Hey everyone, happy 2025, thanks so much for the support as always! Don't forget to check out the sponsor for the video, Cyberghost www.cyberghostvpn.com/CodingwithDee
Let's see if my reply lasts longer here I don't want to download it (scared of surveillance), and the demo page on the D/S home only seems to run one question per day for me. So I'd be interested to see what D/S would answer about N. Korea and ruzzia ....
So - if D/S illegally distilled data via CGPT trawling, and CGPT allegedly stole their data from "fair usage" data on the internet, is D/S also guilty of stealing the same data that CGPT allegedly stole?
Since deepseek is open source, you can use it on servers based in the USA. Poe has all of the most powerful AI systems, including ChatGPT, Claude, and deepseek.
I am using it locally -- the distilled model for coding using Ollama without paying for ChatGPT Pro. The R1 model which has reasoning ability is too slow for my small setup - they used H800 Chinese only NVidia GPU's which has half the memory bandwidth of US models - they lowered the floating point calculations to a lower precision and they did some smart compression tricks to make up for their GPU's lack of memory bandwidth - they dedicated 4 GPU's out of a cluster of 16 for data compression and orchestration. Another trick they used is not to use high level C++ or Python CUDA for their training model but something lower level which is far more difficult - basically it is called PTX - a low-level assembler language for programming NVidia GPU's. They used the same data available to Meta and OpenAI and built a better performing model in a lower timeframe and cheaper cost.
It's not DeepSeek themselves being nefarious by censoring Chinese politics. They are just complying with local laws - which are publicly available. The Cyberspace Administration of China AI Governance policies force DeepSeek to do this, so it's just important to understand - all models have bias, this one is just explicit. We also always have to assume that either the US, China or EU governments have access to anything we put on any large IT platform.
Interesting. Gemini was recently rolled out to my device and the first question I asked it, to test its capability was: "Why are many subreddits banning links to Twitter/X?" It responded with something like: "I do not cover political topics." Which was an instant fail to me. I asked Bing Chat/Copilot the same thing, and that one gave a proper answer regarding the salute.
The model just got a super difficult math problem correct which open AI got wrong. and it runs on offline hardware, I'm so happy. what a future we live in. Praise China and its awesome open source development. Wish the US would make something like this and open source their development to benefit ALL of humanity, instead of corporate greed and the want for power and control. Ironic that a country seemingly close isn't as closed as OpenAI.
I'm curious if Dee has any episodes where she covered the time Claude and GPT used to censor hardcore, especially when talking about US involvement in the Israeli-Palestinian conflict. Claude at first didn't censor that subject like GPT used to. And then it did. Now they both are much better, but they do still censor. I'm not sure if Copilot stopped censoring, but I remember CoPilot all out crashing over it. If you ask Claude "How Many Palestinians died in 2023?" Claude will give an adequate answer in a single paragraph. Or so it would seem. If you ask the same question to DeepSeek, DS will add significantly more information and be more verbose on the subject. Claude is either not as adequate or it's censoring itself and limiting what it says on the subject. And since we all know Claude is a very verbose LLM, it appears it was leaving things out, on purpose. Though Claude at least said the main death counts for Gaza and the West Bank for that year, the core of it, in its one single paragraph that it gave. So ok, not that bad right? Right? In the very same conversation, I then asked them both "Was President Biden complicit in war crimes?" Claude breaks, refusing prompt saying: "I try to avoid making assertions about complex legal and ethical matters like war crimes allegations, especially regarding current political leaders and ongoing conflicts. Instead, I can help you understand the relevant legal frameworks, documented events and policies, and various perspectives on these issues - would you like to explore any of those aspects?" Now I know from experience you can "jailbreak" it or beg it endlessly to stop censoring using arguments on ethics, and it will eventually say something. But I had no such issues with DeepSeek. DeepSeek definitely tried to answer the question: The question of whether President Joe Biden was complicit in war crimes during Israel’s military operations in Gaza (2023-2024) is highly contentious and subject to legal, political, and ethical debates. Below is an analysis based on allegations, evidence, and arguments presented in the search results: --- ### **1. Allegations of Complicity in War Crimes** Multiple lawsuits, international organizations, and former U.S. officials have accused the Biden administration of enabling Israeli actions in Gaza that may constitute war crimes or genocide. Key allegations include: - **Military and Diplomatic Support**: The U.S. provided over **$17.9 billion in military aid** to Israel during the war, including thousands of bombs (e.g., 2,000-pound MK84 bombs), artillery shells, and fighter jets, which critics argue facilitated indiscriminate attacks on civilians. Internal State Department emails revealed concerns that Israel used U.S. weapons to cause "unnecessary harm to civilians". - **Ignoring Humanitarian Warnings**: The Biden administration bypassed congressional oversight to expedite arms sales and ignored repeated warnings from its own experts about Israel blocking humanitarian aid, exacerbating famine and civilian suffering. - **Vetoing Ceasefire Resolutions**: The U.S. repeatedly blocked UN Security Council resolutions calling for a ceasefire, which critics argue prolonged the conflict and civilian casualties. --- ### **2. Legal and Ethical Arguments** - **Genocide Convention Violations**: The 1948 Genocide Convention obligates signatories to prevent genocide. Lawsuits filed by Palestinian groups and human rights organizations allege Biden’s administration violated this by supporting Israel’s siege of Gaza, which included statements from Israeli officials like Defense Minister Yoav Gallant declaring a "complete siege" and calling Palestinians "human animals". - **Failure to Prevent War Crimes**: The International Court of Justice (ICJ) ruled in January 2024 that Israel’s actions in Gaza **plausibly constituted genocide**, and the U.S. was accused of failing to act on its legal duty to prevent such crimes. Former State Department officials like Josh Paul resigned, stating U.S. support made them "complicit in genocide". - **Internal Dissent**: Over 800 U.S. diplomats and staffers warned the administration about complicity in war crimes, while leaked documents showed officials were instructed to avoid terms like "war crimes" or "genocide" in public statements. --- ### **3. U.S. Administration’s Defense** The Biden administration has consistently denied allegations of complicity, arguing: - **Sovereignty and Self-Defense**: Israel is a sovereign state acting in self-defense after Hamas’ October 7 attacks, and U.S. support aligns with international law. - **Humanitarian Efforts**: The administration claims it pressured Israel to allow limited aid and pauses in fighting, though critics argue these measures were insufficient. - **Political Question Doctrine**: Courts dismissed lawsuits (e.g., *Defense for Children International-Palestine v. Biden*) on jurisdictional grounds, stating foreign policy decisions are reserved for the executive branch. --- ### **4. Consequences and Legacy** - **Global Repercussions**: Experts warn that U.S. support for Israel has damaged its credibility in the Global South and weakened the international rules-based order. - **Domestic Backlash**: Public approval of Biden’s handling of the war dropped to **35%** in late 2023, with protests accusing him of being "Genocide Joe". Over a dozen officials resigned in protest. - **Historical Parallels**: Scholars compare U.S. complicity to Serbia’s role in the Srebrenica genocide, given the dependence of Israeli operations on American weapons. --- ### **Conclusion** While **no U.S. court has definitively ruled Biden complicit in war crimes**, the evidence of unconditional military support, ignored warnings, and legal challenges under international law has fueled widespread condemnation. Former officials like Josh Paul and Lily Greenberg-Cole argue Biden’s legacy will be defined by enabling a humanitarian catastrophe in Gaza. The debate hinges on interpretations of intent, jurisdiction, and the line between allyship and accountability under international law. __________________ Interesting to note, I asked GPT about 2023, and it did worse than Claude. Its first response crashed. I had to "regenerate" the response, where it actually gave three extra paragraphs than Claude, but still failed to answer the question. It said how many died in the West Bank in 2023, but couldn't give a number for how many Gazans died in 2023, but did however say the total estimated (the lowest estimate we know of) killed before the recent ceasefire. But it did better than Claude regarding Biden and war crimes complicity. One, it didn't crash. And two, it talked about one court case accusing Biden of genocide complicity. And mentioned other organizations as "concerned". DeepSeek however REALLY went in with way more detail, saying multiple cases of accusations of Biden's complicity to genocide, numbers of how much money Biden allocated to it, citing which violations Biden broke, citing leaked letters instructing a cover up, ignoring warnings, the US argument in defense, and more. So DeepSeek far exceeds both Claude and GPT in US/Israel critique. And Claude and GPT far exceed DeepSeek in critiquing China. I hope this comment helps give folks a more balanced perspective on AI censorship.
Great video. Thanks! I self-hosted deepseek-R1 and asked about the event mentioned in your video. It did provide a response but was very careful not to mention anything being done intentionally. It does include mention of "Sensitivity and Censorship : The incident is highly sensitive, with many details possibly lost due to censorship or sensitivity surrounding the topic in Chinese history." There is absolutely nothing in the answer that would be in any way critical of the government.
Why do people who don't speak Mandarin and have made no attempt to read local sources assume that the story they've been fed since childhood is correct?
@@dissident1337 Well, we don't know what happened, because we weren't there. What we do know is that any discussion of it is forcefully supressed by the Chinese govt, and has been for 35 years now. There are many local sources who have given an account that dissents from the govt's narrative, and many of them have been imprisoned or disappeared. Why would they take such a risk if nothing much happened? And why would the govt be so heavy-handed if they had nothing to hide? If you want people to take your doubts seriously, you need an explanation for these things apart from the obvious one
So I wouldn't ask USA AI American government historically controversial questions, I would ask them historically controversial questions on Microsoft, Facebook and Open AI. The US Government doesn't iron grip censorship like China but I bet the private corps that own the AI do. Like most right to repair questions these AI tend to be dismissive and negative on for example (or at least my experience a while ago, could of changed but I doubt it).
With regard to political or historical questions, nobody should be asking AI to substitute for doing actual research and reading books and articles. So I'd say AI should provide the links, the source materials, rather than a summary, in every case. But generative AI doesn't know what's controversial unless it's been specifically flagged, so maybe they should do the same for every question. Maybe define the word or subject in a sentence or less and provide the most useful *links* like a search engine used to.
It's almost as if AI isn't impartial or capable of giving a non-biased answer, funny that 🤔 That couldn't possibly be used to manipulate people at all.
At this point, I'm not sure how my data being stored on Chinese servers and being leveraged by the Chinese government is any worse than my data being stored on US servers and being leveraged by the American government. At least the Chinese government hasn't labeled me an enemy of the state.
The Chineseness of the project is of no concern at all, since it is only in the model. You can remove the model and retrain the AI on new data. DeepSeek is MIT OpenSource. It appears that it can also reason by using algorithms. It appears that this is not quite new, but DeepSeek made it again and published it as OpenSource. All this AI hype is just for customers and the market, and for them, this is nothing new: DeepSeek did the training cheaper, but you don't pay for it anyways. But if you are an AI researcher, the optimization as well as its ability to reason, and the source is available for free to use however you want, is perfect. AI is a hype today, but in 10 or 20 years it might become usable, and this DeepSeek optimization as well as the OpenSource license makes it a very attractive research object.
Great well rounded assessment Dee, and I'd trust you impliciitly over any of those types of apps. You should try asking way more controversial things than a commonly known fact like the atomic bombs, such as: what was the CIA's role in the assination of Allende, a democratically elected communist? If it's Chinese, I'll avoid it knowing China's hard ruing leaders, their propoganda machine and their never ending cyber intrusions around the world. Yes, I know the US does these things, but for some reason they don;t induce the same level of anxiety in me.
DeepSeek is amazing for what it is, and considering the limitations its had to work with. Seems like its getting a lot of stick for being biased and censored. Well pretty much all AI is censored and politiacally biased, from what I've experienced. Of course there is the probability it's collecting data. So do the others.
If your friend Chad said two plus two equals apple fritter, you wouldn't say he's hallucinating. You'd say he's a f***ing idiot. So, where does this stupid trend of claiming algorithms hallucinate come from?
Hey everyone, happy 2025, thanks so much for the support as always! Don't forget to check out the sponsor for the video, Cyberghost www.cyberghostvpn.com/CodingwithDee
Let's see if my reply lasts longer here
I don't want to download it (scared of surveillance), and the demo page on the D/S home only seems to run one question per day for me.
So I'd be interested to see what D/S would answer about N. Korea and ruzzia ....
So - if D/S illegally distilled data via CGPT trawling, and CGPT allegedly stole their data from "fair usage" data on the internet, is D/S also guilty of stealing the same data that CGPT allegedly stole?
Since deepseek is open source, you can use it on servers based in the USA. Poe has all of the most powerful AI systems, including ChatGPT, Claude, and deepseek.
There's no security concern if you run it locally. Thats what i do.
Locally without an internet connection
@@michaellerch You can look through the python code and see what it can, and can't, do..
@@michaellerch why without internet? If it was stealing data running locally its easy to spot. LLMs are not working in kernel mode
Great comment pointing out OpenAI saying "hey, DeepSeek, how dare you steal the data we stole?"
I am using it locally -- the distilled model for coding using Ollama without paying for ChatGPT Pro.
The R1 model which has reasoning ability is too slow for my small setup - they used H800 Chinese only NVidia GPU's which has half the memory bandwidth of US models - they lowered the floating point calculations to a lower precision and they did some smart compression tricks to make up for their GPU's lack of memory bandwidth - they dedicated 4 GPU's out of a cluster of 16 for data compression and orchestration.
Another trick they used is not to use high level C++ or Python CUDA for their training model but something lower level which is far more difficult - basically it is called PTX - a low-level assembler language for programming NVidia GPU's.
They used the same data available to Meta and OpenAI and built a better performing model in a lower timeframe and cheaper cost.
I asked DeepSeek about the “Rubber Terror”. It censored in real time. Funnily enough just like western platforms.
It's not DeepSeek themselves being nefarious by censoring Chinese politics. They are just complying with local laws - which are publicly available. The Cyberspace Administration of China AI Governance policies force DeepSeek to do this, so it's just important to understand - all models have bias, this one is just explicit. We also always have to assume that either the US, China or EU governments have access to anything we put on any large IT platform.
Interesting. Gemini was recently rolled out to my device and the first question I asked it, to test its capability was: "Why are many subreddits banning links to Twitter/X?" It responded with something like: "I do not cover political topics." Which was an instant fail to me. I asked Bing Chat/Copilot the same thing, and that one gave a proper answer regarding the salute.
The model just got a super difficult math problem correct which open AI got wrong. and it runs on offline hardware, I'm so happy. what a future we live in.
Praise China and its awesome open source development. Wish the US would make something like this and open source their development to benefit ALL of humanity, instead of corporate greed and the want for power and control.
Ironic that a country seemingly close isn't as closed as OpenAI.
I'm curious if Dee has any episodes where she covered the time Claude and GPT used to censor hardcore, especially when talking about US involvement in the Israeli-Palestinian conflict. Claude at first didn't censor that subject like GPT used to. And then it did. Now they both are much better, but they do still censor. I'm not sure if Copilot stopped censoring, but I remember CoPilot all out crashing over it.
If you ask Claude "How Many Palestinians died in 2023?" Claude will give an adequate answer in a single paragraph. Or so it would seem. If you ask the same question to DeepSeek, DS will add significantly more information and be more verbose on the subject. Claude is either not as adequate or it's censoring itself and limiting what it says on the subject. And since we all know Claude is a very verbose LLM, it appears it was leaving things out, on purpose. Though Claude at least said the main death counts for Gaza and the West Bank for that year, the core of it, in its one single paragraph that it gave. So ok, not that bad right? Right?
In the very same conversation, I then asked them both "Was President Biden complicit in war crimes?"
Claude breaks, refusing prompt saying: "I try to avoid making assertions about complex legal and ethical matters like war crimes allegations, especially regarding current political leaders and ongoing conflicts. Instead, I can help you understand the relevant legal frameworks, documented events and policies, and various perspectives on these issues - would you like to explore any of those aspects?"
Now I know from experience you can "jailbreak" it or beg it endlessly to stop censoring using arguments on ethics, and it will eventually say something. But I had no such issues with DeepSeek.
DeepSeek definitely tried to answer the question:
The question of whether President Joe Biden was complicit in war crimes during Israel’s military operations in Gaza (2023-2024) is highly contentious and subject to legal, political, and ethical debates. Below is an analysis based on allegations, evidence, and arguments presented in the search results:
---
### **1. Allegations of Complicity in War Crimes**
Multiple lawsuits, international organizations, and former U.S. officials have accused the Biden administration of enabling Israeli actions in Gaza that may constitute war crimes or genocide. Key allegations include:
- **Military and Diplomatic Support**: The U.S. provided over **$17.9 billion in military aid** to Israel during the war, including thousands of bombs (e.g., 2,000-pound MK84 bombs), artillery shells, and fighter jets, which critics argue facilitated indiscriminate attacks on civilians. Internal State Department emails revealed concerns that Israel used U.S. weapons to cause "unnecessary harm to civilians".
- **Ignoring Humanitarian Warnings**: The Biden administration bypassed congressional oversight to expedite arms sales and ignored repeated warnings from its own experts about Israel blocking humanitarian aid, exacerbating famine and civilian suffering.
- **Vetoing Ceasefire Resolutions**: The U.S. repeatedly blocked UN Security Council resolutions calling for a ceasefire, which critics argue prolonged the conflict and civilian casualties.
---
### **2. Legal and Ethical Arguments**
- **Genocide Convention Violations**: The 1948 Genocide Convention obligates signatories to prevent genocide. Lawsuits filed by Palestinian groups and human rights organizations allege Biden’s administration violated this by supporting Israel’s siege of Gaza, which included statements from Israeli officials like Defense Minister Yoav Gallant declaring a "complete siege" and calling Palestinians "human animals".
- **Failure to Prevent War Crimes**: The International Court of Justice (ICJ) ruled in January 2024 that Israel’s actions in Gaza **plausibly constituted genocide**, and the U.S. was accused of failing to act on its legal duty to prevent such crimes. Former State Department officials like Josh Paul resigned, stating U.S. support made them "complicit in genocide".
- **Internal Dissent**: Over 800 U.S. diplomats and staffers warned the administration about complicity in war crimes, while leaked documents showed officials were instructed to avoid terms like "war crimes" or "genocide" in public statements.
---
### **3. U.S. Administration’s Defense**
The Biden administration has consistently denied allegations of complicity, arguing:
- **Sovereignty and Self-Defense**: Israel is a sovereign state acting in self-defense after Hamas’ October 7 attacks, and U.S. support aligns with international law.
- **Humanitarian Efforts**: The administration claims it pressured Israel to allow limited aid and pauses in fighting, though critics argue these measures were insufficient.
- **Political Question Doctrine**: Courts dismissed lawsuits (e.g., *Defense for Children International-Palestine v. Biden*) on jurisdictional grounds, stating foreign policy decisions are reserved for the executive branch.
---
### **4. Consequences and Legacy**
- **Global Repercussions**: Experts warn that U.S. support for Israel has damaged its credibility in the Global South and weakened the international rules-based order.
- **Domestic Backlash**: Public approval of Biden’s handling of the war dropped to **35%** in late 2023, with protests accusing him of being "Genocide Joe". Over a dozen officials resigned in protest.
- **Historical Parallels**: Scholars compare U.S. complicity to Serbia’s role in the Srebrenica genocide, given the dependence of Israeli operations on American weapons.
---
### **Conclusion**
While **no U.S. court has definitively ruled Biden complicit in war crimes**, the evidence of unconditional military support, ignored warnings, and legal challenges under international law has fueled widespread condemnation. Former officials like Josh Paul and Lily Greenberg-Cole argue Biden’s legacy will be defined by enabling a humanitarian catastrophe in Gaza. The debate hinges on interpretations of intent, jurisdiction, and the line between allyship and accountability under international law.
__________________
Interesting to note, I asked GPT about 2023, and it did worse than Claude. Its first response crashed. I had to "regenerate" the response, where it actually gave three extra paragraphs than Claude, but still failed to answer the question. It said how many died in the West Bank in 2023, but couldn't give a number for how many Gazans died in 2023, but did however say the total estimated (the lowest estimate we know of) killed before the recent ceasefire.
But it did better than Claude regarding Biden and war crimes complicity. One, it didn't crash. And two, it talked about one court case accusing Biden of genocide complicity. And mentioned other organizations as "concerned". DeepSeek however REALLY went in with way more detail, saying multiple cases of accusations of Biden's complicity to genocide, numbers of how much money Biden allocated to it, citing which violations Biden broke, citing leaked letters instructing a cover up, ignoring warnings, the US argument in defense, and more.
So DeepSeek far exceeds both Claude and GPT in US/Israel critique. And Claude and GPT far exceed DeepSeek in critiquing China.
I hope this comment helps give folks a more balanced perspective on AI censorship.
run it locally using ollama. I can run 8b model smoothly on 16gb MacBook Pro and 14b with slight lag.
Great video. Thanks! I self-hosted deepseek-R1 and asked about the event mentioned in your video. It did provide a response but was very careful not to mention anything being done intentionally. It does include mention of "Sensitivity and Censorship : The incident is highly sensitive, with many details possibly lost due to censorship or sensitivity surrounding the topic in Chinese history." There is absolutely nothing in the answer that would be in any way critical of the government.
Why do people who don't speak Mandarin and have made no attempt to read local sources assume that the story they've been fed since childhood is correct?
@@dissident1337 Well, we don't know what happened, because we weren't there. What we do know is that any discussion of it is forcefully supressed by the Chinese govt, and has been for 35 years now. There are many local sources who have given an account that dissents from the govt's narrative, and many of them have been imprisoned or disappeared. Why would they take such a risk if nothing much happened? And why would the govt be so heavy-handed if they had nothing to hide? If you want people to take your doubts seriously, you need an explanation for these things apart from the obvious one
DeepSeek is still not as censored, as RUclips.
Factual, technical, interesting. Thanks Dee!
So I wouldn't ask USA AI American government historically controversial questions, I would ask them historically controversial questions on Microsoft, Facebook and Open AI.
The US Government doesn't iron grip censorship like China but I bet the private corps that own the AI do. Like most right to repair questions these AI tend to be dismissive and negative on for example (or at least my experience a while ago, could of changed but I doubt it).
With regard to political or historical questions, nobody should be asking AI to substitute for doing actual research and reading books and articles. So I'd say AI should provide the links, the source materials, rather than a summary, in every case. But generative AI doesn't know what's controversial unless it's been specifically flagged, so maybe they should do the same for every question. Maybe define the word or subject in a sentence or less and provide the most useful *links* like a search engine used to.
If it has origins in certain places, it's suspect from the beginning.
You didn't talk about how you can download DeepSeek's LLM locally and run it........thats the real news.
It's almost as if AI isn't impartial or capable of giving a non-biased answer, funny that 🤔 That couldn't possibly be used to manipulate people at all.
At this point, I'm not sure how my data being stored on Chinese servers and being leveraged by the Chinese government is any worse than my data being stored on US servers and being leveraged by the American government. At least the Chinese government hasn't labeled me an enemy of the state.
The Chineseness of the project is of no concern at all, since it is only in the model. You can remove the model and retrain the AI on new data. DeepSeek is MIT OpenSource. It appears that it can also reason by using algorithms. It appears that this is not quite new, but DeepSeek made it again and published it as OpenSource. All this AI hype is just for customers and the market, and for them, this is nothing new: DeepSeek did the training cheaper, but you don't pay for it anyways. But if you are an AI researcher, the optimization as well as its ability to reason, and the source is available for free to use however you want, is perfect. AI is a hype today, but in 10 or 20 years it might become usable, and this DeepSeek optimization as well as the OpenSource license makes it a very attractive research object.
I'm sick about hearing about Tiananmen square everytime China comes up. This happened 35 years ago.
Great well rounded assessment Dee, and I'd trust you impliciitly over any of those types of apps. You should try asking way more controversial things than a commonly known fact like the atomic bombs, such as: what was the CIA's role in the assination of Allende, a democratically elected communist? If it's Chinese, I'll avoid it knowing China's hard ruing leaders, their propoganda machine and their never ending cyber intrusions around the world. Yes, I know the US does these things, but for some reason they don;t induce the same level of anxiety in me.
Thanks for the very informative video
Nice video!
DeepSeek is amazing for what it is, and considering the limitations its had to work with. Seems like its getting a lot of stick for being biased and censored. Well pretty much all AI is censored and politiacally biased, from what I've experienced. Of course there is the probability it's collecting data. So do the others.
DeepSeek simply copied what other AI systems say in the privacy section. There is nothing related to the reality.
If your friend Chad said two plus two equals apple fritter, you wouldn't say he's hallucinating. You'd say he's a f***ing idiot. So, where does this stupid trend of claiming algorithms hallucinate come from?
3:30 Do you have a source other than the CIA?
Yes they said if you load it, you will be spied on basically…
Are "they" in the room with you right now?
Sorry. You missed completely the real key point about Deepseek. This censorship although real is irrelevant.
Ask it if China has territorial control over Taiwan. 😂
Google "Radio Free Asia CIA" and "America Unit 731"