Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Exactly. This is why I'm benefitting. I can do those things well enough for my projects that I don't need to hire anyone to write the laborious parts of the code. I'm spending far more time doing design and architecture now than I ever did coding, to an enormous degree. The productivity gain blows my mind every time I stop and think about it.
I'm in a similar situation. I'm a scientist who needs to do a fair amount of coding (mostly scripting) to do my research analyses. I can do way more in the same or less time than before. I can plug into ChatGPT what I need and with some back and forth, end up with the code to do it. Sometimes it takes many iterations and days to get it, other times it's almost exactly what I need the first output.

A benefit is I don't need to know all the languages or have someone code for me. I'm very comfortable with bash. I am decently comfortable with R and have some familiarity with Python. I have minimal Matlab knowledge. Now I can use any of those languages to do what I need to do without really knowing the languages. Some people might be concerned about that but I know what to ask and what the output should be. It's freeing of so much time to let the LLM produce and document much of the code. It will even take an approach I would not have thought of.

I know some people don't trust the output (I don't either until I verify it) and don't like AI -- I get that and understand that -- but I have a specific use for them, at which they excel. These tools are one of the greatest things to happen in my career. They free me up to do much more science. I can have more elegant analyses than I and a team of people could create in a reasonable time.
 
I have no sympathy for closed AI. There is nothing open about it, after getting hundreds of millions in spirit of being open. Deepseek is a game changer.
Open AI: we can train on data from internet, but others can’t.
Deepspeak: we are stealing content from websites and stealing technology developed by other companies.
 
I don’t like Chinese products or services. But, by releasing Deep seek they have proved that it’s the skill of developers important, not just the purchasing power of corporate. If Apple had hired such developers, we could have had Apple Intelligence on all devices which can run iOS 18.
 
It has done much better at math and coding, very helpful to use it locally with visual studio code. Much better than meta Llama models which are open source and were trained with lot more compute and cost.

This is just the beginning, huge barrier to entry in costs for training is starting to crumble.
costs will be low when they don't have to invest in R&D, copy technology developed by other companies.
 
I downloaded the app and I was using it a bit last night. I went to fiddle with it this morning and it got brain freeze. Couldn’t even answer basic questions.

I smell a lot of hype. The short sellers probably made giant bags of money the past few days. They’ll throw those winnings back into the market and ride the rebound.

I think it’s a flash in the pan, much the same as AI in general. Until it’s smarter than the cat, it’s just a bunch of hype.
It’s a small company they just got overwhelmed by the volume and wasn’t prepared to handle this type of usage, it will be sorted over time.

What they did is a game changer not just due to open source and hardware agnostic but showed a different approach at a significant cost reduction. Think of it as a foundation building block that anyone can now leverage been open source to continue build on top in any speciality rinse and repeat, it will accelerate progress significantly.

Their result is also impressive (when it’s not getting timed out), I ran a set of test through it like creating a specific infinite mana magic gathering card game competitive deck, writing complex connected recursive functions in different languages, explain in English thousand year old Chinese poems (written in Chinese) etc.. results are all better than the $200 ChatGPT model.

This does two things: 1) wrestles some of the control from private tech such as openai meta and google back to public domain to level the playfield 2) starts a spark that maybe you don’t need to bow down to nvida hardware monopoly (tbd)

Lastly the amount of racism and bias against China in those responses are mind boggling at this day and age. The sooner you take your head out of the sand and realize how fast China is advancing vs the shtshow here the sooner we can wake up. Go take a trip to Shanghai and shenzhen, it makes nyc looks like a city in the 80s.
 
Last edited:
Using the output from a "trained model" to train another model is the issue here. That's technically theft. The investment went into training the original model to begin with. Training is a highly-involved and expensive process, and it sounds like DeepSeek might have jumped the line. I'm not surprised if that's the case, but I'm really hoping it's not true.
It sounds like DeepSeek took OpenAI's hard work and bypassed the training process entirely. Totally different from siphoning data for the original training, of which OpenAI may be guilty of indeed.

No, that is not what they did. They did not "bypass the training process entirely". They used OpenAI as a tool to improve their already existing model. From reading up on this today it appears distillation is done all over the industry by every company. It’d be stranger if they hadn’t done it.
 
The US firms didn't "steal" data. It was freely available on the web.

And the Chinese are stealing IP, which is not the same as copying something being freely available to read, download, etc, off the internet.

Man, lot of America haters on MR.

So I publish a website loaded with data and put in place terms of use which forbid the consumption of that data by robots, web crawlers and so on. When OpenAI scrape my website they violate those terms.

If DeepSeek has violated OpenAI’s terms of use by training their model the way they did, and you conclude that such behaviour amounts to theft, then OpenAI’s original sin must amount to the same thing.
 
As a software developer, I use AI all day every day for extremely complex tasks. The results border on miraculous. The more experienced I get with AI development, the better it gets. I'm an experienced late-career dev, and this is better than having multiple junior developers working for me, right now, today. I can only imagine what things will be like a year or two from now. It's amazing.
And when you retire, where will the next generation of experienced developers come from?
 
costs will be low when they don't have to invest in R&D, copy technology developed by other companies.
Except this is not copy like usual Chinese phones or cars. They invested in R&D to do some thing more efficient and in an innovative way. Open AI, Microsoft and Nvidia drank too much kool aid thinking barrier to entry is too high. Deep seek did this with low end chips, what others were struggling to do.
 
I use Cursor w/ Claude Sonnet 3.5, what do you use?
I use llama and chatGPT depending on my requirements. I tried using agent style, but prompting tokens/sec with context is slow on my M1 Max with 64 GB. I may upgrade to M4 max 128 GB with my new work starting soon. It will pay itself in a week. Might try for couple weeks and decide to keep M4 or not.
 
  • Like
Reactions: ILoveCalvinCool
And when you retire, where will the next generation of experienced developers come from?
And not just that. What about the ownership issues of the IP vested in the code being produced by AI. Is it original work? It may be well be part of a proprietary algorithm snagged from god-knows where.
 
And when you retire, where will the next generation of experienced developers come from?

The next generation of experienced developers will not be coders. They'll be a more professional version of what is currently called a "prompt engineer". They'll be using much more sophisticated tools to guide AI in extremely complex development environments. I feel like I'm in training for that job right now, in one sense, though I'm past the point where these skills are going to be necessary for me to hold a job.

The ones who learn to code with AI.

Agreed. It's clear that you and I are both doing this type of development right now.
 
This just shows that openAI never had much of a defensive moat to begin with, if their product can be copied so easily.

You see this in the way Apple designs their products. People can copy the look of their hardware, but they will be hard pressed to replicate the software and the overall ecosystem. The end result is that if you want the unique Apple ecosystem, you pretty much have to buy Apple products. That is by design.

Something for openAI to think about as they figure out where to go next. They were so busy copying other people’s stuff that they didn’t stop to think that others could well be doing the same to them.
 
But it's okay for OpenAI to steal our data and train is LLM on it? They wanted to corner the market and make obscene profits from our data and DeepSeek has shaken all that up. If AI is unavoidable then it should be freely available for all.
 
Hey but where are OpenAI’s open sourced models for others to use under MIT license? Ain’t OpenAI open when they stole copyrighted materials from others to build their LLM in the first place?
 
  • Like
Reactions: ILoveCalvinCool
This is rich. OpenAI have been stealing copyrighted material on the internet to train their model. They say is legal cause is on the internet, so is open to the public. Irony is completely lost on them.
 
  • Like
Reactions: Chuckeee
Say what you want about it using OpenAI to train its model, OpenAI used all of us without our permission to train theirs.

And honestly this isn't even the biggest feat, it seems everyone is ignoring the fact that China pulled this off while being locked out of chips that Nvidia told the world we'd need if we ever wanted to go anywhere in the computer world. This is the actual breakthrough, not the fact it used another open source AI to train it. This makes something as dense and powerful as OpenAI more affordable to run offline, making its uses nearly endless.
 
  • Like
Reactions: Velli and surferfb
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.