xAI Launches Grok-3 AI Model, Claims Superior Performance Over GPT-4

[turbineseaplane]

Exactly. They are forcing something, that would need more time. But no, the want results right now. Like a startup company that needs to be sold within a years, because if not, it will go down under.

100% -- Everyone is trying to capture a first mover advantage and be the entrenched early "winner"

 

[Rigby]

I'm fascinated by the AI denialism that I see here on MR. I've mentioned several times recently that in the past 6 months my work has required me to use AI, and going from essentially zero understanding to where I'm at now has completely convinced me that AI is both here to stay and will be massively disruptive. And the speed at which the models are improving means this is all happening sooner rather than later.

$40/month is a pittance compared to the productive work that AI does. And I too pay the $20/month Chatgpt fee; and it's a steal.

I agree. I was skeptical too based on the early models, but I've been using ChatGPT for several months now and it's been a huge time saver. Obviously you have to verify the results, but that is true for results you find with a search engine too. And the new reasoning models are starting to do this on their own now to an extent.

I recently had the opportunity to test OpenAI's new "deep research" model. It's not perfect, but it kind of blew me away.

 

[teak421]

I tested the new Grok against the old Grok, Google Gemini, and ChatGPT to construct a description in a movie script with some general details about the subject. I then copied each response into the AI Detector. Everything except Grok 3 came back as an AI Generator, but Grok 3 did not. It was impressive.

Regardless if you hate Musk, his team did a great job.

 

[turbineseaplane]

Obviously you have to verify the results, but that is true for results you find with a search engine too. And the new reasoning models are starting to do this on their own now to an extent.

That has some really slippery slope concerns to it for me

 

[turbineseaplane]

I tested the new Grok against the old Grok, Google Gemini, and ChatGPT to construct a description in a movie script with some general details about the subject. I then copied each response into the AI Detector. Everything except Grok 3 came back as an AI Generator, but Grok 3 did not. It was impressive.

Regardless if you hate Musk, his team did a great job.

Because it evades being detected by an AI generator, it's impressive?

I'm not tracking why that's a metric of value here

This is as likely to be a flaw in the AI Detector as anything else

 

[Timo_Existencia]

It's much more nuanced than outright "denialism"

AI is very useful for certain things, right now

But it's being force fed into situations that it's not ready for and it's a bad move as it's souring people on it, for good reason

Those situations that "it is not ready for" are being rapidly improved and anyone who rests their arguments about the (near) future of AI on its weaknesses don't really understand how quickly it is evolving.

I've had opportunity to use Chatgpt in an extremely subjective field (not STEM) and AI combined with my personal expertise catapulted my productivity immensely. Not even close. AI could not replace me (yet) but combined with my expertise, it changed everything about how I worked on and completed my project.

 

[turbineseaplane]

Those situations that "it is not ready for" are being rapidly improved and anyone who rests their arguments about the (near) future of AI on its weaknesses don't really understand how quickly it is evolving.

I would conversely argue that anyone who assumes a linear (or greater) tangible improvement curve is potentially over optimistic

I guess we'll find out - so no need to banter much about it really
👍

 

[teak421]

Because it evades being detected by an AI generator, it's impressive?

I'm not tracking why that's a metric of value here

This is as likely to be a flaw in the AI Detector as anything else

Then why did it properly detect AI-generated text in the others? It was a BASELINE.

In terms of you not tracking the value... do you think I care? I don't.

 

[cardfan]

I tested the new Grok against the old Grok, Google Gemini, and ChatGPT to construct a description in a movie script with some general details about the subject. I then copied each response into the AI Detector. Everything except Grok 3 came back as an AI Generator, but Grok 3 did not. It was impressive.

Regardless if you hate Musk, his team did a great job.

Meanwhile, Apple Intelligence is always an option..lol

 

[turbineseaplane]

Then why did it properly detect AI-generated text in the others? It was a BASELINE.

It's evading the current tools to detect AI -- that doesn't mean much qualitatively other than it beats the detectors

In terms of you not tracking the value... do you think I care? I don't.

This is not a response that fosters good debate or dialogue
Can we please disagree while still being polite?

 

[WarmWinterHat]

I'm fascinated by the AI denialism that I see here on MR. I've mentioned several times recently that in the past 6 months my work has required me to use AI, and going from essentially zero understanding to where I'm at now has completely convinced me that AI is both here to stay and will be massively disruptive. And the speed at which the models are improving means this is all happening sooner rather than later.

$40/month is a pittance compared to the productive work that AI does. And I too pay the $20/month Chatgpt fee; and it's a steal.

I would really like everyone who posts such things tell what job/position they have. It seems to be almost universally IT people and coders, which is fine and great, but it doesn't mean it's useful in other areas.

 

[webkit]

$40 for an AI? I heard that DeepSeek is free and it's practically the same as ChatGPT (since they all "learn" from themselves).

I heard LLM need a lot of computational but I was expecting text-to-image to be more demanding. You have all these generated images plaguing the internet. So much generated garbage that it's hard to not be concerned about blocking all that crap.

Again, $40? Who pays for it?

Many AI products have basic "free" versions as well as pay or subscription options including ChatGPT, DeepSeek, Google Gemini, etc. Paying or not paying can depend on the level/features you want access to and/or amount you use it.

 

[Timo_Existencia]

I would really like everyone who posts such things tell what job/position they have. It seems to be almost universally IT people and coders, which is fine and great, but it doesn't mean it's useful in other areas.

My professional career is in Crisis Communications and political consulting.

Non-professionally, I'm working on a very large project in political philosophy and a new way to organize community.

 

[Sigsgaard]

Personally I would rather use the Chinese one than anything from Musk 🤬

 

[Rigby]

I tested the new Grok against the old Grok, Google Gemini, and ChatGPT to construct a description in a movie script with some general details about the subject. I then copied each response into the AI Detector. Everything except Grok 3 came back as an AI Generator, but Grok 3 did not.

That doesn't mean anything since AI Detector has not been trained on Grok output so far.

 

[turbineseaplane]

That doesn't mean anything since AI Detector has not been trained on Grok output so far.

That was the point as I was getting at before above -- thank you for articulating it better than I did 🙏

 

[ProbablyDylan]

Man, Anthropic needs to come out with something. Can't believe they're letting Altman and Musk run circles around them.

 

[teak421]

It's evading the current tools to detect AI -- that doesn't mean much qualitatively other than it beats the detectors



This is not a response that fosters good debate or dialogue
Can we please disagree while still being polite?

Let me help you.

Which AI detector tool did you use? Did you compare that against other tools? How did it fair? What specific text did you use? Maybe you should try some additional types of text besides screen descriptions. Let me give it a try to compare my findings. I'll get back here to my conclusions. Thanks for posting!

THAT"S how normal people do it.

 

[Rigby]

That has some really slippery slope concerns to it for me

How so? Basically it ensures better consistency of the answers in the context of the training data and reduces the error probability.

 

[turbineseaplane]

Let me help you.

Which AI detector tool did you use? Did you compare that against other tools? How did it fair? What specific text did you use? Maybe you should try some additional types of text besides screen descriptions. Let me give it a try to compare my findings. I'll get back here to my conclusions. Thanks for posting!

THAT"S how normal people do it.

This 👇

That doesn't mean anything since AI Detector has not been trained on Grok output so far.

 

[ThailandToo]

The same reason most fanboys on here give Tim credit. We don’t actually believe Tim is coming up with all the ideas for new products and services himself.

I give Tim no credit for any innovation out of Apple only value to AAPL shareholders, the board and himself.

 

[8KYUP]

Can’t wait for this to be available in Teslas.

Yes, that way you can ask it why you were getting those odd salutes everywhere you go.

 

[It’s always something]

I asked ChatGPT what it knew about me. It answered with some correct information, but then concluded with the “fact” that I am dead. 💀🙃

I then asked it about my most recent blog posts, which it correctly identified as dated after my “death”. When I pointed out the discrepancy, it agreed and “realized” that there was a problem. It resolved the problem by supposing that my last blog post may have been scheduled, or that someone else had posted it. In other words it doubled down on my death 😬

I’m impressed that it could be (seemingly) led to recognize the problem with its own answer, but less impressed by its digging in its heels.

 

[turbineseaplane]

How so? Basically it ensures better consistency of the answers in the context of the training data and reduces the error probability.

I guess it all comes down to my concerns about objective truth

I'm just unclear on how machine learning is going to parse that situation in a way where I'd ever be comfortable assuming the output is accurate to the point of not questioning any aspects of the chain that got it there

It's a little bit like building on an imperfect foundation -- the taller the building on top, the more magnified tiny issues at the bottom are.

It's unclear to me that systems are being built in to understand how output was derived and why

Have I missed reporting on that?
It seems like they are all just blowing by that with basically no concern

Regardless of the usefulness at the moment, I do think everyone should be way more skeptical about all this as you don't put the toothpaste back in the tube if this gets rolled out sufficiently before ready for primetime

 

[Howard2k]

Not using Grok, but ChatGPT saved me probably 40 hours of work over the past two weeks on a project that had an extremely tight turnaround. Was absolutely worth the $20 a month I pay; in fact would have been worth the $200 a month for their Pro version that I don't subscribe to (but really glad that $20 option is there 😂).

Yep, it's saved me countless hours too through some automation stuff for repetitive tasks. Probably more than 50 hours over the last few months.