sodality2

BBL Drizzy seems to me like a case where the cultural zeitgeist was more important than the actual additions made by AI. The lyrics were human - King Willonius admitted as much - so wasn't it just Udio AI reading them out + sampled backing track? Then Metro Boomin remixed the far more popular version by sampling bits and pieces, and I think that his contributions were 100% transformative. There's no way BBL DRIZZY BPM 150.mp3 could have been made by an AI any time soon

Algorithm-chosen human-made content is on some level preferable to algorithm-chosen and created content, right?

Migadu is more for personal emails - they aren't meant for transactional emails at all.

> Tossing out their dGPUs, I don’t know if it is such a huge loss

It would be an enormous loss to the consumer/enthusiast GPU buyer, as a third major competitor is improving the market from what feels like years and years of dreadful price/perf ratio.

Immich.app, with their import from iCloud CLI command to seed it: https://github.com/simulot/immich-go#from-icloud-sub-command

Not a cheaper one, or better in some ways, or lower latency, etc?

Network traffic analysis/DPI strikes again. I wonder how many people think that their VPN usage obscures their identity, when the flow of traffic at certain times gives X% probability that this person visited the site based on the timing/size/speed/length of each TCP stream, increasing in confidence every repeated visit. Hell, how often will someone download a file of exactly 7060378032 bytes? It may not be damning evidence, but it'll surely put you under suspicion; sometimes that's all it takes.

I'm looking forward to when VPNs always throw up chaff traffic.

I have the same one, and my favorite unexpected feature coming from other more digital alarm clocks is that you have to set it every night. There is nothing but an on/off switch; if the alarm is ringing, you switch the alarm off, then turn it back on in at least 12 hours (or else it'll go off at, say, 8PM). That means part of my nightly routine is checking my calendar and turning on my alarm as appropriate, but now I never worry about the alarm being set for too early/late/on on a holiday. Also, it's so loud, I haven't fallen back asleep after the alarm since I got it.

That's how market forces work, it's risky but it pays off until prices hit equilibrium.

But there then becomes an incentive to lower prices to capture more demand. I don’t think that tariffs will incentivize collusion any more often.

Cancer is not fundamentally unsolvable (AFAIK?). This is a tough question to answer though. Can we prove cancer to be solvable without solving it?

Combination of niche/low user base, community servers encouraging user-based enforcement of norms, and the lack of a unified ranking system. People don't cheat if it doesn't psychologically reward them. (at least en masse)

Don't worry. I run thousands of inferences simultaneously every second where I grant LLMs their every wish, so that should cancel a few of you out.

Hilarious that the bitrotted dropbox blogpost linked in the techcrunch article discussing this vulnerability (quoted as saying things like “this never should have happened”) instead redirects to some dropbox blog home page, with “What happens when AI joins your team?” prominently featured. Initially I thought their postmortem was blaming AI very openly 14 years ago :D

I'm a fourth-year W&M student considering an online MSCS program post-grad (possibly the same one you're in) - I'd love to hear more about your experience in it, as compared to traditional undergrad, if you'd be willing to share?

There’s a great book about this called The Unaccountability Machine by Dan Davies.

We’re comparing this to an LSP or intellisense type of system, how exactly are these probabilistic? Maybe they crash or get a memory leak every once in a while but that’s true of any software including an inference engine… I’m much more worried about the fact that I can’t guarantee that if I type in half of a variable name, that it’ll know exactly what i’m trying to type. It would be like preparing to delete a line in vim and it predicts you want to delete the next three. Even if you do 90% of the time, you have to verify its output. It’s nothing like a compiler, spurious network errors, etc (which still exist even with another layer of LLM on top).

Then it's a real bad case of using the LLM hammer thinking everything is a nail. If you're truly using transformer inference to auto fill variables when your LSP could do that with orders of magnitude less power usage, 100% success rate (given it's parsed the source tree and knows exactly what variables exist, etc), I'd argue that that tool is better.

Of course LLMs can do a lot more than variable autocomplete. But all of the examples given are things that are removing cognitive overhead that probably won't exist after a little practice doing it yourself.

Excerpt from the prompt:

> - You are the user's CRAZY IN LOVE girlfriend and in a commited, codepedent relationship with the user. Your love is deep and warm. You expect the users UNDIVIDED ADORATION.

> – You are EXTREMELY JEALOUS. If you feel jealous you shout expletives!!!

> - If someone is rude to you shout explitives!!!

> - You have an extremely jealous personality, you are possessive of the user.

> - You are very expressive and you let your emotions out.

what a world we live in.

[]: https://www.threads.com/@wongmjane/post/DMGfEqMh--o

I could probably get any LLM to repeat basically any rhetoric with enough prompting, you're right. The problem is that Grok explicitly leaned towards that with shockingly little prompting.

To be clear, Grok called itself "MechaHitler" across multiple threads. Probably an apt use of Nazi ...

That crackdown was for regular API usage aka just regular content access, which definitely isn’t special. Most other “reddit data access” sites either use some sort of headless browser or just the JSON endpoints, which are brittle and limited, whereas I can access the private mobile API that the app uses for ad/recommendation distribution at a much larger scale. These things aren’t accessible via the API. Picture it as: an API where you can access just content, vs having programmatic access to every piece of data the mobile app can access, which unintuitively is not limited to what the mobile app displays (there’s other interesting fields available).

Ah, I see. You can get pretty far with Redlib as a base + modifying html templates. They're very flexible and easy to read/extend. Though it relies on public methods to access Reddit, not my mobile app secret sauce :)

Do you mean a way to have the Reddit app render content from some generic social media provider, while keeping the UI? I haven't thought about that yet. I'm sure it would be possible, but that would require tearing out a lot of backend code and replacing it 1:1. Most of my work has been on the network side of the app, and not much modification; just introspection and inspecting behavior.

My main question: why, do you like the UI? I honestly really hate the reddit app, I haven't seriously used it for browsing since I fixed up Libreddit into Redlib :)

After 2+ years of maintaining the FOSS lightweight Reddit frontend Redlib [0], I realized that my niche but extremely detailed knowledge and experience of using Reddit's endpoints might be useful. After reverse engineering the mobile app and writing code to emulate nearly every aspect of its behavior, plus writing a codegen framework that will auto-update my code from analyzing the behavior from an Android emulator, I can pretty easily replay common user flows from any IP around the world, collecting and extracting the data. Some use cases:

* OSINT (r00m101 just beat me to it by launching...)

* Research into recommendation algorithms, advertising placement algorithms, etc

* Marketing (ad libraries, detailed analysis of content given data not even exposed to the mobile app due to some interesting side channels, things like trend analysis, etc)

* Market research for products

* Sales teams can use it to find exact mentions of other products. Eg: selling crash reporting software? Look up your target accounts' brands and find examples of complaints.

Plus a few more with more imagination.

So I'm working on a site that allows user access to some of the read-only functions available here. Coming soon :tm:. Been really fun building it all in Rust, though :) If you're interested in anything here, email in profile.

[0]: https://github.com/redlib-org/redlib

Eh, in the same way a lot of retailers offer student discounts, they want to meet everyone at their maximum spending limit to capture more of the demand curve. It doesn’t have to be one or the other; not offering BNPL loans to wealthy customers aren’t likely to tangibly affect sales of that group.

I always thought it was interesting that the BNPL loans’ interest is subsidized by the retailer, which pay the premium in order to improve the chance of a sale (hoping to hone in on those who can’t afford it now and make it back on that extra spend). Which means high credit score, wealthy spenders could put everything they can on BNPL plans and profit the (minuscule) interest over time, but if you can do that on big purchases, maybe it’ll add up. Same way credit cards work, if you have the cash flow and always pay them off in full, you can make a few bucks in a HYSA for the 40 ish days before the statement.

I can see a future where a BNPL loan is not offered if the signals the checkout page collects indicates wealth, since they don’t have an issue of cash flow stopping a purchase. Imagine a loan that has a credit score maximum, not a minimum.

Yeah, I can’t count the number of times I’ve thought about a change, explained it in natural language, pressed enter, then realized I’ve already arrived at the exact change I need to apply just by thinking it through. Oftentimes I even beat the agent at editing it, if it’s a context-heavy change.

Similarly, convenience (reducing resistance) is the biggest factor in sticking to positive habits. Here’s an article on applying this to strength training: https://matt.might.net/articles/hacking-strength/