3 ms·

Apple does their own language model development and worst case could host eg Qwen or Deepseek. That would be an absolutely insane waste of cash. Their problem

Apple does their own language model development and worst case could host eg Qwen or Deepseek. That would be an absolutely insane waste of cash. Their problem isn't inability to develop or lack of access to technology, they're extremely good at that, They're like a vertically integrated combination of Microsoft and Nvidia. Their problem is the lack of vision and inability to productise what they develop. IE their problem is that they're Apple, buying other companies won't change that.

znpy 1y ago
> Apple does their own language model development and worst case could host eg Qwen or Deepseek. That would be an absolutely insane waste of cash. At those levels you don't buy other companies to only get the tech, you also acqui-hire the talents.
- msgodel 1y ago
  That's the thing though. Apple's internal ML team has amazing output for the small amount of money they're given. There are very few companies with better talent. If they scaled up the Capex for ML they could probably match other big tech's language models. They don't have to though. Language models are commodities and the weights are free.
  - bitpush 1y ago
    > Language models are commodities and the weights are free That's like saying chips are a commodity. You can buy from Intel, or AMD. Why bother making something yourself. And yet, Apple has proven that unless you vertically integrate you're not going to differentiate.
    - msgodel 1y ago
      Eh. Apple's chips are good but not exactly game changing. At the end of the day if they're doing anything useful they still run the same software everyone else's do with roughly the same performance. It's a great analogy for the situation with language models. Can you get a slight edge with the particular RL tuning on some models? Sure. Does it actually result in qualitative changes in what the model can do? No not really, all of them can do MCP, all of them still hallucinate etc.
  - znpy 1y ago
    > Apple's internal ML team has amazing output for the small amount of money they're given. You're probably missing the point. The output is certainly good when compared to the number of people involved. But if you need "more output" at some point you'll need more people. Also, new people usually bring new know-how. No company ever died from having too much know-how.

znpy 1y ago

> Apple does their own language model development and worst case could host eg Qwen or Deepseek. That would be an absolutely insane waste of cash. At those levels you don't buy other companies to only get the tech, you also acqui-hire the talents.

msgodel 1y ago
That's the thing though. Apple's internal ML team has amazing output for the small amount of money they're given. There are very few companies with better talent. If they scaled up the Capex for ML they could probably match other big tech's language models. They don't have to though. Language models are commodities and the weights are free.
- bitpush 1y ago
  > Language models are commodities and the weights are free That's like saying chips are a commodity. You can buy from Intel, or AMD. Why bother making something yourself. And yet, Apple has proven that unless you vertically integrate you're not going to differentiate.
  - msgodel 1y ago
    Eh. Apple's chips are good but not exactly game changing. At the end of the day if they're doing anything useful they still run the same software everyone else's do with roughly the same performance. It's a great analogy for the situation with language models. Can you get a slight edge with the particular RL tuning on some models? Sure. Does it actually result in qualitative changes in what the model can do? No not really, all of them can do MCP, all of them still hallucinate etc.
- znpy 1y ago
  > Apple's internal ML team has amazing output for the small amount of money they're given. You're probably missing the point. The output is certainly good when compared to the number of people involved. But if you need "more output" at some point you'll need more people. Also, new people usually bring new know-how. No company ever died from having too much know-how.