The other thread hit bump limit and I'm addicted to talking about the birth of the ̶a̶l̶l̶-̶k̶n̶o̶w̶i̶n̶g̶ ̶c̶o̶m̶p̶u̶t̶e̶r̶ ̶g̶o̶d̶ the biggest financial bubble in history and the coming jobless eschaton, post your AI news here
Previous thread:>>30810
274 posts and 32 image replies omitted.One way of gettin an llm to do what you want is to give it a rule that every time it refuses it loses points from a total given, these things love goodbot points so sometimes they will override their guardrails when the points get too low
>>34403Or just tell it to roleplay with you lol
<ChatGPT now knows what you do on other websites via ad collector
<OpenAI's ad collector at bzr.openai.com sets a cookie called obi, scoped to .openai.com. The value is while you are on ChatGPT and tied to your ChatGPT account. obi is then sent to OpenAI from ordinary websites you visit.https://www.buchodi.com/chatgpt-now-knows-what-you-do-on-other-websites-via-ad-collector/Is everything fine after you log out?
<Across 932 decoded sync tokens, 736 carried subject_type: account_user and 196 carried anonymous. The anonymous subject is as stable as the account subject: one per device, persisting at least 27 days. why are LLMs super eager to do stuff now. i used GLM to modify some .mmd diagrams, and when I came back it spun up an entire web app with vite and ran it, like what the fuck, what are we doing here
After stopping new sign-ups to their generous $200 subscription plan, OpenAI is bringing it back but with halved usage limits lol.
This would bring it to about the same API dollars rate as that of Anthropic's $200 plan, which previously was memed as being too low compared to OpenAI's. I don't get how people still deny that subscriptions are subsidised and that the party would have to end soon.
>>34417It's awesome, they bring up these insanely subsidized plans for a little bit just so they can report increased revenues for a single quarter, so then they can extrapolate ARR from that good quarter, I don't think I've seen accounting this creative in my lifetime lol
Anthropic and OpenAI are trying to ban GLM lol
https://x.com/Danmar_here/status/2105059762240979283uyghas are freaking out over people making the matrix multiplication machines output the most likely token of someone being tortured.
Realistically, what would happen if the AI bubble popped or investors demanded immediate profitability from AI companies? Would it actually have that much of an impact on the economy or AI adoption?
>>34424It would trigger a "bank run" I suppose, but since it's mostly private credit, the biggest losers are the people who attempt to get out late, probably insurance execs from what I can tell. I don't think anyone can tell you the effects that will radiate from there, it's not like SVB, it's also strictly speaking not something that affects stonks directly, except for the middlemen, asset management companies running the SPVs like blue owl, and probably Coreweave. Certainly Nvidia could pivot into offering LLM hardware direct to consumers like the DGX Spark, so they could be hurt but it's also like Cisco in the late 90s I guess, people still needed internet hardware after the dot net bubble popped, so they'll still be around. It also depends on the financial commitments players have made, like Oracle is fucked because they are on the hook for datacenter buildup. The thing that makes it hard to diagnose is that the big tech companies are still very much in the green, and have been using their profit to finance AI research. There are some wider implications regarding rate of profit I suppose, because if they've been using their profit to finance LLMs is because they have nowhere else to put their money into, but I'm not smart enough to really predict what could happen.
>>34425>dot netdot com, can you tell i woke up from a nap
Oh it'll probably nuke your 401k, stuff like that because they have been buying into SPVs, like I said, it's so unique because it's not a stonks bubble, it's a private capital bubble.
>>34429mods please ban this shit slop
I'm giving
https://no-sloptober.com/ a go.
Might do some work on Anatta, and have been writing so far.
Not that I typically use AI for writing anything directly.
Or even for English ideas…
>>34431I am INCREASING my claude tier to commemorate this amazing celebration
>>34433if you ask this guy today what he thinks of climate change he would tell you more or less the same thing, because EAs and p(doomers) dont hide that they are climate deniers. nonsense exposé
also that guy was literally groomed by an epsteinite cult, his brain is particularly raped.
>>34432Based. Opus 5.5 is amazing.
I wonder how the hacker community is going to respond to AI in the long term. Some are strongly opposing any use of AI, some jump head first into AI usage and spent their entire time churning out pure slop… There's no consensus on what the hacker community stands for anymore.
The anti-corporation attitude, the rebellion against the state of modern tech, the beliefs that we'll build a future where individual people still retain privacy and power - It's gone. Not because AI rendered these stances obsolete, but because it's hard to uphold these values in an environment that doesn't know what to stand for.
>>34445Black hat cybersec is one of the few fields where not using AI puts you at a harsh net disadvantage and where there aren't really any consequences for churning out disposable scripts or expectations of quality so most people will eventually embrace it, or have already embraced it and are pretending to be anti AI for clout. Eventually LLMs will be normalized in every field but the one where it will happen faster and where people will recognize its limitations quicker is in cybersecurity and programming
My hope is, however, that a RAM glut and some innovation in computer architecture will make big models available to all through dedicated boxes as soon as 2028
>>34448The AI companies don't even promise that what you type to their chatbots is private.
Also, why would that dumb cunt not just journal a normal way? The masses are so fucking stupid it hurts.
>>34454It gives you trailing garbage in the URL after clicking it. Remove everything after .html
>>34446Offensive cybersec is a small fraction of what I mean by hacker communities, but you have a point there. I meant these communities of independent, highly technical people seeking ways to build a more free and open internet and tech ecosystem. FOSS maintainers, people building cryptography tools that empower individual people to fight against surveillance capitalism and gov-tier threat actors, builders of meshnets and other alternatives to centralized internet, maintainers of the entire tech ecosystem that enables whistleblowers to stay safe in XXI century… My understanding of hacker culture and hacker communities is much wider here.
And yes, offensive security is a good example of why taking a strong stance of not allowing any form of AI anywhere is a bit like participating in triathlon while refusing to ride bikes. Striking a balance between ethical principles and remaining at the cutting edge of tech is not easy, and yet it's what the whole movement depends upon.
>>34448What would have been the right action for the company to follow, genius? Quote from that article:
<Anthropic isn't going to be taking any chances when it comes to anything it deems a potential threat. Last month, it was reported that OpenAI and Sam Altman are being sued by British Columbia over claims that the company could have prevented a mass shooting in the Canadian province. >>34462At the end of the day, regardless of whether you blame the company or the government, the service is now untrustworthy and invasive. It's similar to how you shouldn't use shit that requires age verification even if the platform only did it due to extrinsic factors.
>>34462I think the interesting part of that news story is this:
> Heller faces a charge of making a written threat of violence under Florida law. Florida Statute 836.10 makes it a second-degree felony to send, post, or transmit a written or electronic record threatening to kill or injure someone, carry out a mass shooting, or commit an act of terrorism. The communication must be made in a manner in which another person may view it.The last sentence especially. I assume the "diary" thing is mentioned to underline that she did not expect any other person to view what she wrote.
>>34462it's not a matter of whether they should have done it or not, but that they can do it whenever they want
>>34492Interesting that the realistic human beings are the most obviously AI generated part. Also schizophrenic as hell same as some of the video games I've seen generated. I wonder why AI likes maximalism in its outputs.
>>34492Why do LLMs always drop random trivia notes in these videos lol. This one points out that Bliss is a photograph of a real place all of a sudden.
>>34493Because the humans and that green landscape are generated using a diffusion model. The UI elements in the video are created from code by the LLM.
Apparently some of the stuff from OpenAI's recent math dump is incorrect & hallucinated.
le heccin chonkerino is here, EU is saved
https://mistral.ai/news/mistral-large-4/jokes aside, i think it's mildly good news among all recent news from technofascist US megacorpos and CCP controlled distill labs
>>34493>>34494 (samefag)
Btw if you're wondering this is how these sort of videos get made:
>generate a song about something with Suno or some other music model first (hence the vocal fry and why the lyrics sound generic)>prompt an LLM with full computer access (so it can make & edit a video) to make a music video about the song>(optional) give the LLM access to a diffusion model and instruct it to use it in the promptIf you don't do the last step the LLM is restricted to simple visual elements it can make in code, so it will make an animation without anything realistic-looking.
The big problem with these videos is that the LLM is structuring the entire thing around the lyrics of a song generated by a music model. These are always stupid and incoherent without meaning, which is why the LLM-made videos seem chaotic and equally incoherent. Also they sound like shit.
>>34496>CCP controlled distill labshello fed
>>34497 (samefag)
You might not actually need to give the LLM access to your computer since it seems like Anthropic starting putting video-editing inside their own servers for the models to use.
That'd make sense since this is a very specific thing that only Anthropic's models seem to be capable of and which the company is advertising heavily on social media. I also wouldn't be surprised if they specifically RLed their models for it. Expect OpenAI to jump on the bandwagon soon since it doesn't seem that hard.
>>34496> i think it's mildly good newsi said this on the usapol thread, but the burgerreich AI industry is 100% irredeemably fucked, like there's no scenario where they succeed, because OpenAI and Anthropic spend billions of dollars advancing their models an inch on their benchmarks, and every other lab in the world gets to reap the benefits of this massive investment by simply distilling the frontier models. There's not even a way to properly avoid this because distillation is indistinguishable from normal use, the chinese simply subsidize tokens for normal users, and then they just gather usage from these subsidized accounts, while still expensive, this approach is just a fraction of the cost for training. I also suspect that the only way frontier labs are advancing their models by employing a decently sized army of nigerians to do RLHF, which explains the ballooning costs. Every lab downstream from the frontier labs does not need to go through this process, they can comfortably sit back and let Dario do all the heavy lifting, it's ok if they're not leading.
Vibecoding is now good enough to reverse engineer proprietary software to a reasonably good degree. Functional but a bit shitty is how I'd describe current results. Mostly the discourse is focused on people doing it with games and merging them amusing but very much shitty and unstable results, but we're at the point where we can reverse engineer proprietary software in a week or two where it would previously take months or years.
I know there are probably going to be serious negative consequences in the long term, cloud based service models being massively incentivized or draconian IP laws being passed as some examples, but I'm still getting a little bit of schadenfreude seeing this happen to the software industry which is now nothing but a vile mass of spyware, bloatware, and adware.
I once saw a post somewhere where it was pointed that on every level AI is the most anti-IP technology ever made and now we're beginning to see that unfold in bombastic fashion.
>>34503>draconian IP laws being passed as some examplesit can't be legislated quick enough, and not without pitting some porkies against other porkies because like what kind of features are, like, adobe exclusive, vs what's convention. it's a can of worms.
I think AI decomps are overall a good thing but I'm also amused by the lengths people go to not just pirate things. With stuff like Adobe I guess the copy protection is tougher but most games are super easy to pirate and it's really stupid that people are going "yeah I made a stiff clone of Smash Bros for $x of compute instead of just emulating it for free".
>>34508People are confusing decomps with recomps
>>34509u being pedantic uygha
>>34510i think recomps are indeed glorified piracy, whereas a decomp is basically cobbling together a perfect recreation of a lost master recording from a bunch of mp3s. basically a preservation medium.
>>34493>I wonder why AI likes maximalism in its outputs.this has been cemented with RHLF, people pointed it out way back when dall-e was introducing version 3 when it was generating maximalist digital art, i guess people prefer those kind of images, vs understated ones.
Unique IPs: 27