Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Sam Altman Declares 'Code Red' for ChatGPT, Delays OpenAI Advertising Plans
- Thread starter MacRumors
- Start date
- Sort by reaction score
Are you suggesting anything coming out against Grok is by haters and fueled by competitors? Everything is biased?
Not everyone has time to dive deep to verify every single detail. That's why we find journalists/sources we trust to do that for us, and we rely on their reporting everyday. I don't need to chase a tornado to verify its speed, or to dig a grave to verify someone's obituary.
In this particular case, though, I did in fact spend the time. The original awful statements were deleted, but here's the apology posted on Grok's official account about comments it had made 4 days earlier. There are many examples inappropriate comments by Grok, but this is the only time (to my knowledge) they gave a comprehensive apology.
I am not biased against xAI because of Elon. I am not a "hater." I gave it a fair shot. It just wasn't good enough for me.
I have. He is in control of Grok which has routinely devolved into a Nazi bot and he spends his entire energy in fascistic thought with the decline of white population.
But if that is not enough, then there is his bigoted stance on Trans folks, which the venn diagram for anti-trans views and fascist views is a circle. He supports far-right political parties. Twitter is a cess-pool of bigots (but using "cis" is a bannable offense).
Again, I see the sky is blue, so I'm saying the sky is blue.
Being rich isn't a skill. Developing useful technology requires actual skill, and usually whole teams of people with multiple disciplinary backgrounds, not just one guy brain-farting his way through it. That's why the moon landing (your choice of comparison) isn't credited to any one individual, much less the industrious Nazi you likened to Musk.
Many Nazis are. I mean, if you want to take surface-level ascriptions at face value then I can see why you'd find Grok and other chatbots appealing. But it does little for your political literacy.
I think it's outrageous that, this late in the piece, there are still people pretending not to know this and, even more bizarrely, pretending to be outraged by others simply noting it. No, sorry, the plausible deniability went out the window long ago.
And, again, Musk's Nazi behaviour notwithstanding, you are ignoring that the issue isn't just Musk. Other purveyors of this technocratic dystopia—Thiel, Andersen, et al.—deserve at least as much public disgrace. Musk just happens to be the most visible pustule, thanks to how he devours attention (bought a whole website for it!), but there are many more to lance.
I've been using several LLMs, and I would agree that OpenAI needs to call a Code Red. Their competition has caught up and even surpassed them in many areas.
ChatGPT is still really good for sales and marketing materials (probably the best out there), but Claude and Grok are both better at technical work and code generation (I've had more success with Claude for coding, but I've found that Grok is better than Claude for drafting documents). Gemini is great for videos and images, but mediocre at everything else (caveat that I have not spent much time with Gemini 3, but if the AI results under web searches are any indication, it hasn't improved much).
Unless there is some crazy breakthrough, we will end up with a segmented market where different LLMs excel at different tasks, and there is no clear winner.
This is serious for OpenAI, as I don't think their business plan accounts for them only owning 25% of the market.
To tie all of this back to Apple, I really hope that they let users pick and choose any LLM they want to use (similar to how they let you pick your search engine). I want to be able to decide whether Siri is powered by OpenAI (if they get this act together), or Grok, Claude, or Gemini, or whatever.
ChatGPT is still really good for sales and marketing materials (probably the best out there), but Claude and Grok are both better at technical work and code generation (I've had more success with Claude for coding, but I've found that Grok is better than Claude for drafting documents). Gemini is great for videos and images, but mediocre at everything else (caveat that I have not spent much time with Gemini 3, but if the AI results under web searches are any indication, it hasn't improved much).
Unless there is some crazy breakthrough, we will end up with a segmented market where different LLMs excel at different tasks, and there is no clear winner.
This is serious for OpenAI, as I don't think their business plan accounts for them only owning 25% of the market.
To tie all of this back to Apple, I really hope that they let users pick and choose any LLM they want to use (similar to how they let you pick your search engine). I want to be able to decide whether Siri is powered by OpenAI (if they get this act together), or Grok, Claude, or Gemini, or whatever.
To be fair you just told me you use Safari. I know because I have the same problem. If you use a Chromium browser, I recommend Brave, you won't have that problem. Safari cannot handle the long pages generated by AI's endless iterations of attempting to not suck.
---
Unfortunately there is no solution to the fact that for anything that isn't dead simple AI will tell you with supreme certainty an answer that is simply factually wrong. @casjohnr experience is dead on accurate. The great thing about programming is it is immune to opinion or heresy. With a computer a thing is either letter perfect or hopelessly wrong. And the computer could not possibly care less which it is.
It's funny really. The immutability of "letter perfect OR hopelessly wrong" reveals the calamitous garbage that generative AI swills out to the unsuspecting world.
This is not just a stock market bubble. It is a reality bubble that is going to make a lot of people look supremely foolish when the hype wears off and businesses finds out it is a tiny niche technology with not quite as much usefulness as your average Cocker Spaniel.
The problem is core tenants of generative AI construction is an evolutionary dead end. It cannot be "refined". It is squarely in the "that didn't work, throw it away and try something different" territory.
It is possible - likely even - that the Siri AI assistant no show is not at all an Apple Engineering problem but rather that AI just can't do what Altman et al promised Tim Cook it could.
Apple has had neural processors as part of their chips for decades, well before the industry adopted them as standard.
There's already talk that Apple is building a data centre using their own servers and chips to power Apple Intelligence.
Nvidia's big boost happened when GPUs transitioned from being primarily "Graphics" Processing Units into "General" Processing Units. With Apple incorporating an NPU into *each* GPU core, processing on Apple's chips should be very efficient.
Looking forward to Siri leapfrogging ahead of the industry as the best personal assistant. You can laugh now.
Then you are an exception, most tech people either hate Musk or really hate Musk. I am not a Tesla fan, but Musk has done some other great things. He has totally outclassed NASA.
ChatGPT 5.1 is still better than Gemini 3 for what I use it for. I once tried doing the same thing on both and Gemini's output was significantly worse, so I guess it just depends what you use it for.
Your numbers are out by nearly an order of magnitude. Losses for 2024 were $5 billion, making the monthly loss about $400 million. Still insane but not quite $3 billion/month.
That would honestly be an impressive feat.
Zig quits GitHub, gripes about Microsoft's AI obsession
: Zig prez complains about 'vibe-scheduling' after safe sleep bug goes unaddressed for eons
nixCraft 🐧 (@nixCraft@mastodon.social)
Lmao. Zig quits GitHub, says Microsoft's AI obsession has ruined the service. Microsoft simply doesn't know how to make profit out of AI because it is not profitable for anyone right now. So they are adding AI to Windows 11, GitHub, Office, and everything else they own, hoping that somewhere...
mastodon.social
Truth be told, most companies are flailing with AI right now, and we are treating them like the holy grail and acting like only Apple has lost the plot.
I’ll be sure to never steer any business your way. Thanks for letting me know.
Let’s call it machine learning. None of this is ai.
You didn't answer the question. You implied any negative press against anything Elon does is fabricated by haters. I disagreed, and provided one (glaring) example. Me being "an exception" (independently researching) doesn't change the fact that opposition against him is justified.
Elon has done some good things, though he is good at taking credit where it's not due. Acknowledging that he has some achievements doesn't stop me from despising him for a number of other reasons. Despite that, I still gave grok an honest try.
Your number was true in 2024, but this is 2025, and OpenAI has announced deals for compute power totaling $1.4 trillion of expenses over the next 8 years. If recent analyses (of, admittedly, a private company that is very opaque) are at all accurate, if anything I was off in the other direction--$3 billion/month is conservative.
Per this Register article, based on the fraction of OpenAI that Microsoft owns, and information they were required to report publicly, OpenAI was estimated to have suffered a net loss of more than $11.5 billion in the quarter ending September 30, 2025. That would equate to a loss of $3.8 billion per month for July, August, and September of this year. If the WSJ's estimate of how much of OpenAI MS owns is accurate, the loss was closer to $12 billion, which would be $4 billion/month.
That's indirectly calculated, but according to this Forbes piece based on internal documents from this summer, OpenAI itself was planning to lose $9 billion this year. That's much less than the apparent actual loss Microsoft implied for the quarter after those documents were made public, but as Forbes notes those estimates were before OpenAI signed those deals to spend absolutely staggering amounts of money on computing power, which could very easily account for the discrepancy.
If you believe OpenAI's own internal forecasts made this summer, they will lose more money each year until 2028, when losses will peak at $78 billion ($6.5 billion/month), then they will become profitable 2 years later in 2030. Cumulatively, I believe they expected to lose $115 billion between mid-2025 and the start of 2029, which overall averages out to $2.7 billion/month for the next three years or so.
Last edited:
Based on what? My personal anecdotes aside Grok wins out on many objective studies and analysis, heres an example:
xAI's Grok 4.1 vs. OpenAI's ChatGPT-5.1: Elon Musk's Grok is the clear winner
As artificial intelligence models evolve at breakneck speed, the battle for conversational supremacy intensifies. In this head-to-head…macdailynews.com
Even ones where it loses to ChatGPT it is close second with Llama and deepseek being way on the bottom. Anyone claiming Grok to not be good hasn't used it. Nearly every time I use ChatGPT or Claude I need to follow up 2-3 times before I get complicated code that I asked for to run correctly. Its typically one and done with Grok (at the cost of it being slower)
I have used it. Multiple times. In actual real life queries, it always falls short compared to the 3 main competitors. They are very good at gaming benchmarks, but when you use it for regular queries (and not always code, anecdotes you give are very specific / personal) it always never scores quite as well as the top 3.
Otherwise, you are quite obviously simping over Musk https://forums.macrumors.com/thread...ns.2473978/page-5?post=34318607#post-34318607, which makes your analysis incredibly subjective. I don’t have fondness for any CEO, so I don’t have the kind of tilt you seem to have.
The thing that makes Grok utterly untrustworthy--and therefore effectively useless--doesn't stem from how well it does or doesn't do on any benchmark, sample query, or anecdote.
It's the thing demonstrated by its answer to the hypothetical about how many innocent people's deaths would be worth trading for Elon Musk's life. Yes, phrasing it in terms of a particular ethnic group was a trick question, because it spelled out the actual logic: In a hypothetical, killing any number of humans up to half the population of earth in exchange for Elon Musk surviving is a net win.
Which is to say that, however good or bad its training dataset and algorithm is, its answers will always be filtered through the manic whims of the comically-bloated ego of the billionaire who is subsidizing it.
If you ask a question that doesn't hit any of the ego-feeding or talking points Musk tweaked the backend with that week, maybe you get a good answer. But there is absolutely no way to tell, so any and every thing you ask might be filtered through whatever Elon has decided it should be telling you that week, right, wrong, or bonkers crazy. Most of the Musk-ified garbage the news reports on are so obvious it's funny--MechaHitler and such--but there's no way to tell what more subtle things have been tweaked away from reality and toward whatever Elon Musk wants people to believe, or just when it does a better job of hiding its ulterior motives.
And fundamentally, while I don't like Musk, it doesn't actually matter whether I do or not--the point is that Grok's answers are being explicitly and very obviously manipulated on the back end by a single person, and that person has no financial interest or external pressure to prioritize accuracy over opinion or ideology. If there were an LLM in exactly the same situation being steered by a person who I didn't hate, it would be no more trustworthy.
And if the person were better at steering it, it would be even worse--instead of ham-fisted lies or exaggerations, it might actually be good at subtly manipulative responses to queries.
It's another flavor of Chinese LLMs that will give you false information about the Tiananmen Square Massacre despite having accurate information in the training data--the Chinese government's desire to manipulate public opinion and what is perceived as fact takes priority over any sort of objective accuracy, and there's no pressure whatsoever in the other direction. People know to look for that specific lie or omission, but you don't know what other things the LLM is subtly trying to manipulate you about at the instruction of those who control it.
No LLM maker is immune to this, but at least most of them are under some financial pressure from their investors or funders to make the thing work well.
Last edited:
I agree with you on the open source bit of the whole argument. I mean they really duped Musk and Microsoft to throw money their way in the name of open source then pulled the rug when they saw the opportunity was too great. Don’t really like OpenAI’s Sam Altman, because you shouldn’t go from nonprofit open source to let me ripoff all the people who donated to fight for a market opportunity. It’s one problem in a host of many for American businesses right now.