TechBro8615

IIRC, Twilio offers a paid STUN/TURN service. But you could also self-host it using some open source TURN server, since all it requires is a public IP address (with the caveat of course that it might not be geographically close to the end user).

Thanks. I assumed it was Margin of Error. The article doesn't expand the acronym until midway through the post, where it appears almost accidentally. Perhaps the intended audience is a mixture of experts, of which I'm not a part.

I don't mind reading research papers, but they're really annoying to read on a phone screen. I remember a few years ago, an HN comment shared a link to some tool that could convert a PDF to single column text and make it more readable on a phone screen, but I can't find it. Anyone remember this or have the link?

I can type just as fast (if not faster) on my phone as on desktop. I write almost all my HN comments from my phone, primarily because it's my leisure device and I try to use my desktop only for work (or at least "work") activities. My phone has a high DPI screen and a responsive touch interface that makes it easy to browse content and write comments. The main downside is that it's more immersive and usually can't display as much content on the screen as a desktop monitor, but for this I mostly blame responsive design and its tendency to favor single-column text rendering; reading sites in desktop mode on a high DPI phone screen actually isn't too bad, aside from the tiny buttons and links. But the reason this is a major downside is that it encourages doomscrolling as the screen typically only displays one "piece" of content at a time.

I expect they'll lose value in the step-function shape that's typical of technology, where each new generation decreases the value of the previous one. But that's fine, as long as each chip generates more revenue over its lifetime than it cost. And the lifetime can extend through multiple generations, as long as each successive generation is only a marginal improvement over its predecessor.

Personally I think the bigger risk is software innovations making CPU training (and/or inference) sufficiently viable that it's cheaper to train models on a commodity CPU cluster than on some proportionally expensive GPU cluster. I don't know enough about the space to say whether that's likely, but it seems like a low risk, since pretty much any parallel algorithm will always be faster on GPU than CPU - it's just a question of the marginal benefits and cost (e.g. maybe it takes more CPU to train same model in same time, but cost of CPU is so much lower that it's worth buying more of them).

I love mitmproxy. It's been consistently part of my toolbelt for nearly a decade, it's stable, and still has plenty of new features that I can use if I need them, but it still works the same for its core use case as it did ten years ago.

Were they ever active in the first place? I only signed up to claim my account (which I later realized was pointless since the name is reserved by your Instagram handle anyway).

I'm perfectly willing to believe usage has flatlined, and that matches my personal experience where I've opened it maybe twice since installing it, only to see posts from mostly brands, celebrities, influencers and meme pages. There was an initial surge where I got a notification every time one of my Instagram followers also followed me on Threads, but that only lasted about a week and it's been quiet since then. I suspect a lot of people were behaving similarly to me, either signing up to claim their account or opening the app when they got a notification of a new follower.

But I'm skeptical of the utility of comparing the flatline to a baseline from the first week, and of using data from Sensor Tower and Similarweb, neither of which has much insight into usage data beyond what's publicly available through metrics like app installs, reviews and maybe DNS. It's not like Meta is using any analytics frameworks other than their own.

Is this normalized by number of each demographic that sits for the test? (Sorry if that's answered in the PDF but maybe someone with more time or statistical understanding than me can check it.)

That's because Europe doesn't have as many world renowned institutions. And for the few that it does have, people absolutely care. There's a difference between EFPL and University of Delft, or Oxford and University of Southampton...

If you don't believe me just look at where members of your government went to school.

Relax. There'd be no need for presumptions if you wrote more clearly than "phone calls is 99% of what I do," unless you're suggesting it's a reasonable presumption that you spend 99% of your personal time on the phone, or that "what I do" doesn't colloquially mean "for work."

It sounds like a dumb gimmick, but I don't see anything morally wrong with it as long as the interns are aware of the deal before they accept the offer to work there.

Also, is it actually public? Or just shared with everyone inside the firm? If it's public, anyone got a link to the leaderboard? And if it's not public, then it sounds less like a gimmick, and more like a transparent stack ranking incentive program.

Predicted [0] three years ago by user A4ET8a8uTh0 and myself (I've also written similar comments for the past six years).

[0] https://news.ycombinator.com/item?id=23324576

What do you mean calls is 99% of what you do? For work? Doesn't that just prove my point? You're not using the phone system for personal calls. You must work somewhere doing outbound calls for some reason or receiving them for another. And yes, I would include sales in this classification.

I haven't answered the phone in years. All calls from unknown numbers go straight to silent, and anyone in my contacts knows not to call me anyway. What kind of real person uses the phone nowadays? It's literally scams, customer support, billing departments, credit collection agencies, and generally people you don't talk to as long as you can avoid it. Normal people use FaceTime, WhatsApp, Messenger, Zoom, etc.

There's basically zero upside to phone calls, and frankly nor is there much for SMS either. I would suggest that phone companies completely deprecate their telecom networks and go all-in on internet, but then the spam would migrate to the apps (it's already begun with whatsapp), so I don't want to give them any ideas.

You're unemployed and have $25,000 in credit card debt. This is some impressive post-hoc rationalization, but it would have more credibility if you wrote about it before you went into significant personal debt. Otherwise it just reads like cope.

And btw, I've been where the author is, so I recognize the signs. Let's check back in three years.

My advice: get a job. (I see the author is currently an intern in college. So I'll amend that advice to be: line up a job for after graduation.)

And/or maybe demand has decreased as people have settled on alternative boards...

Does enabling a VPN disable private relay or does it tunnel to the VPN through private relay? For some reason I have it in my head that it "double wraps" the traffic but I've never actually confirmed that for a fact.

I felt the same, but I also feel like it was the perfect (least bad) time in my life for it to happen. I can't imagine being in the middle of High School or University during Covid - those kids got the absolute worst end of the deal. I was also recently single, and counted my days feeling grateful I was not living with my ex when the pandemic hit.

> because they trained with data derived from chatGPT

Can you expand on this (genuinely curious)? Did Facebook use ChatGPT during the fine-tuning process for llama, or are you referring to independent developers doing their own fine-tuning of the models?

I can empathize with the author. The first time you write some code collaboratively with GPT and it actually works, you feel a burning need to shout about it. Because it's one of those moments where something "clicks" and you suddenly feel like you've discovered fire. Once you figure out how to work with them, it makes you excited for the future and you can clearly see where LLMs will fit permanently into your toolbelt. They're far from perfect now, and sometimes the time savings is a wash - you get instant specialized expertise that can produce code like a senior engineer, but you need to goad and coax it like it's a high maintenance intern. But the thinking power expended is still somehow lower - it's a new way of working with technology and deferring some of the grueling parts to the machine. This becomes especially obvious when the code requirements depend on an esoteric API or conventions that you'd normally need to spend time researching and manually enumerating.

First, it's worth noting the code in the blog post is not "production code," but rather one-off or periodically used scripts for accelerating manual business processes, with results that are easy to manually check.

But in regards to production code, I agree. When code is committed to a codebase, a human should review it. Assuming you trust your review process, it shouldn't matter whether the code submitted for review was written by a human or a language model. If it does make a difference, then your review process is already broken. It should catch bad code regardless of whether it was created by human or machine.

It's still worth knowing the source of commits, but only for context in understanding how it was generated. You know humans are likely to make certain classes of error, and you can learn to watch out for the blind spots of your teammates, just like you can learn the idiosyncrasies and weak points of GPT generated code.

Personally, I don't think we're quite at "ask GPT to commit directly to the repo," but we're getting close. The constant refrain of "try GPT-4" has become a trope, but the difference is immediately noticeable. Whereas GPT-3.5 will make a mistake or two in every 50 line file, GPT-4 is capable of producing fully correct code that you can immediately run successfully. At the moment it works best for isolated prompts like "create a component to do X," or "write a script to do Y," but if you can provide it with the interface to call an external function, then suddenly that isolated code is just another part of an existing system.

As tooling improves for working collaboratively with large language models and providing them with realtime contextual feedback of code correctness (especially for statically analyzeble or type-checked languages), they will become increasingly indispensable to the workflow of productive developers. If you haven't used co-pilot yet, I encourage you to try it for at least a month. You'll develop an intuition for what it's capable of and will eventually wonder how you ever coded without it. Also make sure to try prompting GPT-4 to create functions, components or scripts. The results are truly surprising and exciting.

This guy is operating a profitable business, creating value for customers, shipping features, and openly publishing details and learnings about the threats he mitigated. He used ChatGPT to generate scripts to help him throughout this process. I don't know if he's non-technical or if he just wanted to save some time, but frankly he should be commended for his hustle and get-shit-done attitude. These scripts were not determining life or death, or even making business critical decisions - they were filtering bulk data and making his life easier by producing results that are easily manually checkable, but save tons of time either coding the scripts or hiring a programmer to write them.

To me it reads like a great example of where ChatGPT is most useful: as a force multiplier for time-constrained entrepreneurs who have a specific goal and need specialized knowledge for short periods of time (e.g. to write a script). It's now basically free and instant to produce what would previously require a multi-week process of sourcing, hiring and communicating with contractors to write a script that leads to the same end result.

The kneejerk reaction to call this "surprising" or irresponsible, while understandable, gives major "get off my lawn" energy. This is the future and as coders we should support the increased self-sufficiency of non-technical people. If you want to adapt to the change then maybe think about how to improve the process for entrepreneurs of asking ChatGPT to write a script.

Cloudflare is one of the operators of egress nodes for iCloud Private Relay, along with Akamai and Fastly and maybe some others (IIRC). So you probably have issues with Archive.is when you're connected through Cloudflare. Although I'm not sure how DNS resolution works on private relay (since it's the DNS server rather than the egress proxy that causes the issue - but if the egress proxy is making the DNS requests then that would explain it).

I think you're taking their choice of a single word a bit too seriously, and dare I say, personally. I don't think they're claiming to be the first to have ever discovered this attack vector, nor are they trying to steal credit from you. And while "novel" might not be the best word choice, in common parlance it need not mean "unique," and can just as well mean "unusual." In fact the dictionary definition of "novel" literally includes "unusual" ("new or unusual in an interesting way").

Probably. I got distracted by the game so I never found out. And I only visit those websites when I'm watching a stream anyway. Really it's quite a clever distribution vector, since it involves long sessions and any experts in the audience will be too distracted to look too deeply, or at least unmotivated to investigate and publish research against a site that helps them watch the games they want to watch. It's a symbiotic relationship in a way...

I did notice the ad serving infrastructure seemed quite sophisticated. There were so many domains and proxies and redirects. Luckily uBlock Origin blocks almost all of them. And usually, I can avoid any of the "bonus" features by opening the video player iframe in its own tab (but sometimes this isn't possible, or the video player tab has some scripts to make it annoying to run in isolation).

One thing I like to do during the commercial breaks is paste the URL of the site into GitHub Code Search. This always leads to interesting results, including blocklists, people's personal media scrapers, or sometimes even the (re-)publishing infrastructure of the sites themselves. It's also a great way to find alternative URLs or other streaming sites.

I believe much of economics can be summed up as: most people don't want to do most things. (This can probably be rephrased to align with Maslow's hierarchy of needs.)

Also note that as far as the sketchiness scale goes, this is basically a 5/10. Now consider the same tools in the hands of malware distributors. For example, I've seen these anti-debugger techniques on NFL piracy websites when I tried to investigate why my CPU was pinned to 100% while I was streaming the game.

I'm confused about the comment referring to a WebCam? What does that have to do with 2FA?

FYI you can create a virtual WebAuthn device in Chrome DevTools and use that as a 2FA method (just don't lose it lol). I did this with Google where it was required to provide a phone number or hardware token in order to unlock the ability to switch to a simple TOTP method. I created the WebAuthn device, generated a TOTP, set TOTP as the default method, and then removed the WebAuthn device.

I'm not sure this applies to the GitHub situation (I am having trouble understanding the linked forum post), but it's a useful tip regardless.

In software, projects take years and they usually send a few guys to work on it for eight hours every day. A few years later they suddenly decide to rewrite everything and never finish it.

Congrats on the launch. I just want to say that I love the name :) It's fun, memorable, evocative of the product's purpose, and rolls off the tongue.