yokto

One of the approaches I'm taking is to try hard to simplify the code generated by LLMs. If you can reduce its complexity, it usually means you have a fairly good idea of what's going on in the code change. I consider which edge cases are getting unnecessarily handled, which new vocabulary was introduced where existing or simpler vocabulary would be fine, etc. etc.

I also have a /simplification-check skill I run after pretty much every large change, and that already helps surface the murky areas of the code and sometimes remove complexity before I even start reviewing, thus accelerating the review.

Will give it a try! How is LiveKit working for you in terms of achieving low input2video latency? Did it lend itself to getting tuned for this application?

However, not paying does not mean that you necessarily become "the product".

Offering loss-leading products is a common business strategy and I'm perfectly okay taking advantage of this for hosting needs that fit a vendor's free tier.

Static hosting is very cheap to offer and gathering free tier users will gain developer mind-share, which helps the vendor can sell their truly money-making products. To pick a clear-cut example: AWS Lambda does not collect any additional data on free tier usage compared to paid usage.

I would love to try it, can I sign up for a Linux desktop app notify-list?

It's most likely two or more separate attackers operating. The first malware, Shai Hulud 2, exfiltrates credentials from the infected dev machine to new public GitHub repositories. As the repositories are public and searchable via GitHub's interfaces, any malicious attacker aware of the attack can easily grab the credentials and launch any attack, whether it's a noisy destructive script or some sophisticated ransomware.

Yarn has a configurable npmMinimalAgeGate setting as well since version 4.10: https://yarnpkg.com/configuration/yarnrc#npmMinimalAgeGate

That's my bad, apologies. This job post will be published next week and I somehow removed this mention from my comment while editing it. I've put it back now. In the meantime, feel free to apply through the Platform Engineer role while stating your targeted role and I will re-categorize your application later. Thank you!

This is (in part) what "world models" are about. While some companies like Tesla bring together a fleet of small specialised models, others like CommaAI and Wayve train generalist models.

What issues did you encounter with WorkOS? I'm currently evaluating it and would love to avoid repeating knowable mistakes.

It does, shockingly well in my experience. Check out this blog post outlining such an approach, called Literate Development by the author: https://news.ycombinator.com/item?id=43524673

Whenever an LLM struggles with a particular library version, I use Cursor Rules to auto-include migration information and that generally worked well enough in my cases.

This reminds me of this quote I love:

> "[They] placed too much weight on the introspections that they generated at that moment in time, and thus lost sight of their more enduring attitudes.” [1]

The quote refers to this study [2] in which subjects had to chose a poster to take home. The group who was instructed to think about their reasons for their initial choice, and had the option to change it, were less satisfied with it three weeks later. As the abstract says:

> When people think about reasons, they appear to focus on attributes of the stimulus that are easy to verbalize and seem like plausible reasons but may not be important causes of their initial evaluations.

This suggests that satisfaction is more correlated with initial gut feeling than reasoning, at least for aesthetic choices, but I think in many other cases as well.

[1] https://sci-hub.st/10.1016/S0065-2601(08)00401-2

[2] https://journals.sagepub.com/doi/abs/10.1177/014616729319301...

Not a reading guide, but Computerphile have a good introductory video on Gaussian Splats: https://youtu.be/VkIJbpdTujE?si=8hoMbMx6tKuMZo2S

Individualkex also has a couple videos on the high level ideas: https://youtu.be/GQXDjzNWuPc?si=zlAN7dO9STGATKad

It generates the pixels, including the blurry UI at the bottom.

It is, but only within the same origin, which already enables users to not re-download jquery.js or Google Fonts if they previously visited another website that downloaded the same file from the same (usually cross-) origin.

fal.ai is nice and fast: https://news.ycombinator.com/item?id=39360800 Both in performance and for how quickly they integrate new models apparently: they already support Stable Cascade.

I'm not sure how relevant that threat model is (OS level security would probably be enabled for people susceptible to be targeted in such a way. Support could advise to do it before toggling the flag, etc.), but anyway the hypothetical flag would only be about making sure the automation doesn't happen and the ticket goes to support. Support can then manually handle the rare edge case and place more burden on the person attempting to deactivate the account.

What is the abuse your referring to?

With your suggested approach, the attacker is free to use the account to impersonate the victim until they get a new SIM card, which could easily take days or weeks.

This seems like a degredation compared to the current abuse potential which is mostly limited to logging you out.

This is trivial to mitigate with per-account rate limiting.

On top of that, if a specific account is targeted at the rate-limit, a flag could be put in place to let support disable the automation for that account.

Whoops, my comment isn't very clear, sorry. I meant: "but their account would still be active and in the hands of the thief, if there is no way to quickly deactivate it, e.g. before receiving a new SIM card from their operator that would enable you to prove your identity to WhatsApp."

As YetAnotherNick said, logout might be the better word to describe the impact here (plus, a fairly aggressive inactivity deletion period).

I agree with you in principle, but I still don’t understand how else to mitigate this: WhatsApp must get a lot of cases of stolen unprotected phones. The victim can ask their operator to lock the SIM card, but their WhatsApp account would still be out in the open.

With the continuous improvements in mobile OS security defaults, I’d expect this scenario to become less and less of a problem, but it must still be accounted for.

The process still goes through support ticketing, so I’d expect a spike to be noticed and stopped.

Isn't this flow what more ore less what you would expect? Could someone suggest what would be the appropriate alternative here?

- The inconvenience to the deactivated account is minor: one SMS verification code and the account is back, queued messages get received, etc.

- Persons who lost their phones probably don't have a good fast way of proving their identity, as their identity is tied to their phone number in WhatsApp's model.

- Needing to quickly lock out spammers, thiefs or hackers is probably far more frequent than abuse of this feature.

- If abuse of this feature becomes a recurring problem, I'd expect WhatsApp to react and adjust the flow to place more burden on its user.

The auto-delete part is slightly more worrying, but if you don't use WhatsApp during 30 days, your account and group membership probably isn't very precious. Backups are automated and separate. You can still easily re-create an account with the same number then.

The story might be "Apps should stop using SMS and phones numbers as the source of identity", and while I generally agree, most comments don't seem to be about this and WhatsApp is maybe _the_ one app whose success was based on this very idea.

To offer a counter datapoint: I moved from engineer to manager while tracking my hours fairly accurately. I can pull off 50 hours of development, including late night emergencies, some client pressure, etc., and still feel energized the weekend, while after 40 hours of management, meetings and firefighting, I become mostly useless and have to take some serious breaks. I enjoy both.

However, there are so many parameters affecting this result that I wouldn't dare to make a call on which role is more tiring in general at a regular company.

- I'm less experienced as a manager, I have to continuously learn a lot and grow fast.

- This is in a young start-up context.

- This is with a very broad scope of responsibilities.

- I'm me.

- etc.

YMMV

The largest fatigue factor to me seems the amount of context switch I have to do as a manager.

The text chat experience is something that has been so refined over the last 10 years that to meet today's expectations towards it, you need a very surprising amount of engineering.

It's something I didn't realized until I started coding it: every detail that you omit from your chat UX is a tragically noticable pain to the user used to WhatsApp and Slack 100 times a day.

I've been paying for Raindrop for a few years and its organisation capabilities are really good: tags, collections, folders, search, etc. All in a quite polished UI!

Nice! The rendering is more polished than what I did a few years ago in another similar project: apply CycleGAN style transfer on Google Maps / OpenStreetMap tiles to redraw them interactively as old hand-drawn maps https://github.com/nathanvogel/deepmaps

Hands are notoriously hard to even photograph. You very quickly get weird unnatural results with a camera in front of hands, so in a way I'm not surprised AI models struggle to produce satisfying imagery there too.

If you like TypeScript/JavaScript, I recently discovered RedwoodJS [1], and while I have never tried it, it looks like a simple and rather complete framework for basic CRUD apps, while leaving room for growth.

[1] https://redwoodjs.com/

Just F2, but I had to get used to that already with my previous keyboard (Microsoft Sculpt). But I let all F keys bound to the Layer 2 as per the ZSA defaults.

And Alt + F4 is bound to a double tap of the right red thumb key :D

My ZSA Moonlander keyboard [1].

When I bought it, I was just looking for a sleeker and more ergonomic keyboard with a split design, but the ability to easily reconfigure every key on the layout brought a new meaning to the word "ergonomic" for me.

It means that when a particular motion or shortcut that I frequently use is puts too much strain on my hands, I can simply change the layout to make the keys more natural too use. And it's just an overall incredibly well made product.

[1] https://www.zsa.io/moonlander/