teawrecks
@teawrecks@sopuli.xyz
- Comment on Valve’s hero shooter Deadlock leaks with screenshots, gameplay details - Polygon 16 hours ago:
The only part of this that interests me is how they decide to handle anti-cheat on linux.
- Comment on Self-balancing commuter pods ride old railway lines on demand 16 hours ago:
Seems like a train that uses both sides of the track fulfills different requirements. A train can only be made to go one way at a time, but can hold more people (increased bandwidth), but these smaller half-cars can be moving people in both directions at the same time (lower latency). Seems quite clever if it works out.
- Comment on Let's discuss: Half-Life 1 day ago:
The original graphics, physics, and performance were incredible for the time, but to be fair, that’s not what you’re running when you download HL2 on steam today. The textures have been silently updated many times over the years. Your mind’s eye says “yeah, this is how I remember it”, and I’ve seen multiple streamers playing it for the first time thinking they’re seeing the original textures from 2004.
- Comment on Tesla is already pulling back Supercharger plans after firing team 2 weeks ago:
Maybe he’s looking to retire and the board wants to make whoever follows him look like a godsend.
- Comment on Why do mobile games suck nowadays? 2 weeks ago:
These were originally the exception, but all eventually followed the rule.
- Comment on A personal argument for a benefit of gaming 2 weeks ago:
Seems orthogonal. I don’t care how regulation is accomplished, just that it is. I feel like the tax levels and safety nets we had in place ~70 years ago were fine, until red scare propaganda convinced everyone to vote against their own interests.
Also I feel a bit like you’ve hijacked this discussion about the importance of video games in child development.
- Comment on A personal argument for a benefit of gaming 2 weeks ago:
In theory, capitalism is supposed to create exactly that environment. The problem is we have a society that doesn’t believe in proper regulation to prioritize the wellbeing of the society over that of the achievements and desires of individuals.
So much like how modern WoW has transformed into this uninteresting, solved meta that requires weakauras to do your thinking, gold buying to have gear and reagents, and no interesting competition for loot, our society is now an uninteresting solved meta where the wealthy nullify any possibility of competition, everyone is employed as wageslaves with a corporate handbook doing their thinking for them, and there’s no safety net to allow anyone take a chance at working together on interesting projects to actually compete.
The problem isn’t that we have capitalism, it’s that capitalism is synonymous with patriotism.
- Comment on A personal argument for a benefit of gaming 2 weeks ago:
I’m glad you can recognize how important this is to a kid. So many wow raiders in the 00s were ostracized by society for being this dedicated to a team of other humans and a shared goal. It really is something we need to learn to embrace and harness. I love the unique emotional responses that video games are capable of eliciting in people that movies and tv never could.
- Comment on Alan Wake II Has yet to Recoup Development and Marketing Expenses; Tencent Raised Stakes in Remedy to 14% 2 weeks ago:
I haven’t finished it yet, but AW2 is a dramatic step up in the entire experience. They still “pay homage” to the original combat, but there aren’t nearly as many enemies. If you’re familiar with the Control story and like that universe, I’d say it’s a must play.
- Comment on Alan Wake II Has yet to Recoup Development and Marketing Expenses; Tencent Raised Stakes in Remedy to 14% 2 weeks ago:
That’s wild. The combat in the first one felt mind numbing to me. Just constant padding with the same flashlight/gun combo over and over and over.
- Comment on Alan Wake II Has yet to Recoup Development and Marketing Expenses; Tencent Raised Stakes in Remedy to 14% 2 weeks ago:
I have to assume they did the math and concluded an epic exclusive was still the more lucrative option.
- Comment on Safety in typing, no cloud needed 2 weeks ago:
Cool, will do. It’s weird to me that open boards need to pull the swype binary. Is it really that hard to replicate?
- Comment on Safety in typing, no cloud needed 3 weeks ago:
Unfortunately (and incredibly) Gboard is the only keyboard that fits all my needs. I’m on graphene so I feel ok about just blocking its network access. This means voice transcription doesn’t work, but otherwise I get swipe, predictions, and other languages.
- Comment on Weekly “What are you playing” Thread || Week of April 21st 3 weeks ago:
Just finished Chants of Sennaar. Great game, highly recommend for people who liked Outer Wilds or Return of the Obra Dinn.
Now I’m trying out the new RimWorld Anomaly expansion.
- Comment on Someone got Gab's AI chatbot to show its instructions 4 weeks ago:
Yeah, I suppose you’re right. I incorrectly believed that a defining characteristic was the generation of natural language, but that’s just one feature it’s used for. TIL.
- Comment on Someone got Gab's AI chatbot to show its instructions 4 weeks ago:
Oh I see, you’re saying the training set is exclusively with yes/no answers. That’s called a classifier, not an LLM. But yeah, you might be able to make a reasonable “does this input and this output create a jailbreak for this set of instructions” classifier.
- Comment on Someone got Gab's AI chatbot to show its instructions 4 weeks ago:
Because it’s probibalistic and in this example the user’s input has been specifically crafted as the best possible jailbreak to get the output we want.
Unless we have actually appended a non-LLM filter at the end to only allow yes/no through, the possibility for it to output something other than yes/no, even though it was explicitly instructed to, is always there. Just like how in the Gab example it was told in many different ways to never repeat the instructions, it still did.
- Comment on Someone got Gab's AI chatbot to show its instructions 4 weeks ago:
Ah, TIL about instruction fine-tuning. Thanks, interesting thread.
Still, as I understand it, if the model has seen an input, then it always has a non-zero chance of reproducing it in the output.
- Comment on Someone got Gab's AI chatbot to show its instructions 4 weeks ago:
Any input to the 2nd LLM is a prompt, so if it sees the user input, then it affects the probabilities of the output.
There’s no such thing as “training an AI to follow instructions”. The output is just a probibalistic function of the input. This is why a jailbreak is always possible, the probability of getting it to output something that was given as input is never 0.
- Comment on Someone got Gab's AI chatbot to show its instructions 4 weeks ago:
Yeah, as soon as you feed the user input into the 2nd one, you’ve created the potential to jailbreak it as well. You could possibly even convince the 2nd one to jailbreak the first one for you, or If it has also seen the instructions to the first one, you just need to jailbreak the first.
This is all so hypothetical, and probabilistic, and hyper-applicable to today’s LLMs that I’d just want to try it. But I do think it’s possible, given the paper mentioned up at the top of this thread.
- Comment on Someone got Gab's AI chatbot to show its instructions 4 weeks ago:
Oh, I misread your original comment. I thought you meant looking at the user’s input and trying to determine if it was a jailbreak.
Then I think the way around it would be to ask the LLM to encode it some way that the 2nd LLM wouldn’t pick up on. Maybe it could rot13 encode it, or you provide a key to XOR with everything. Or since they’re usually bad at math, maybe something like pig latin, or that thing where you shuffle the interior letters of each word, but keep the first/last the same? Would have to try it out, but I think you could find a way. Eventually, if the AI is smart enough, it probably just reduces to Diffie-Hellman lol. But then maybe the AI is smart enough to not be fooled by a jailbreak.
- Comment on Someone got Gab's AI chatbot to show its instructions 4 weeks ago:
I think if the 2nd LLM has ever seen the actual prompt, then no, you could just jailbreak the 2nd LLM too. But you may be able to create a bot that is really good at spotting jailbreak-type prompts in general, and then prevent it from going through to the primary one. I also assume I’m not the first to come up with this and OpenAI knows exactly how well this fares.
- Comment on Apple users "don’t know what is going on": New study shows that Apple's default apps collect data even when supposedly disabled, and this is hard to switch off 4 weeks ago:
I’ve been running graphene for almost a year I think, and I haven’t had to make any functional sacrifices. Though I was already not using Google’s voice assistant features.
- Comment on Apple users "don’t know what is going on": New study shows that Apple's default apps collect data even when supposedly disabled, and this is hard to switch off 4 weeks ago:
Depends on whether you consider anti-patterns to be “lying”. If a normal user would reasonably think that data isn’t being collected based on the settings they chose, then is it dishonest for them to still be collecting data? Is it good enough for them to say “well we technically never said that disabling X disabled all the invasive functionality needed to do X.”
- Comment on Apple users "don’t know what is going on": New study shows that Apple's default apps collect data even when supposedly disabled, and this is hard to switch off 4 weeks ago:
Obviously a Google phone will be at least as bad, but I’m curious how de-googled Android forks like graphene would fare.
- Comment on Discord is nuking Nintendo Switch emulator devs and their entire servers 5 weeks ago:
I think their point was, all things being equal, a server on discord vs a community on a Lemmy instance doesn’t make a difference. In both cases, the people who ultimately own the platform have to decide whether to just delete them or go toe-to-toe with Nintendo in court.
Hosting everything themselves is a different story. Though…is it possible for a federated instance to exist inside the tor network? Maybe that’s already a huge thing and it never occurred to me.
- Comment on Kobo's new color E Ink eReaders start at only $150 5 weeks ago:
We’re looking at 40 days based on 30 minutes of daily reading at 30% brightness, and with Wi-Fi and Bluetooth turned off. Dropping the brightness down to 10% nets an increase to 53 total days of runtime.
So as long as you don’t use it, the battery will last a long time!
- Comment on Steam is a ticking time bomb 1 month ago:
Yeah, totally agree that we shouldn’t go all in on trusting valve, but apple is definitely the anti-consumer one here. I don’t think valve would support DX if they could get away with it. Apple deprecating everything but metal without making it an open spec basically said, “we don’t want anyone gaming on our platform”.
- Comment on Roblox Studio boss: children making money on the platform isn't exploitation, it's a gift 1 month ago:
“The children yern for the mines!”
- Comment on A very late "Weekly “What are you playing” Thread || Week of March 31st" 1 month ago:
Just finished Rain World.
What a brave game to make. It is not afraid to scare players away. I admit, I ended up having to look up a map in order to find the ending before I threw my PC out the window entirely, but acknowledge I was not in the right mindset to be playing. You cannot play the game to finish it, you must enjoy the gameplay for what it is, because it is not going to funnel you through the story at all, and you’re going to have a LOT of deaths that feel like total bullshit.
But the atmosphere, and the sound design, and the art, and the creatures, and their AI, and the world building are all top tier. I don’t know if I can recommend anyone play it, but it is a very well made game.