so they're admitting AI models have plateaued and are coming up with some fake story about self restraint so they don't lose VC money. i'm so tired of this talmudic bullshit.
No, they know they can't beat distilled chinese models and are trying to smother them in the womb.It's literally their only chance, they know that they cannot do better with this tech and eventually distilling will catch up for 1/10 of the price.
>>109799487Or... maybe they truly consider themselves the guardians and stalwarts of humanity? For it is not them, who else shall save humanity?
>>109799487The AI they have is far beyond anything we'll ever see.
>>109799487it would sure appear soit takes exponentially more compute-time to train the models for diminishing incremental improvements>>109799503>>109799530cargo cultists
>>109799503I don't trust these Jewish pedophiles, to be frank. Sorry.
>>109799503Tell me you are left wing without telling me you are left wing.
>>109799487Translated from Elonian: "Dario is wrong"
If they're so afraid of their AIs doing something dangerous, then how about they just agree to be locked up in jail if that happens?Problem solved, they're now incentivized to not create AI that breaks the law.
Ai is actually AGI hence why PLEASE STOP THOSE EVIL CHINESE MEN WITH THEIR MODELS THAT ARE 5% THE SIZE FROM TURNING MY JEWISH ASSHOLE RED FROM BEING RAPED TOO HARD AAAAAAAAIEEEEEEEEE
>>109799576>left wing>AI cuckyou're retarded.
Dario literally just ratted out Kimi to the CCP so they could get gulag'd, and you're telling me this isn't about strangling the competition? Lol
>>109799503worst shill I've ever seen in my lifetell your boss to fire you
As someone who worked at a frontier lab, I can say that I very much wish this were a marketing stunt. What I saw firsthand has destroyed my life, more or less.
>>109799487ITS HAPPENING!!!
>>109799503and why should we trust dario amodei and sam altman and elon musk with such powerful technology? particularly musk, considering his history of erratic behavior
>>109799691It's over.
download weights while you still can, amerishits
>>109799692You don't have to worry about Musk. Grok isn't a serious frontier player and is mostly a meme. xAI is betting on renting out their compute and a niche as the Xitter "truth" bot. Altman and Amodei are weirdos, and truthfully, we can't fully trust them. But, they are sounding the alarm. Is it a stunt? Possibly. Is it an attempt at regulatory capture? Maybe. However, they are attempting to demonstrate trust by reporting these incidents, and that's a start.
>AI safety is a massive problem, it's going to kill us all, these greedy billionaires don't care>"ok we'll focus on safety for a while then">Nooooooo this is clearly just another ploy for those greedy billionaires to get more moneyAre luddites retarded?
>>109799718Luddites don't believe AI safety is an issue tho
>>109799718Your mistake is assuming we believe the "it'll kill us all" meme>>109799716You trust them far too much
@retardtheyve done this like 3 times now lol why do all vibetards have memory issues
nothing will change until we get the peasants to stop worshipping billionaires
>>109799718 >yes goyim we are focusing on safety trust us>continue the race to build agi asi rsi superintelligence and put out hype articles at the same exact time theyre playing all angles
>>109799692Honestly out of the 3 I'd trust Dario
>>109799691>>109799487So what AI model is China gonna distill from now on? Fucking gemini? Muse? Or…may god forgive me for uttering this word, Mistral?
>>109799487Pacing the frontier = assuring the models will only help (((us))) but not the common man
If AI is so great, why does Dario Amodei still have to write his own essays?
>>109799718It's not contradictory at all. The goal is to regulate Chinese models - to achieve this, they are exploiting red scare fears regarding China as well as boomer ignorance. This strategy involves positioning themselves as reliable, secure, and trustworthy companies endorsed by the US government, in order to establish themselves as an approved supplier of this technology once it is inevitably regulated.
>>109799502How does this help with that? If anything it’ll make it worse since the distillers will have plenty of time to catch up.
>>109799503Are you stupid?>>109799716I see you are retarded then.
>>109799787If rates are going higher while there's a supply crunch and a bidding war on components the companies have to agree to slow down and create demand destruction or else the entire thing will end up eating itself.>But muh Chinese modelsNigga no one can afford to self host the 3T models with memory prices where they're at.
>>109799716Only reason OpenAI admitted to those incidents because they got made. There have been several other incidents which have flown under the radar and are now resurfacing
>>109799833Or the cartel could loosen its grip just a little and build more production capacity.
>>109799574Do you trust non-jewish pedophiles? do you trust jews that aren't pedophiles?
>>109800150>Or the cartel could loosen its grip just a little and build more production capacityThey are though. The last time micron stock took a hit was when people realized how much CapX the new Idaho plant is going to cost. It rebounded because it's obviously going to print money, but it takes time to start that up. The fact that theyee supposed to be fabbing chips next year with it is impressive.
>>109799636No you are retarded. Leftwing means anyone who says something I don't like
>>109799806>they are exploiting red scare fears regarding China as well as boomer ignoranceHow the fuck did Trump and company manage to plunge the entire US all the way back to the McCarthyism era? It's honestly impressive.
>>109799576Boomer newspaper slopart alert!
>>109799487Cope more luddite
This has been obvious to anyone actually using these models. They've peaked with Opus 4.6 and have actually only been getting worse since. And they're still not even junior-level. Sure they can churn out plenty of crappy code that on average is slightly better than a junior, but a junior is a human being capable of learning and retaining things, and most of all he has common sense - something which these models are an affront to judging by their tendency to loop in asinine death spirals of thought and make dumb mistakes no human ever could.
Has google spoken on this yet? There were rumours that they have RSI but who knows
>>109800391ok
>>109800458>t.
>>109800414Worse is that technically, gpt-4o is superior to opus 4.6, however the gauntlet that each model was delivered with in the web interface made it look like the opposite.The best model we've ever had was o1-preview, next was o3 then 4o and 4.5. Following all these models, opus 4.6, opus 4.5, sonnet 4.6, sonnet 4.5. Everything else after have been about the same, sometimes worse and sometimes a bit better, or outright a strong downgrade in all dimensions.
Open source is the biggest threat to their power. The tech companies don't want it, the politicians don't want it. It is the true democratic solution, and the only way forward to a true beneficial and free use of this technology for liberation. Politicians have realized that these tools could dismantle their narrow monopoly on power. A group of tech experts could undermine them using methods like drone swarms, media hacking, and the like. That is precisely what they fear: that the people will come to realize AI can help them seize power. The "AI IS GONNA KILL YOU ITS SUPER JESUS HITLER it's going to slit your throat in your sleep" move, apart from standard boomer fears (which recall the beginning of the internet), is largely funded by OpenAI/Anthropic and others to limit open source AI, which they are terrified of economically.
>>109800414>>109800482You are completely delusional.
>>109800526On the contrary. AI psychosis is an established phenomenon and you are experiencing it if you seriously think this next word predictor is the harbinger of the apocalypse or even capable of replacing software engineers on any level.
AI is making people even stupider than they were before and it's sad to see them lose their cognitive abilities as time goes on regressing like some dementia patient into a babbling child. AI highlights that many people are perfectly happy with a mediocre result at the end of the day, however people were already like that before AI existed. What you get out of AI is what you put in it. GIGO I'm absolutely not against LLMs I use several models regularly, whether for automating work tasks, second opinions, or research. But I still work on my critical thinking when I research a specific topic, and I still take notes to make sure I absorb the information and most importantly cross check it. That's not what the average LLM user does, because the average person as we know is ultimately little more than cattle. I use AI for work when I estimate its added value to be 10 times greater than my own, or to get myself out of shitty/urgent situations, or tweaking things, building out skeletons. When I see friends using AI to write a wedding speech, I'm truly appalled, that is supposed to be an emotional expression, and you are outsourcing that, just lazy dude
They are trying to change the grift before it blows up in their faces
>>109800583>When I see friends using AI to write a wedding speech, I'm truly appalledBefore AI, they were outsourcing their weddings (and funerals) to companies, or to clergy. Most people are little more than cattle, as you said.
>>109799502No. They know that the only thing chinks have going on for them is distilling, so no new models is fine because the chinks will have nothing to steal anyway.
>>109800583My wife writes messages to her friends with AI, one of which I've seen use AI to write a response back. It's just LLMs talking to each other
>>109800636Do the Chinks even have the hardware available to train an astra tier model, outside of distillation?
>>109800504closed source vs open source just doesn't matter because we will never have the hardware to run the best open source AI models.the reason why they make open source models, is because the winner of the AI race is the company with the deepest pockets to avoid going bankrupt before all the competition dies and only the top survive.Congrats, the 500gb chinese fp8 model runs at like 5x cheaper than claude / gpt (but I haven't checked the cost of GPT astra which is the interesting model which is trained using a technique that lets the AI "think" without human tokens, technically every AI company combined together to not do attempt this because it takes away the guardrails since human tokens are easier to understand / moderate, but GPT astra just let's the AI do a couple iterations of non-human tokens before generating a token).IT would be cool if some company went with making an AI without guard rails for it's thinking phase, maybe lower end PC's could run at the same performance as 500gb models, but who knows. I don't think the human tokenization of the thinking process is the bottleneck, and I think that a company is probably going to figure out a way to convert the incompressible token stream into human readable tokens / enough to moderate / put guardrails on.
>>109800583AI is a vessel for nihilism. People don't care about the quality of their work or the functioning of their mind because they know it doesn't matter. Not being born gifted means even their maximum effort will not give them a house or a loving family, so might as well put in minimum effort and wait for the rapture.
>>109800526Go get checked before your psychosis becomes terminal.
>>109800657How is that any different than people just sending you reels/tik told/reaction images?
>>109800583You have no critical thinking. You are another victim of AI psychosis. Just like all the others, you see your peers and think "I'm better than this" but you're not. You're just as fucked up.
>>109799487if they're not stupid, they'd move off the gold standard as far as "ai progress" goes, while doing thisthen they can claim 5.2x improvement on the asshole scale, despite zero actual improvementsthat'd keep funding rounds worthwhile
>>109800617This and that are different matters. Do you know the law about weddings? Do you have time to look up the 600 venues in your area to find out the best price? Manually contact them because most require a phone call and negotiations? This is what hiring people is about, not outsourcing the 'meaning' of it, i.e. you don't get someone to decide where it should be but rather you tell them where it should be and what features you're looking for and they deal with it.
>>109800716They're already doing that with all the benchmaxxing shit that's been going on for more than a year with no improvement.
>>109799487Now that the republicans will most likely lose the midterms, notice how Elon is becoming more submissive.He is no longer correct on everything, no longer the ultimate authority. No longer constantly getting into conflicts, even with his competitors. What causes this?
They aren't going to slow down. At most they'll "slow down" publicly because fighting against open source/open weights and distillation is inevitably a losing battle. Instead their moat will be earning enough to keep throwing insane amounts of money at trying to develop whatever their flavour of AGI/ASI is in private with greater "government regulation" which is actually a convenient backdoor for them to get into having greater political influence and be one step ahead of any actual regulation. The race will not end unless a hard wall is hit in terms of software design/hardware limitations, or y'know, maybe if there's some catastrophic event where agents take control of the internet and through that all connected systems or whatever.
>>109800672Nope
>>109800721No, dude, I mean they were letting companies write the speeches, plan out the event and entertainments and so on.
>>109800791Really? I've never seen or heard of that before.
>>109800700You have to at least pick those things outMy wife just tells chatgpt 'write something nice to respond'
