I like how everyone is suddenly taking China at face value. Disregarding their history of lying to save face and make geopolitical opponents look bad. Much like our own Government come to think of it.
I’d imagine this is going to be a Sputnik moment though. And a lot of people in the AI field are getting very sternly worded messages and promises of more funding.
This coding was very likely not stolen from any of the big players in the West. If you follow ai development closely you will know that the main track of progress for the big dogs has been spending more money on more robust chips and compute.
Meanwhile, Chinese companies like the one behind deepseek have been denied access to these cutting edge chips, and they supposedly trained this model on inferior chips. This model's improvements were made through algorithmic progress, not by just throwing more compute or expensive hardware at the well.
What's more, this model is completely open source with open weights. There are people running the local models on their computer and even got it to output info about tiananmen square and the censorship surrounding it. Whereas their webchat version is much more censored.
And how is that supposed to make me feel better? Because recent events and embarrassments within the OSS community has severely damaged my trust in the community's ability to detect malicious code on a timely basis. Sure nothing too damaging has occurred yet, but it is more to dumb luck than actual strength in the OSS community.
And with the CCP likely backing this DeepSeek project, I will refuse to rely on dumb luck to keep my systems safe from CCP attacks.
Of course, those looking to improve things have the issue of making the first investment.
When you visit Ollama.io, you will notice that a lot of models are available for download and use. You could even use a product like: Jan, to host your own LLM-model.
Ollama allows you to download its software and language models and use it on your ow computer, train it the way you want.
I tried a 16B model, and I can tell you, it was slow as molasses in February. SO, a smaller model would serve me better. Despite that, I am training the model to recognize Latin words, and do an etymological analysis on it, to seek the equivalent that actually means what is means. It is a fun experience.
The opensource community in general is what it is: once you put up the code, others can use it and improve on it.
If I had to choose between language models and paying for tokens, I would not want my money to go to the CCP. As far as I am concerned, the CCP is a bad egg.
Looks like DeepSeek China is getting attacked. Could this be AI against AI?
The cyberattack on DeepSeek, while not fully detailed, appears to have been a sophisticated and large-scale operation targeting the rapidly rising Chinese AI startup.
According to DeepSeek's own statements, the company fell victim to "large-scale malicious attacks" on its services, prompting immediate action to limit user registrations and mitigate potential damage12.
The incident occurred shortly after DeepSeek's AI chatbot surged to the top of Apple's App Store downloads in the United States, overtaking established competitors like OpenAI's ChatGPT3.
While the full extent of the attack remains unclear, the swift response from DeepSeek highlights the severity of the threat.
The company has engaged cybersecurity experts to investigate the breach and assess its impact1. This incident underscores the growing cybersecurity challenges faced by AI companies, particularly those experiencing rapid growth and increased public exposure.
It also raises concerns about the security of user data and the potential vulnerabilities in emerging AI platforms, especially as they gain prominence in the global tech landscape
I like how everyone is suddenly taking China at face value. Disregarding their history of lying to save face and make geopolitical opponents look bad. Much like our own Government come to think of it.
I’d imagine this is going to be a Sputnik moment though. And a lot of people in the AI field are getting very sternly worded messages and promises of more funding.
I am not.
I have known China for lying about their tech for years, why wouldn't they lie about their AI achievements?
I bet they stole the AI coding too.
^^This^^
This coding was very likely not stolen from any of the big players in the West. If you follow ai development closely you will know that the main track of progress for the big dogs has been spending more money on more robust chips and compute.
Meanwhile, Chinese companies like the one behind deepseek have been denied access to these cutting edge chips, and they supposedly trained this model on inferior chips. This model's improvements were made through algorithmic progress, not by just throwing more compute or expensive hardware at the well.
What's more, this model is completely open source with open weights. There are people running the local models on their computer and even got it to output info about tiananmen square and the censorship surrounding it. Whereas their webchat version is much more censored.
And how is that supposed to make me feel better? Because recent events and embarrassments within the OSS community has severely damaged my trust in the community's ability to detect malicious code on a timely basis. Sure nothing too damaging has occurred yet, but it is more to dumb luck than actual strength in the OSS community.
And with the CCP likely backing this DeepSeek project, I will refuse to rely on dumb luck to keep my systems safe from CCP attacks.
Of course, those looking to improve things have the issue of making the first investment.
When you visit Ollama.io, you will notice that a lot of models are available for download and use. You could even use a product like: Jan, to host your own LLM-model.
Ollama allows you to download its software and language models and use it on your ow computer, train it the way you want.
I tried a 16B model, and I can tell you, it was slow as molasses in February. SO, a smaller model would serve me better. Despite that, I am training the model to recognize Latin words, and do an etymological analysis on it, to seek the equivalent that actually means what is means. It is a fun experience.
The opensource community in general is what it is: once you put up the code, others can use it and improve on it.
If I had to choose between language models and paying for tokens, I would not want my money to go to the CCP. As far as I am concerned, the CCP is a bad egg.
I think you are correct. I think they are hyping it to get a lot of code for free. Kind of an empty shell at this point.
Or, I think it is a trap for important computer engineers and companies to allow a potential backdoor for CCP hackers to start stealing tech again.
That was my first thought. Download a chinese made "super intelligent" AI onto my computer? Nah, no thanks
Except for https://x.com/Cointelegraph/status/1883918055866081298
Looks like DeepSeek China is getting attacked. Could this be AI against AI?
The cyberattack on DeepSeek, while not fully detailed, appears to have been a sophisticated and large-scale operation targeting the rapidly rising Chinese AI startup.
According to DeepSeek's own statements, the company fell victim to "large-scale malicious attacks" on its services, prompting immediate action to limit user registrations and mitigate potential damage12.
The incident occurred shortly after DeepSeek's AI chatbot surged to the top of Apple's App Store downloads in the United States, overtaking established competitors like OpenAI's ChatGPT3.
While the full extent of the attack remains unclear, the swift response from DeepSeek highlights the severity of the threat.
The company has engaged cybersecurity experts to investigate the breach and assess its impact1. This incident underscores the growing cybersecurity challenges faced by AI companies, particularly those experiencing rapid growth and increased public exposure.
It also raises concerns about the security of user data and the potential vulnerabilities in emerging AI platforms, especially as they gain prominence in the global tech landscape