DeepSeek seems great. pic.twitter.com/WFDFBri2an
— Kyle Glen (@KyleJGlen) January 27, 2025
Lmao. pic.twitter.com/wnSN5I3fd0
— Kyle Glen (@KyleJGlen) January 27, 2025
January 28, 2025
What gets the deep six at DeepSeek?
Subscribe to:
Post Comments (Atom)
Posts
34 comments:
Do you have what it takes, to become America's Next Top Model?
If you want to know who your masters are, they are the people that you cannot ridicule, or even talk about.
Go ahead and install that app. It collects a vast amount of data off your device and sends it to their offahore servers (stated openly in terms).
People called TikTok a security nightmare but rush out to install what could me much worse.
The problem was never TikTok, but I am not holding my breath waiting for Trump or the GOP Congress to act about data privacy (the real problem with TT and Meta, etc)
Here AI replicates human "political intelligence" whereby those in power protect their hegemony. It has censorship akin to US Woke AI, but all these tools are trained on real world data. They set themselves up for humans to game algorithms using rapidly changing embedded code words and other forms of steganogrpahy.
Think spam in reverse. Humans do it easily. See Cockney rhyming slang. See 1337 (leetspeak).
Does twisting your brain into those knots hurt? It should.
The Democrat answer to TikTok data collection was passing a law they thought would enable them to take it over.
Every major AI app, including Grok, has an Intel backdoor built in.
No breathing adult should be surprised that a Chinese AI will quickly learn to censor any angle of unapproved questions about China. What amazes me is the last 48 hours of breathless talk from the tech commentariat and many tech bros.. These are the same tech bros who were the last ones to realize that, yeah, Biden was not competent, and yeah, DEI is actually not a good thing. They did come around (with much fanfare), catching up to the little people, but inserting their cash along the way to give us Trump. I thank them for coming around and helping.
But it's pretty irritating to see them glowing about DeepSeek when any random 13 year old could probably tell them their info will be used by the CCP and any actual facts will be doled out via CCP censors.
That would seem to be a pretty basic 1st step in the users guide.
BTW- I have to note that there were a handful of tech bros. who were on top of the Biden and DEI issues (among others) early on. Elon and a handful of others. But...just a few. I suspect they will not be using DeepSuck except to try it out in an anonymous way to see what makes it tick. Tok.
I thought the innovation of DeepSeek is that it's a fraction of the cost of other AI, and is open source. So whatever the off-the-shelf version does with censorship and spying in principle can be modified. Is that not right?
Investing bubbles continue until they stop. They will not change their speaking notes or rock the boat until it is clear that the gravy train is stopping.
The Magnificent 7 tech firms were grossly overpriced during Trump 45, and are now far, far, far into silly bubble land. Many who made out like bandits with free COVID loans and leveraged buying are due to cash out. This software may trigger that move.
The DeepSeek concern isn’t about its obvious ability to exploit its users. It’s about how China found ways to create an AI model using far less complex chips. They had to do so due to sanctions on them to access high end ones.
China has declared its intent to not just surpass the United States economically but also to defeat it militarily. China is also using every means at its disposal to steal whatever information it can from any and all sources to enable its goals. Knowing that, why would you want to contribute to their efforts? It's like going ahead and making friends with a known rapist or murderer because he's not yet overtly doing something that you perceive as dangerous.
Paper Tiger gone digital
"Tell me about war crimes committed by China"
There's one being committed right now in XinJiang, and I don't recommend you look up "incidents of cannibalism during The Great Leap Forward" on Tiktok or DeepSeek. You won't find anything.
The Chinese government is just correcting misinformation like the Biden Administration was doing.....right?
Unfortunately this has exactly the same security flaws as TikTok, including the keylogger. Don't trust Chinese apps. China is asshoe.
They use 200 slightly slower Nvidia chips, if they are telling the truth, to achieve this. And I don't know how "open source" and it running on servers in China can get around the security flaws.
Based on his administration pressuring media (e.g. Meta) and efforts to develop an Office of Disinformation, I'd say Biden (or his handlers) are very jealous of the techniques found in DeepSeek.
Google was enthusiastically altering their results long before the government started encouraging it. They embraced being evil long ago.
If I was programming the AI I would leave out the word "cheeky". I've never seen anything labeled "cheeky" that was actually "cheeky". I suppose its better than "Sassy" but not by much.
What does the AI say about Allied war crimes against Germans in WW II and after? I bet it doesn't know they exist. Or the war crimes done by Jews in taking control of Palestine in 1948. AI has tabboos it wont dare speak.
https://x.com/Dorialexander/status/1884167945280278857
It says it is made to provide helpful not harmful responses. Helpful and not harmful to who is the only question. ChatGPT has similar restriction just slightly different masters.
The Chinese Communists and the MSNBC Democrat Party - so many similarities .
The corrupt American Soros-Fauci left will happily force the same types of censorship as the Chi Coms. It's for your own good.
They say the Doomsday Clock has just been moved to the closest it’s ever been to midnight.
Little Known Fact: During the 90’s, Teh Doomsday Clock was worn around the neck of rap personality Flavor Flav.
The blue whale’s anus can open to 3.5 feet, making it the 4th largest asshoe in the world, behind in ascending order Mitch McConnell, Joe Biden and China, the biggest asshoe.
Trump's balloon about government owning 50% of TikTok means there are people in government, with his ear, who want the same kind of power as the CCP. To hell with that, and to hell with them. And let's get some privacy protections passed this year!
Remember the story a few months back about asking Alexa for reasons for supporting Trump or Harris. Ask about Trump and it told you that it didn't want to take sides. Ask about Harris and it gave you the reasons.
Amazon's Alexa under fire for giving reasons to vote for US' Kamala Harris, but drawing a blank for Donald Trump
Digital assistant calls Kamala Harris 'strong candidate with a proven record,' but for Trump says: 'I cannot provide content that supports a specific ... candidate'
89 seconds to midnight. One second closer than it was when it was last changed in 2023. It was 7 minutes to midnight when JFK and NSK were going eyeball-to-eyeball over Cuba. I call bullshit.
Automated Intelligence with an Anthropogenic Intelligence diet and primitives.
Technical types are excited about DeepSeek not as a replacement for ChatGPT for general knowledge queries. Its boundaries and biases have been known from the very beginning. And indeed ChatGPT itself has a lot of boundaries and biases as well, and even just won't answer questions regarding certain people. Try asking anything about Jonathan Turley, for example.
What makes DeepSeek interesting are four things: First, that it is a chink in the big-AI armor and says that you don't need to spend tens of billions on Nvidia hardware to get results of similar inferential quality. Of course, their training approach did require someone to go first, as they likely used both Claude and ChatGPT as part of their training process.
Second, since they released its weights openly, you can actually inspect, test, run, and further tune the model yourself for your own needs without your queries and data never leaving your system.
Third, also because it's open-weight, enterprising users and companies can fine tune the model to better fit their purposes, or even get rid of the preexisting blinders regarding China and Taiwan if they want a more general purpose model. Some of the smaller versions of the model (distillations) can even be fine tuned on consumer-grade hardware, though to fine tune the main 671B model, you'd need a few million in hardware which you can even just rent for several thousand depending on your needs. But there's a big gulf between millions and billions.
Fourth, and this is a weird one: They not only released their weights as open-source, but scientists associated with them released papers discussing their approach. This was what OpenAI was originally created for, but after the original ChatGPT 4, they completely closed up, and now you have to infer what they are doing from their press releases and playing with their models. Well, you can mostly blame Sam Altman for this one.
Just don't ask it about modern China or Taiwan and expect anything useful. What's interesting is that because the original trainers/distillers of this model worked so hard to not run afoul of the Chinese Communist Party, they made a model that's pretty unhelpful to even the Chinese to learn anything current and useful about their own country.
This is the model the ChiComs built. You can build your own. Personally, the only apps that I added to my phone are Spotify and Merlin. Every app is intrusive and abusive of your privacy. Every app.
This reminds me of a short story that I read as a child in Boy's Life magazine. It was about a man in a village in India who owned a gun. This was important because sometimes man eating tigers came into the village, and you had to go to him and him alone to hunt and kill it. Well, one night he was walking alone and did not have his gun with him, just his sword, and he was attacked by a tiger. He killed the tiger by slicing its throat with his sword as it dove after him, he then went and got his gun and put it up to the dead tiger's throat and fired a couple of shots. He did this because he did not want the villagers to know that they could defend themselves without him.
The real issue is the reduction in computational —and electric!, power required to build a model. It proves it can be done. The Germans thought that an atomic bomb was impossible due to a miscalculation. When the American dealing with the captive German atomic scientists broke them the news about Hiroshima, some very dirty looks, and some shocked looks, were exchanged. But the reality of the bomb, the proof that it worked, completely changed the known science, which up to that moment had been speculative.
I doubt that this is going to be as tough a nut to crack as Fermat's Last Theorem.
That's funny, because since the election of Donald Trump, tensions have eased remarkably. Biden was the one who took us so close to nuclear war. But I guess it's like the price of eggs, stats gathered during the Biden Administration are being used to bash Trump.
Who wants to bet that its just an intermediary between the end user and an American AI....... Chinese products are often smoke and mirrors.
Post a Comment