click on them, and then advertisers would’ve never pay for advertisements since they wouldn’t benefit them.

Ads are human reviewed. Either you’re exceptionally unlucky with reviewers or there’s more to the story here.

> [At larger companies...] Talking to users is for PMs, silly! You stick to what you’re good at. At best you get a summary of user insights and a reasonable task priority list derived from it. At worst you get a confusing task list built off a mistaken understanding of users and the manager’s selfish vision, and no one can explain why each task matters.

Not necessarily. At the large companies I've worked at (>200 people), a UXR will drive the interview process with users, starting with compiling a list of questions from engineers, then conducting the interview sessions with users in which engineers can sit in, and then disseminating the insights and setting up a meeting to make the insights into a actionable engineering projects.

Working with UXRs is a dream, as they're trained to conduct interviews in an impartial way, leading to insights that are less biased and higher quality. Contrast to when I worked for startups and I or someone on my team would interview users, very often the feedback would end up contaminated because the interviewer would ask the wrong question, projected their biases into the questions, or even worse into the insights.

Building products and slowly grinding away at improving DAU is Meta's bread and butter. This has been the case with IG Stories vs Snap, now IG Reels vs Tiktok (IG Reels rev set to exceed tiktok as early as 2024 https://www.mbi-deepdives.com/meta2q23/). Meta is setting up the exact same playbook for Threads, and given their track record, I wouldn't bet against them.

This type of hiring is most likely reserved for Yann LeCun caliber talent. If you walk in to FB HQ without an appointment, 1) they wouldn’t let you in, and 2) you’d probably just make it awkward for the security guard

This is a perfect example of how GDPR can present challenges to innovation. The fact the top comment in this announcement revolves around GDPR compliance and associated fines raises questions about whether companies will be motivated to share research and open source datasets in the future.

In fact, I would go further and say we would actually be worse off without personalized advertising. Personalized ads help connect innovative products with the people who would benefit from them. Without a way to reach these niche audiences, these products might not be able to exist, and the community would be worse off.

1) This is the largest launch of a social media platform this decade. Higher growth than ChatGPT, and already over 50MM signups https://www.quiverquant.com/threadstracker/

2) Zero downtime on a day 1 launch at the scale is a very impressive technical feat. If it had been rushed or released prematurely, this level of app stability would not have been the case. If I had to guess, this would be the reason why Bluesky hasn't been able to release to the general public

In terms of the app being a foundation that will be built on and cannibalize Twitter, I'm very bullish on the future for Threads!

This has already be debunked https://www.threads.net/t/CuXRXDdNOtH/

You can "deactivate" your account and delete all of your posts individually, which is effectively equivalent to deleting your Threads account.

Tangential, but I think strong social skills are very important if you want to improve as a software engineer. Examples of strong social skill would be being able to set project direction, set up cross team collaborations, communicate impact, mentor and grow your colleagues, etc. Exclusively having strong technical skills can lead you down the "evil genius" archetype, where you're able to deliver technically complex software, but in a way that alienates contributors and collaborates (ie Haskell elitists).

Targeted Advertising serves a valuable role in free markets by giving exposure to innovative and novel products that improve people's lives. Without advertising, there would be no playing field for new products to challenge incumbents. Everybody would end up sticking to products they know without ever knowing about superior alternatives like in some kind of communist dystopia. Food for thought.

This is slightly off-topic, but this post makes me feel sincerely nostalgic for building games in Eclipse and Swing back in the day. Growing up, Notch's live streams[1] where he would code up a game in Java Swing for Ludum Dare were one of my first exposures to programming, and reasons for eventually studying CS. I wish there were some modern alternative to Swing that would be as simple as Swing, but would allow you to create games for the web.

[1] https://www.youtube.com/watch?v=4J_8HkQj9mU

That's a two way street. I equally wouldn't want to work with someone who makes sweeping generalizations.

> For moral reasons, I don't want to hire someone who worked at Facebook.

That's quite a hostile and closed minded take, but an understandable cope for someone who could never hire a FAANG engineer in the first place.

Depending on how successful your job searching is going, you should be able to negotiate offers with a postponed start date (say, 3 months in the future after your company folds). A proper job search can last one or two months, so your timing is great to start now.

Job searching while unemployed can be nice, but it's better to be safer than sorry and have a future job lined up.

Another thing to keep in mind is that not many companies hire during the holidays, so the best window you'll have to conduct your job search is now. Otherwise you're likely going to likely have to wait until January to begin job searching.

Remove half the jobs and drop the month from the start - end date range

When I worked there me and at least 2 other engineers I knew didn't have college degrees and joined at entry level. If a job listing says it requires certain credentials, I've mostly gotten by by just ignoring that and applying anyway.

Besides Pytorch, Tensorflow, gRPC, React, AWS, Kubernetes, GraphQL, C++ Swiss table, etc...?

You might be interested in checking out https://www.ggpo.net/. It's an open source C++ library for game networking that's used mainly in fighting games and it implements support for input prediction and rollback. They also have articles and documentation that cover the various implementations of netcode, how they're implemented, and their trade-offs.

I used what I learned from reading the articles on their site to implement my own input prediction and rollback netcode for my personal side project. It's a pretty useful resource, and helps organize my project in other ways too such as informing me of ways I can structure event messages, handle input valuation, etc.

Location: SF / NYC

Remote: Ideally no

Willing to relocate: Yes

Technologies: Professional: Typescript, React, GraphQL, gRPC, Postgres, Rust; For Fun: Go, Kotlin, Python

Résumé: fermidirak.github.io/resume.pdf

Email: manuelebryan[at]gmail[dot]com

I'm a full stack developer with 3.5 years of experience. My area of expertise is in frontend and in frontend tooling. In the past, I've worked on frontend infrastructure teams as well as product teams at startups and early stage companies. I like working on ambitious projects in interesting domains.

I find competitive programming to be creative and expressive. For most problems, there actually can be multiple ways to solve the same problem, with each approach having both hard (perf, memory) and soft (simplicity) trade offs. Leetcode problems remind me a lot of Celeste levels or rock climbing, where even though there might be an optimal solution, everyone solves the problem in their own unique way. No two solutions look the same. Even if everyone is solving the problem with the exact same algorithm, every submission has its own code style, and that's really cool.

I also find that Leetcode translates really well into the code I write in my freetime (a roguelike. Real work is boring fullstack crud). I personally hate reaching for libraries in my side projects, so being able to have these data structures and algos at my fingertips and being able to implement say A* or a weighted rng function or something else on the fly has been super handy. I honestly don't think I would be able to implement a vertex fragment shader w/ lighting without the skills I've picked up from Leetcode grind.