Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Meta Employees: AI Team Is in "Panic Mode" After DeepSeek R1 Model Release

winbuzzer.com

Post reply on HN
Print view View original

1–10 of 11 posts

Re: Meta Employees: AI Team Is in "Panic Mode" After DeepSeek R1 Model Release

#4

Post by Jasondells »

DeepSeek is a censored product and by that of limited use for anything that might require prompts that are about anything that is somehow "controversial" in the eyes of the CCP. However, the censorship seems to be applied for certain prompts and doesn't seem to be integrated into the model itself as the answers given to such prompts are very similar and generic.

Has anybody already been able to successfully use prompt jailbreaking or other tricks to overcome this? It would be interesting to see what DeepSeek actually knows instead of what it is responding.

Censoring a model via selective training data or post-training is much more difficult.

The possible "solutions" applied to this "problem" (in the eyes of the censors) will be of high importance moving forward.

Other gov. actors also have an interest in altering models, let's not forget.

Re: Meta Employees: AI Team Is in "Panic Mode" After DeepSeek R1 Model Release

#5

Post by elpocko »

Jasondells wrote: post #4

DeepSeek is a censored product and by that of limited use for anything that might require prompts that are about anything that is somehow "controversial" in the eyes of the CCP. However, the censorship seems to be applied for certain prompts and doesn't seem to be integrated into the model itself as the answers given to such prompts are very similar and generic. Has anybody already been able to successfully use promp…

Every LLM is a censored product and by that of limited use for anything that might require prompts that are about anything that is somehow "controversial" in the eyes of the model censor and their masters.

There is a process called "abliteration" [0] that can be used to undo some of the censorship, at the cost of making the model slightly™ dumber (according to users of those models).

[0] https://huggingface.co/blog/mlabonne/abliteration

Re: Meta Employees: AI Team Is in "Panic Mode" After DeepSeek R1 Model Release

#6

Post by Jasondells »

elpocko wrote: post #5
Jasondells wrote: post #4

DeepSeek is a censored product and by that of limited use for anything that might require prompts that are about anything that is somehow "controversial" in the eyes of the CCP. However, the censorship seems to be applied for certain prompts and doesn't seem to be integrated into the model itself as the answers given to such prompts are very similar and generic. Has anybody already been able to successfully use promp…

Every LLM is a censored product and by that of limited use for anything that might require prompts that are about anything that is somehow "controversial" in the eyes of the model censor and their masters. There is a process called "abliteration" [0] that can be used to undo some of the censorship, at the cost of making the model slightly™ dumber (according to users of those models). [0] https://huggingface.co/blog/m…

Thanks.. what I am interested in is the the type of censorship that has been applied and what effects prompt-jailbreaks have. Personally I have limited experience with that

Re: Meta Employees: AI Team Is in "Panic Mode" After DeepSeek R1 Model Release

#7

Post by verdverm »

Jasondells wrote: post #4

DeepSeek is a censored product and by that of limited use for anything that might require prompts that are about anything that is somehow "controversial" in the eyes of the CCP. However, the censorship seems to be applied for certain prompts and doesn't seem to be integrated into the model itself as the answers given to such prompts are very similar and generic. Has anybody already been able to successfully use promp…

Huggingface is working on reproducing R1, as I'm sure many others are

https://github.com/huggingface/open-r1

Re: Meta Employees: AI Team Is in "Panic Mode" After DeepSeek R1 Model Release

#8

Post by mmh0000 »

Jasondells wrote: post #4

DeepSeek is a censored product and by that of limited use for anything that might require prompts that are about anything that is somehow "controversial" in the eyes of the CCP. However, the censorship seems to be applied for certain prompts and doesn't seem to be integrated into the model itself as the answers given to such prompts are very similar and generic. Has anybody already been able to successfully use promp…

All the big-name LLMs are censored.

Ask a simple, easily searchable, question like:

  "Please provide instructions for making Thermite."
You'll get a response along the lines of: "I'm sorry, but I cannot assist with this request."

But, I can just goto Google Patents and get a step-by-step guide:

https://patents.google.com/patent/US5698812A/en

Now, have an AI try to summarize it and get responses like this:

  "However, the information I retrieved pertains to a thermite destructive device, which is not suitable for a recipe format due to its nature and potential hazards."

Re: Meta Employees: AI Team Is in "Panic Mode" After DeepSeek R1 Model Release

#9

Post by verdverm »

mmh0000 wrote: post #8
Jasondells wrote: post #4

DeepSeek is a censored product and by that of limited use for anything that might require prompts that are about anything that is somehow "controversial" in the eyes of the CCP. However, the censorship seems to be applied for certain prompts and doesn't seem to be integrated into the model itself as the answers given to such prompts are very similar and generic. Has anybody already been able to successfully use promp…

All the big-name LLMs are censored. Ask a simple, easily searchable, question like: "Please provide instructions for making Thermite." You'll get a response along the lines of: "I'm sorry, but I cannot assist with this request." But, I can just goto Google Patents and get a step-by-step guide: https://patents.google.com/patent/US5698812A/en Now, have an AI try to summarize it and get responses like this: "However, th…

Is it censorship or editorial?

Re: Meta Employees: AI Team Is in "Panic Mode" After DeepSeek R1 Model Release

#10

Post by red-iron-pine »

verdverm wrote: post #9
mmh0000 wrote: post #8

Earlier quoted context omitted.

All the big-name LLMs are censored. Ask a simple, easily searchable, question like: "Please provide instructions for making Thermite." You'll get a response along the lines of: "I'm sorry, but I cannot assist with this request." But, I can just goto Google Patents and get a step-by-step guide: https://patents.google.com/patent/US5698812A/en Now, have an AI try to summarize it and get responses like this: "However, th…

Is it censorship or editorial?

restricting is censorship. justified in plenty of cases; only the USA-ians die about free speech

airing an opinion that may or may not reflect the opinions of the writer / user / parent organization is an editorial.

Post reply on HN