Large X models lack a crucial component of “open-source”. Freely redistributable and modifiable for any purpose, sure, but there’s no chance in hell of auditing one, let alone if the training data is kept a secret. It’s literally impossible; human beings cannot look at a trillion weights and biases representing a single highly chaotic, unfathomably complex nonlinear function whose input and output space are the totality of human language/images/etc. and say “yup, looks good to me.” Deep learning models – contrasted with traditional machine learning models – learn their own features which almost 100% of the time would be nonsense to a human. You just have a blob of shareware when you run DeepSeek.
(They also just outright steal from billions of copyright-protected sources to create it, so calling it “open-source” is pretty funny.)
Auditing for bias purposes, yea true. But my primary concern is it having the capability to “phone home” which you don’t really need to audit the model itself to be able to detect or prevent
DeepSeek’s model is open-sourced and can be run locally; though I think there some bits related to its training data they have been kept obscured (if I remember correctly) - likely due to the dubious nature of how it was acquired.
AKA ANY details about its training data, and its training hyperparameters, and literally any other details about its training. An ‘open’ secret among LLM tinkerers is that the Chinese companies seem to have particularly strong English/Chinese training data (not so much other languages though), and I’ll give you one guess on how.
Deepseek is unusal in that they are open sourcing the general techniques they used and even some (not all) of the software frameworks they use.
Don’t get me wrong, I think any level of openness should be encouraged (unlike OpenAI being as closed as physically possible), but they are still very closed. Unlike, say, IBM Granite models which should be reproducible.
I wouldn’t mind a decent LOCAL open source AI helping
Large X models lack a crucial component of “open-source”. Freely redistributable and modifiable for any purpose, sure, but there’s no chance in hell of auditing one, let alone if the training data is kept a secret. It’s literally impossible; human beings cannot look at a trillion weights and biases representing a single highly chaotic, unfathomably complex nonlinear function whose input and output space are the totality of human language/images/etc. and say “yup, looks good to me.” Deep learning models – contrasted with traditional machine learning models – learn their own features which almost 100% of the time would be nonsense to a human. You just have a blob of shareware when you run DeepSeek.
(They also just outright steal from billions of copyright-protected sources to create it, so calling it “open-source” is pretty funny.)
Auditing for bias purposes, yea true. But my primary concern is it having the capability to “phone home” which you don’t really need to audit the model itself to be able to detect or prevent
There are a few that are “truly” open like IBM Granite, and a handful of others over the 7B range.
DeepSeek’s model is open-sourced and can be run locally; though I think there some bits related to its training data they have been kept obscured (if I remember correctly) - likely due to the dubious nature of how it was acquired.
Unless training data is made available, a model is not open source. DeepSeek is better described as “open weight”.
AKA ANY details about its training data, and its training hyperparameters, and literally any other details about its training. An ‘open’ secret among LLM tinkerers is that the Chinese companies seem to have particularly strong English/Chinese training data (not so much other languages though), and I’ll give you one guess on how.
Deepseek is unusal in that they are open sourcing the general techniques they used and even some (not all) of the software frameworks they use.
Don’t get me wrong, I think any level of openness should be encouraged (unlike OpenAI being as closed as physically possible), but they are still very closed. Unlike, say, IBM Granite models which should be reproducible.
Firefox can use a local llamafile model, but you have to enable it in about:config first.
Honestly it’s easier to find an addon that’ll hook to ollama instead, fire fox’s inbuilt support is shit