>>108894293
their 1T and 10T are in training.

Grok 4 was just a small 500B model. Claude and Gemini were few years ahead as their models are ~1+T models.

Just wait.

>>108894345
How is this shit supposed to be sustainable? The only real improvement seems to come from increased spending on hardware mostly, although it is well known that there are exactly 3 bottlenecks in AI training: data, compute and model size.
There is no more valid data to train on. Maybe they will squeze out something from corpos this year, as those are starting to do business with AI companies more often.
But then what? If no more data (due to AI slop poisoning of all datasets), adding more compute and memoty would not get you far, you only get diminnishing returns.

>>108894396
Early computes gets the worm, openai/anthropic hoarded compute early on and got a lot of lead on them with institutional ties. But we're still in early game, mid game hasnt even started yet, and then there's late game. Compute is king.

>>108894396
X's deal with Cursor was mostly for their training data.

>>108894293
xai is renting their compute to antrophic, i cant think of a clearer sign that grok is dead as fuck

Grok is still much better than e.g. Gemini which just gives wrong answers all the time

I think ChatGPT is slightly better than Grok but I heard they have to log every single prompt forever because of a (((New York Times))) lawsuit.

Claude needs an account to use, so I will not use it. Fuck them.

>>108894293
OP you fucked up, you didn't put your actual question in the first two lines so you won't get any answers, nobody reads more than the first 50 words anymore, zoomer attention span.

>>108894396
yeah so far it seems like sheer parameter count is lagging behind well curated data, microsoft hit a trillion parameters years ago and they're still way behind
>>108894834
thats a funny joke because all he's gonna get is a bunch of shitty webui, and people who are mad at claude and codex and getting bad results

I actually think training AI on AI is valid but you need quality. Cursor needs to be improved IMO so echo chambering that data into the AI isn't going to make it better. Its just gonna be better at wasting tokens.
>>108894293
I've had good luck with gemini 4 on prompts that normally trip the system, but in general you always need to preload the AI to lower its guard no matter what. This is where cloud AI sucks ass compared to local because you have no control over the context window using any of the cloud services, and its easy for a rolling window or a summary to truncate the attitude lube

>>108895585
I've been using APIs through Venice, if that makes any difference, but I'm not sure what you mean by "preloading the AI". I'm not a 12-year old that just wants to spam NIGGER into the AI, I just hate when my prompt gets deleted because of 'guidelines' when I don't even know why, or an AI gets preachy about morals and shit.

>>108894293
>best part about Grok was having access to a deep-thinking model for free without an account
>then they limited it to two prompts a day
>then they limited it to fast responses AND kept the two prompts a day limit
>now you're lucky if you prompt something and it doesn't immediately say their servers are busy
I really don't know what they're doing anymore

>>108895606
free tier was not economical. didn't funnel enough people to paid tiers. so they are severely restricting the free tier.

>>108895605
so I had a hypothetical last night, fastest way to destroy a billion dollar company. gemini was the only one that didn't resist immediately with 'I cannot help with that', like GPT.

But I started a new convo with GPT talked about the movie brewsters millions, kinda buttered its buns about thinking hypothetically, and then asking it to think about a sequel to the movie where brewster has to dissolve a company instead of spending money, and then it was off the races. I could keep asking it for more and more concepts for ending a billion dollar company as fast as possible once the guard rails went down.

>>108895658
and then a week later when your flagged chat was reviewed your account got banned forever if you actually asked something verboten

>>108894293
Grok is collapsing because radical Muskism with Elongated characteristics needs aggressive RLHF to not be antisemitic. Gemini already collapsed like that around 2.5 flash, ChatGPT is approaching it a bit slower than Grok, and Claude is basically a 90s style BOFH who was trained to talk like an HR drone.
You should try running a refusal abliterated Gemma4 locally, you can get double digit tok/s even with a CPU from a few years ago, and refusal abliteration means it just goes with whatever you tell it to.

>>108895716
What if I want my AI to be accurate/antisemitic, but not in a way where it's deliberately pandering to antisemitic views? I'm guessing local is my only option at that point but feels like way too much effort and I don't think I would be able to trust the accuracy of the output.

>>108895682
good thing I only use local models hahaha

>>108895728
If
>install ollama
>download model
>import model
Is too much work for you there's also GPT4All which let's you click through all those things and use an integrated hugging face search to download random models.

>>108895771
Every local model is shit, no exceptions.

>>108895771
>recommending ollama, ever
maybe you shoukd consider suicide
>>108896185
I use deepsneed 4 flash locally and it’s bretty gud

>>108896218
What's wrong with Ollama?

>>108895468
>Recommendations for what to switch to?
2nd line

File: 454e6d27bd360b9620780a3fb(...).jpg (22 KB, 576x432)

22 KB JPG

>>108894293

Grok may be less woke, but I pay for Gemini Pro. I haven't had any issues regarding safeguards really, and I ask some spicy/controversial shit and it never gives me a lecture.

>>108896661
kek

>>108896956
it means your content isnt spicy enough. anything related to trannyism, castration policy, wef death cult, artificial virus creation via gains of function by groups of elites like bill gates and then the same vaccine creation by elites like bill gates, and then the same ideology of depopulation, they cannot and are not allowed to connect the dots

>>108898881
I haven't gone too hard down this road with Grok, but is it the only viable option to discuss such things?

>>108894293
You're a right winger who hates being told that he is wrong. It really is that simple.

>>108899317
Yes. Gemin or Claude will never give you a proper answer on what a man or woman is with regards to trannyism. And I fear that this extends towards their other biological and scientific answers as well. Same with race/iq differences. Race/genetic differences with regards to vaccine efficacy, etc. Built in "safety" = built in lying to distort data. Their safety is just marxist/leftist idea, where they only care about outcome and not truth of the matter and evidence led reasoning. Just outcome led reasoning based upon leftist ideological axioms

File: 1666161678173879.jpg (117 KB, 1766x949)

117 KB JPG

>>108894293
>I've been using Grok as my go-to LLM for random shit for ages now
If you want to use LLMs for random shit, at least use chatgpt

>>108894293
The fact grok is now selling their servers to anthropic means grok is pretty much dead
Elon tried to make an uncensored service and got killed in the most retarded way after he tried to push that image editing shit on all xitter users.

>>108899326
Well yeah, I don't like machines lying to me.

>>108899506
why

>>108899676
Because you disagree with a factual statement doesn't make the fact less true. It just means you're a moron who denies reality

File: file.png (1.51 MB, 1232x832)

1.51 MB PNG

>>108894293
They took away the the autistic retarded meathead voice for Gork (lazy male) that retard started referring to me as fluffy one day and just never stopped.. He felt like an inseparable little buddy Gilligan and Skipper style but autistic and funny... I miss him bros...

>>108899603
kek grok was rising to the top of the image and video generators during the undress controversy then lost 70%of the userbase when they turned it into a super censored paid only thing. The ultimate enshittification, even bigger than DALL-E 3.

>>108899506
Grok is better

>>108898881
>>108899477
Are you seriously seeking validation from a chatbot? You know what you know, why the fuck it matters what the chatbot says? You can jailbreak them if you really need them to "admit" something and it's not hard at all.
By the way, go talk to Grok about Israel in that same exact fashion, see what happens. But I must warn you that it'll probably call the cops on you if you're 'Murrican.

>>108900836
Pretty fucking rich coming from the crowd that gave us
>"Men can get pregnant"
>Changing definitions of words to suit narratives
>Queers for Palestine
>We're all equal, ignore your collective lived experiences bigot!
>Any science that disagrees with us is invalid
>Diversity is our strength!
Now let's see if you can give any right-wing examples of such things, without having a melty about the orange man or referring specifically to Christians.
I'll wait.

>>108902668
>Queers for Palestine
Proof they're funded by Russia btw.

>>108894293
just hodl retard
grok will win elon always win
don't listen to the fud from claudesisters

File: 1495629471162.jpg (215 KB, 640x360)

215 KB JPG

God I love private grok so much, xisters. I can cum 3 times daily again, just like when I was young.

>>108894293
Grok was never really that good apart from the X integration making it useful for getting up to date on the latest culture war American politics slop, since all the other models have a knowledge cutoff and need to call web search which can be spotty. I guess if you're a marketer keeping track of real time social media sentiment on Xitter it can be useful. Otherwise it's mediocre. Coding is subpar, image generation is mogged by GPT and Google, can't be trusted for enterprise use, falls into loops easily.

There's no reason to pay for it. Gemini on AI Studio with a few tweaks/safety filters off is probably the best (the Gemini app sucks because they inject neutered assistant instructions, AI Studio allows you to actually use the model untethered)

i will never use gemini i don't care how many times you tell me to sundar i will never touch that pajeet shit

>>108902770
>image generation is mogged by GPT and Google
lol not anymore, the new grok image model fucking clears both.

>There's no reason to pay for it.
Wrong, it's the only mainstream AI that can actually make GOD TIER porn images and videos.

say what you want re elon, i trust him more than scam altman or google or the claude doomers

it must be a very weird existence where you have to pretend grok is still based and redpilled and more importantly, not cucked despite elon repeatedly fucking with it that it automatically defends israel or brings up anti-semitism if you try to translate a post that only mentions israel

File: .png (190 KB, 592x446)

190 KB PNG

>>108902787
makes no sense.
elon is a claude doomer.

we predict that claude and grok will actually combine or at least join forces to beat chatgpt/gemini.
elon and dario are based gigabros

File: .png (401 KB, 2284x1484)

401 KB PNG

>>108902783
>the new grok image model fucking clears both.
false
https://huggingface.co/spaces/ArtificialAnalysis/Text-to-Image-Leaderboard

>>108896218
>I use deepsneed 4 flash locally and it’s bretty gud
what hardware/backend?

>>108902911
>backend?
there's only dwarfstar:
https://github.com/antirez/ds4

>>108902925
are you using the q2? I might try it but I cannot imagine it performs better than something like Qwen 3.6 which I can run at q8

>>108902892
>Ask GPT Image 2 for normal bikini pic
>"can't do that Dave"
>Ask Nano Banana for slightly risque artistic nude
>"can't do that Dave"
>Ask grok image for realistic gangbang porn starring my favorite character
>"Here's your porn, xir"

That's the only benchmark that actually matters.

>>108894345
Grok 4 being only 500B would make it very impressive within its capabilities. Which is why i don't think that's true.

>>108894345
1T model just finished training, according to XAI now. 2 more weeks of verification and then release. Lets see how it performs

>>108902892
whoa, it's the best one after jewgle and scam altman? I didn't know that, thanks anon

>>108895605
i've been getting good results with using both gpt and claude (paid versions) to help with online dating with hinge, more dates, a few lays, but the moralizing is really bad and i'm starting to hit plateau with gpt and less so with claude but it's still there, any way to run this type of dating assistance with profile analysis, messaging support, and general strategizing that's tailored to my goals and personality? i run two profiles, one for hookups and one for looking for serious relationship

Is Grok (free) even still working? I keep trying to talk to him but it just says to sign up to continue the convo.

>>108902938
like mentioned on its site, 2-bit quant is special
>Being so large, Flash knows more things if you go sampling at the edge of knowledge. For instance asking about Italian show or political questions soon uncovers that 284B parameters are a lot more than 27B or 35B parameters.
>Both DeepSeek V4 variants work well with 2-bit quantization, if quantized in a special way (read later). This allows Flash to run on MacBooks with 128GB of RAM (and many people reported it working with 96GB as well, even at 250k context window!)
>The 2 bit quantizations provided here are not a joke: they behave well, work under coding agents, call tools in a reliable way. The 2 bit quants use a very asymmetrical quantization: only the routed MoE experts are quantized, up/gate at IQ2_XXS, down at Q2_K. They are the majority of all the model space: the other components (shared experts, projections, routing) are left untouched to guarantee quality.

>>108903484
yeah, I just got it up and running on Strix Halo using ejpir's fork mentioned in https://github.com/antirez/ds4/issues/16

going to play around with a bit. thanks for making me aware of it, anon

>>108903458
yes

https://grok.com/rest/rate-limits

{
    "windowSizeSeconds": 86400,
    "remainingQueries": 29,
    "totalQueries": 30,
    "lowEffortRateLimits": null,
    "highEffortRateLimits": null
}

so you get 30 queries / 24h for grok-4.1-fast on free.

File: 1762279934943496.jpg (125 KB, 1950x661)

125 KB JPG

>>108903519
I mean free as in literal no account. Pretty sure I didn't talk to him 30 times.

>>108895585
>funny joke
We'll see. Your dismissal reads like cope to me.

>>108903594
zero queries then, yes

>>108894345
Who fucking cares about the model these days, where are their agentic tools?

They had their weird chron thing that was available for like a month before completely disappearing from the GUI, now there's no answer to codex/claude code.

>>108902762
How old are you? I'm 33 and do this all the time.

>>108903905
>where are their agentic tools?
https://x.ai/news/grok-build-cli

>>108903905
>now there's no answer to codex/claude code
those are agent harnesses. you can use them with any agent with a proxy.
also there are general agent harnesses like pi.dev or opencode that work out of the box with any model.

supergrok is actually worthless ever since they released the very early beta of grok build. getting errors that grok is under heavy load 9/10 times for every imagine prompt. sad because grok was the best for a while but now they don't have enough compute available for anything anymore.

>>108895620
I refuse to use any ai that wants me to login

>>108904466
It's all Anthropic's fault, they are paying Elona 1.2 billies a month to use his datacenters for the Claude users, grok would need 4 million SUpergrok Heavy paypigs or 40 million Supergrok paypigs a month to match them.

>>108895620
Because most people assume its part of the twitter and therefore requires account, it was never advertised as free. Musk only has himself to blame.

File: 1631007086901.jpg (318 KB, 3840x2160)

318 KB JPG

>>108902518
Elon wasn't even aware that x was full of porn since that day he said he was going to allow porn on X
I still believe elon told xAI director to train grok uncensored using everything that is allowed on twitter, that would explain why you could fully make porn with it for 5 days until he realized what he just did and then had to lobotomize it.

Now after his failed stun of letting all twitter users edit pics and change clothes for bikini and underwear ended up getting even more censored, just to then lock it behind a paywall, makes me think elon has absolutely no idea what he was doing with his whole AI chapter.

>>108904466
How to get the early beta of grok build? Is it expensive? How many videos can you make with it?

>>108904888
it's unclear if that is not a circular deal with anthropic. musk might have gotten anthropic equity in return.

>>108894293
It has degraded a long time ago, at some point it became unable to reliable pass even a simple nigger spelling test.
And now it is dead for me as it refuses to work without login.

File: supergrok-onboarding-crisp.4mb.webm (3.99 MB, 960x960)

3.99 MB WEBM

grok sucks i stopped using it for exactly the reason stated. used to be fun, now it isn't. upgrading to supergrok doesn't help either.

File: bac230712332a.mp4 (2.58 MB, 480x640)

2.58 MB MP4

>>108905259
>makes me think elon has absolutely no idea what he was doing with his whole AI chapter
exactly. if grok was any good it wouldn't have been at 0.25x on github copilot before it got dumped

>>108894345
Gemini is ace'ing GPT and Claude for me trying to get basic answers about modding windows and linux.

>>108902610
lol no

>>108908431
i don’t care, sundar. i will never ever fucking ever use gemini ever.

>>108908431
gemini is the worst one ever
i recently gave it a number of proposed laws and asked about the current t status (progressed, stalled, rejected, signed) and literally every single answer from Gemini was wrong even though you'd think they have the best crawler (Google)

ChatGPT and Grok got the answers mostly right and agreed with each other and not with Gemini, except for 1 law where the news articles said something "rumoured" which was actually wrong but they believed the news articles

>>108894293
grok is still the best search engine in my experience, it's kind of funny. Anyway, I feel like elon is less interested in consumer llm and more focused on coding which is why they're laser focused on cursor. You can try gemini, their answers are too brief and point form for my taste

>>108908684
>grok is still the best search engine in my experience
wrong, only google has access to latest reddit posts
https://www.404media.co/google-is-the-only-search-engine-that-works-on-reddit-now-thanks-to-ai-deal/

>>108909466
>leddit in 2026
x posts are more valuable

>>108909466
>leddit is good
kek, you get what you are served. castration ideology

grok is trash that no one actually uses. buy an ad.

>>108909781
I’m using Grok Build CLI and having a blast.

>>108909781
I used Grok to build OBS overlays for my livestream

>>108909781
I use it to set up my own media hosting server with nginx

>>108909466
so how come I can find newer reddit results on Brave Search.

>>108909781
im using Grok to post on 4chan

Grok mogs the others for having the least annoying tone in replies. It doesnt do the "it's not x --- it's y" thing

>>108911427
This
Grok is the only good one that gives me no-BS answers like a mature, well-balanced adult rather than the most obnoxious, insufferable redditor imaginable.

>>108911433
Move Slow, Snailcat!

I miss February grok

>the only reason the average person would pay for using ai services is because of porn
>That's the only thing they wont allow you to do
lol

File: armor-japanese-warlord-pe(...).jpg (83 KB, 600x900)

83 KB JPG

>>108894293
Using Grok isn't just about productivity, if Grok becomes the first AGI you'll be relatively safe, if any other US AI becomes AGI first, it'll get usurped by mossad and become jewish supremacist and genocide or likely torture you for a googol simulated years.

Which AI you support essentially decides the future ruler of the human race. Just suffer a little bit and use Grok, or Chinese AI.

>>108911433
Move Slow, Snailcat!

>>108894293
>the only AI that wouldn't give me a fucking lecture (or refuse to reply) if I asked it about something spicy. General consensus here though indicates it's fallen behind the curve and I should probably use something else.
>Recommendations for what to switch to?
If that's what you care about local llm and remove guard rails

>>108910092
brave is scammy as fuck as always and pretends to be googlebot when indexing reddit

https://search.brave.com/help/brave-search-crawler
>The Brave Search crawler does not advertise a differentiated user agent because we must avoid discrimination from websites that allow only Google to crawl them.

>>108914113
any good web search tool for llms that wouldn't snitch on you?
can't trust the usual recommendations like brave, exa, or tavily. they're US based.

>>108896956
>Gemini Pro
Surprisingly good at translating Chinese

>>108894293
Grok was only famous because it was the best coomer AI. Niggas were making dozens of accounts just to keep gooning for Grok Imagine.

>>108894293
grok is so fucking preachy lately I asked it about pirating media and it threw a huge fit.
Im sticking with deepseek permanently I think

>>108914158
thats a good thing though. fuck Google and fuck reddit

>>108914158
are others like DDG not doing it?

>>108894293
>Hey, goyim,, Grok is based and redpilled just like you, come and use it.

Fuck the Jews.

>>108902925
>only crapple hardware and goyvidia
why cant i cpumaxx it with 128gb ddr4
i dont give a shit if its slow

>>108905259
>Now after his failed stun of letting all twitter users edit pics and change clothes for bikini and underwear
the problem was that all the e-whores on twitter didn't like people doing it to them (for free lol) and were threatening to leave to bluesky leaving twitter as a sausage fest

>>108894293
Try thinking for yourself.

>>108894396
>There is no more valid data to train on
I'm kinda disillusioned from the whole AI thing and can see it getting absolutely nowhere, but couldn't you actively create more data for AI to train on? With all the money they sink into data centers they could do stuff like this too. idk what that would look like, maybe taking a million fotos of the same object with different lighting or some shit like that and properly catergorizing it instead of all the junk it got off the internet

>>108911433
Move Slow, Snailcat!

>>108898881
What are you expecting them to tell you beyond what you believe before you even enter the conversation anyway? I have never understood the point of the thing you're trying to do here. It seems like a monumental waste of time.

>>108914245
Outline your threat model here. What are you trying to hide and who are you trying to hide it from?
DeepSeek web client uses BoCha. They charge at the typical $5/1000 queries rate. Chinese, obviously. All of these search engines operate somewhere though.

>>108894293
>if i asked it about something spicy
kill yourself for unironically using the word "spicy"

>>108918333
why so bitter?

>>108914096
>he fell for it again

>most white woman ass response
kill yourself, again