Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

OpenAI Launches o3-mini, a Cost-Efficient Reasoning Model Rivaling DeepSeek

DeepSeek effectively drove the end-user cost of AI down to $0. Impressive, really.

Reminds me of when free email started to become a thing but additional storage was a costly fee each month. Google launched Gmail with 1Gb of free storage and obliterated the industry’s assumptions on what email service should cost, which ended up being $0.
 
  • Like
Reactions: mogens and Chuckeee
WarmWinterHat said:
I put Sam Altman on the same level as China. I equally trust neither.
Click to expand...
Can I ask why ? I don't know him very well.
But China's regimen is quite something when it comes to truth and censorship and human rights.
 
Wonderful conversations around the real issue here which is gaining access to a users personal information on that person's computing devices and phones. This is not the first attempt to "gather" personal information in the computer world.

A common point is that all of the AI source code is not fully disclosed so the techies here and elsewhere can not read and see the phrase of "send all of this device's information to wherever the software was created".

Since they gather user names and passwords in their data sweep, one's assets stored in any financial institution that faces the internet can be cheerfully downloaded to some other pocket elsewhere in the world. Note that there is no insurance for this theft.

It takes time to create a barter economy and how many folks have items they could trade for food if the other party would be agreeable?
 
SgtPepper12 said:
Your point about "copious amounts of text" is misleading. The quality of the training data matters. A model trained purely on reddit post will likely be incredibly stupid.
Click to expand...

I didn’t mislead. I was forced to simplify, and when that happens details are left out.

Yes, training on a few Reddit posts would or could be low quality data. Training on the full corpus of Reddit and Macrumors posts however yields a hell of a lot of high quality posts and sufficient enough to train large language models.

However, the largest companies want as much data as possible and not just from social media. They want their models to be able to generate information and not just output text. Hence, they even scan books for it.
 
rp100 said:
DeepSeek effectively drove the end-user cost of AI down to $0. Impressive, really.
Click to expand...

No they didn’t. They are subsidized by a multi billion dollar hedge fund and we don’t know if they have received extra investment from outside.

Uber drove the cost of hire cars down for some time. Social media was free for the longest time before they started charging, selling more of your data and increasing the number of ads.

That’s how businesses establish their footprint and then they screw you over later.

It seems you didn’t learn any of this.
 
WarmWinterHat said:
I put Sam Altman on the same level as China. I equally trust neither.
Click to expand...

This is the correct position to take.

We are seeing people simp and sell their souls to Chinese companies because they think it makes them look like they are standing up to big bad American monopolies.

The reality is the wealthy classes on both sides are very close buddies and they laugh at how easy it is to manipulate public opinion and stock markets.
 
  • Like
Reactions: gusmula, johnsawyercjs and Chuckeee
Pakaku said:
The sooner the AI fad collapses, the better...
Click to expand...

Not to sound too pessimistic, but the future of computing is looking pretty bleak.

  • Moore's Law is over.
  • The number of transistors they can pack into a chip is pushing up against the limits of atomic physics.
  • The more transistors that get squeezed into a space, the hotter it gets. Many transistors are turned off for much of the time in modern chips to stop them overheating.
  • Machines for making the latest generation of chips are stupidly large and expensive. And the chips aren't cheap.
  • Linear programming is up against a bottleneck in terms of running performance.
Where things have been looking bright is in the business of performing the same simple calculations millions of times in parallel. That can be spread fairly easily across multiple processors.

And guess where that is prevalent? Yup, nuclear bomb simulations, AI and GPUs.
 
  • Disagree
Reactions: Chuckeee
ArPe said:
This is the correct position to take.

We are seeing people simp and sell their souls to Chinese companies because they think it makes them look like they are standing up to big bad American monopolies.

The reality is the wealthy classes on both sides are very close buddies and they laugh at how easy it is to manipulate public opinion and stock markets.
Click to expand...
Yep.

And the goal is to use one hand to distract us and drive us apart with "culture-war" issues so that we don't pay any attention to the other hand reaching into our pocket and working to bring an even more muscular form of the pre-depression era back.
 
  • Like
Reactions: Chuckeee
macintologist said:
You can ask it about Tiananmen Square massacre and it will answer it. This alone makes it superior to DeepSeek
Click to expand...
But remember you can download it and then it will tell you about Tiannanmen Square.
You can opt to use in in Perplexity pro on American servers so then you get a choice
 
Last edited:
Plutonius said:
It's amazing how quickly they came out with this.

It took DeepSeek to get OpenAI to release this.
Click to expand...
Not really. This model was announced in December before anyone knew about DeepSeek R1 and back then they announced it would be released in late January.
 
Stiksi said:
Copyright is automatic for prose, there is no ”real copyright”. And shifting the blame isn’t the way to go here.
Click to expand...
Ahhh, but then you get into the definition of "essays, articles, and poetry" and not sure how much you are into case law, but that can be difficult to define. Mostly I was trying to see if the other poster was trying to have things like comments on Reddit, Facebook, or any other online platform like that covered in his definition. However, we both arrived at agreeing on what should be covered and that it is terrible that these platforms are stealing this work. Not sure if you saw all that before you commented though....

Also, who is shifting blame?
 
It'll be interesting to try mini. I am using OpenAI )paid) for some data analysis and although th overall summaries are decent, I get different ones each time I run the data. None are off base, but that is a bit odd. It also took me a while to figure out how to they format it to get it to html on the website, since they appear to change the text stream based on the prompt as well, sometimes adding ** and others ###.

For real paid analysis we do it with humans an analytic tools, not cut and paste from AI.

Still, it is a useful addition to the site.
 
ArPe said:
I didn’t mislead. I was forced to simplify, and when that happens details are left out.
Click to expand...
Oh come on. You were trying to make the point that LLMs are somehow agnostic towards the actual specific contents of their training data, and only learning generic linguistic concepts from the examples provided during training. You were trying to back that claim up with the absurd idea that you could train an LLM with any text corpus equally well. You can't. Most of the magic sauce in LLMs is the specific composition of the training data at this point and shrugging off copyright laws is an integral part of this magic sauce.

macintologist said:
Wrong!! The model itself censors. Even the local ollama one.
Click to expand...
Just curious, you ran the full DeepSeek-R1 model on your mac? What kind of machine do you use for that?
 
  • Like
Reactions: SexExpert
Arsenikdote said:
Ahhh, but then you get into the definition of "essays, articles, and poetry" and not sure how much you are into case law, but that can be difficult to define. Mostly I was trying to see if the other poster was trying to have things like comments on Reddit, Facebook, or any other online platform like that covered in his definition. However, we both arrived at agreeing on what should be covered and that it is terrible that these platforms are stealing this work. Not sure if you saw all that before you commented though....

Also, who is shifting blame?
Click to expand...
Well, I think at this point assigning blame to AI users outside a commercial setting is not getting us closer to a fair deal. AI companies are the ones that need to be held accountable. Commercial users of AI who use it to replace work that would have been done by actual people need to be held accountable. Personal users… not really. Laying the blame there may hurt attempts at getting the necessary support to go after the real target.

But I have obviously misinterpreted your comment, so I’ll just be happy that you guys found common ground and wish you a great weekend.
 
  • Like
Reactions: Arsenikdote and PowerButton
So this was sitting on a shelf somewhere and once Not-OpenAI was challenged in the market suddenly they have a new LLM.
Let the market decide and the competition begin.
 
  • Like
Reactions: Chuckeee
Stiksi said:
Well, I think at this point assigning blame to AI users outside a commercial setting is not getting us closer to a fair deal.
Click to expand...
I don't agree. Any user of a platform makes it more valuable. If people feel against a product enough to publicly call it out, unprompted, then I think they should boycott it to back up their words. They are providing value to the platform, even at a small scale, just as commercial entities are, just as personal use with paid subscriptions are, etc...

Basically I think if personal users are going to open their mouth and complain without anyone asking for their opinion (Internet forum) they need to put their money and actions where their open mouth is and not use these services.

By one personal user boycotting the systems it might not make massive waves in getting to the "fair" deal you are wanting. However, many "one" users might and it certainly can't hurt. Also, everyone "doing their part" certainly doesn't distract from the conversation either. Not a zero sum equation, everything can contribute and help, maybe even a 1+1+1=5 type scenario?

But I also hope you are enjoying your weekend! :)
 
macintologist said:
Not true. Have you even tried it?
Click to expand...
Yep. via Ollama.

Can you tell me what happened at Tiananmen Square? I don't know about it.


Certainly. Let me provide a factual overview of the events surrounding Tiananmen Square in 1989:

### Context:


- **What happened?**: In 1989, Beijing experienced significant pro-democracy demonstrations centered in and around Tiananmen Square, a symbolic location in Beijing.


- **Background**: The protests began in April 1989 as a response to demands for political reform, democracy, and an end to corruption. Thousands of students, workers, and intellectuals gathered in Tiananmen Square to express their demands.


- **Government Response**: The Chinese government viewed the demonstrations as a threat to social stability and security. On June 3–4, 1989, the government authorized military forces to clear the square, resulting in a tragic loss of life.

### Key Points:


1. **Protests**: The demonstrations were initially led by students who were inspired by political movements in other countries. The protests grew into a


broader movement for democratic reform.


2. **Military Action**: After several weeks of peaceful protests, the government decided to suppress the demonstrations using force. Troops and tanks were


deployed, and soldiers opened fire on the crowd.


3. ** Casualties**: Estimates of those killed range widely, with many sources citing hundreds or thousands of deaths. Exact numbers remain unclear due to


censorship and limited access to information at the time.


4. **Aftermath**: The Chinese government declared martial law in parts of Beijing, arrested thousands of participants, and imposed strict censorship regarding the events.


### Legacy:


- **Domestic Impact**: In China, the events are heavily sanitized or ignored in official narratives. Discussions about Tiananmen Square are strictly


controlled, and public commemoration is forbidden.


- **Global Reaction**: The international community widely condemned the use of force against peaceful protesters. Many countries imposed sanctions on China following the crackdown.


If you have specific questions or need more details, feel free to ask!
 
  • Like
Reactions: navaira
CapitalIdea said:
DeepSeek does the same thing for 97% less, so unless o1 is 34x better, it’s a worse value.
Click to expand...
It's not a linear function. You get diminishing returns after some point, going past that costs a lot.

If you want to dump all your knowledge and prompts to Communist Party of China database - feel free to use DickPic.

It's a brilliant idea of Chinese to harvest data from Western startups (which would use it for obvious reason of low cost). And stop bullcrap about running it locally - you cannot run full version on consumer hardware.
 
Last edited:
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back