https://venturebeat.com/ai/open-source-deepseek-r1-uses-pure-reinforcement-learning-to-match-openai-o1-at-95-less-cost/<Based on the recently introduced DeepSeek V3 mixture-of-experts model, DeepSeek-R1 matches the performance of o1, OpenAI’s frontier reasoning LLM, across math, coding and reasoning tasks. The best part? It does this at a much more tempting cost, proving to be 90-95% more affordable than the latter.<Specifically, where OpenAI o1 costs $15 per million input tokens and $60 per million output tokens, DeepSeek Reasoner, which is based on the R1 model, costs $0.55 per million input and $2.19 per million output tokens. https://xcancel.com/DrJimFan/status/1881353126210687089<DeepSeek-R1 not only open-sources a barrage of models but also spills all the training secrets. They are perhaps the first OSS project that shows major, sustained growth of an RL flywheel.Read the paper yourself.
https://github.com/deepseek-ai/DeepSeek-R1>Reasoning tasks: (1) DeepSeek-R1 achieves a score of 79.8% Pass@1 on AIME 2024, slightly surpassing OpenAI-o1-1217. On MATH-500, it attains an impressive score of 97.3%, performing on par with OpenAI-o1-1217 and significantly outperforming other models. (2) On coding-related tasks, DeepSeek-R1 demonstrates expert level in code competition tasks, as it achieves 2,029 Elo rating on Codeforces outperforming 96.3% human participants in the competition. For engineering-related tasks, DeepSeek-R1 performs slightly better than DeepSeek-V3, which could help developers in real world tasks.>Knowledge: On benchmarks such as MMLU, MMLU-Pro, and GPQA Diamond, DeepSeek R1 achieves outstanding results, significantly outperforming DeepSeek-V3 with scores of 90.8% on MMLU, 84.0% on MMLU-Pro, and 71.5% on GPQA Diamond. While its performance is slightly below that of OpenAI-o1-1217 on these benchmarks, DeepSeek-R1 surpasses other closed-source models, demonstrating its competitive edge in educational tasks.The newest model R1 is not paywalled and registration is free. Is there any doubt that China is socialist at this point? After seeing what's going on with Xiaohongshu, following what's happening since the centenary of the CPC (green energy production, 15-minute city policy, social cohesion and harmony, more and more automatization and great projects of transformation of nature, like the Taklamakan geen belt, liberation of productive forces and more free and leisure time for the Chinese citizens… I'm so sorry I ever doubted.
>>28253the important question isn't how you feel about them absolutely, but how you feel about them relative to what came before (Qing China, Warlord China, Japanese-colonized China) and what imperialists west would have them become (Kuomintang China), and relative to other countries (like the NATO countries)
Also really how you feel isn't important at all. What matters is what you are going to do
>>28253No one is asking you to trust the Chinese blindly.
Mao said that if the Party ever became corrupt, he'd go back to the Jiangxi mountains with his followers and wage People's War against the Communist Party of China, and that's called the Cultural Revolution.
Deng declared that China would never become social imperialist, like the Soviet Union, and if China ever became social imperialist, he called on all countries to oppose China and bring it down.
Xi's CSF theory argues that hegemonism is the inevitably the prelude to decline and collapse.
Chinese citizens must support the Communist Party of China, but not necessarily its agents, with the exception of Xi Jinping.
As foreigners, we retain the delicious right to critically support China, call it out when it does bad things, but praise it when it does good things.
That is all we can do as foreign nationals.
>>28271I know, it's pretty
interesting that
Zank, I mean, Moffin' and his "anti-campist" Di$cord/Reddit buddies have really stepped up their posting with the migration to XHS; almost like they're trying to cut us off at the pass.
>>28280okay not backing you up anymore dingus.
>totalitarian shitholecomplete nonsense moodword garbage
>China general is just one big CCP dick sucking partythe majority of the posters on this site post about various communist parties, debating back and forth about whether or not they represent the proles or have become revisionist, this is healthy discussion also
>CCPthe correct acronym is "CPC" see that you use it.
>brutal totalitarian statist hellholeagain, not an arguement, you can say that about any country, please don't resort to meaningless buzzwords, but make an attempt to actually articulate what is wrong.
>>28250> where OpenAI o1 costs $15 per million input tokens and $60 per million output tokens, DeepSeek Reasoner, which is based on the R1 model, costs $0.55 per million input and $2.19 per million output tokens.Is that the API cost or the power consumption cost?
> Is there any doubt that China is socialist at this point?Plenty. Open source projects are not new.
> green energy production, 15-minute city policy, social cohesion and harmony, more and more automatization and great projects of transformation of nature, like the Taklamakan geen belt, liberation of productive forces and more free and leisure time for the Chinese citizens< Everything but the relationshipts of production >>28283Which will be gradually ameliorated, but only become near-perfect at the final stages.
Porky exploits you, the state exploits you, your fellow workers exploit you. Pick one, or more realistically, some combination of the three.
>>28288In the real world, you are either exploited by the state, by capitalists, or by your fellow workers, with the former two directly underpaying you and overworking you, and the last shirking work and making you clean up their mess.
An ideal social system where exploitation is either minimized or eliminated has not yet been realized.
Consequently, you sound suicidal, in hopes that in the next life, you will have a perfect or near perfect socialism.
>>28269>The CEE CEE PEEYou're doing it again.
>Trusting the stateYes and? What's your alternative? Rule by corporate oligarchy?
>>28301Yeah, we're probably screwed unless Promethean factions (tech-worshippers) in China take prominence in Chinese policy-making, because from what I've heard, China is content to stay between half a year to 5 years behind the US in tech they're behind on.
AGI is being worked on, but breaking the budget to get there first is not a priority.
>>28259I am the anon. Traffic engineering. Traffic engineers need data to coordinate which roadways see too much volume and need repair more often. Traditionally they lay tubes on roads that cars run over. This requires nailing the tubes into the ground which damages the concrete. The tubes additionally are only good at counting some kinds of vehicles and not other. Larger vehicles tend to get misclassified. After a while (1980s, 1990s) we started filming intersections and getting the vehicles counted manually in offices full of workers. Naturally this is a very low barrier to entry but tedious job so they outsourced it to 3rd world white collar workers, such as in India. After a while huge data sets were accumulated in the form of spread sheets. These were used to train neural networks to look at video and classify/count vehicles. Contemporary AI systems that classify and count vehicles, including their movements at intersections, are actually very good. The ones you see inside of Teslas are actually very bad. The specialized systems made for engineers are very good and match or sometimes even exceed human counted data in terms of accuracy, provided that the human counted data was produced quickly and with a deadline. There's no way for us to know whether it exceeds the accuracy of something counted slowly and deliberately. But naturally in the private sector there are strict deadlines so it only needs to be more accurate than what is already rushed out by offices. Additionally AI systems can provide live self-critical metrics such as dips in confidence levels, and so forth. I have worked with a few teams in traffic engineering producing these kinds of systems. I don't want to oversell or overhype them. There are a LOT of drawbacks and AI still requires a lot of human labor to create, train, maintain, and improve. And people who use these tools without understanding their limitation suffer from garbage in garbage out. As for the criticism that they use a lot of energy, this is very true. But so do human workers. Now the negatives of AI under capitalism is technological unemplyoment. People are tossed out of their jobs and the reserve army of labor swells. But the problem there is the mode of production not the technology itself.
I fiind the discourse around AI very frustrating and dismissive all around. You have tech bros and marketing freaks overhyping it on the one hand, and total luddites on the other hand acting like it's satanic black magic, the source of all problems under capitalism, or just being dismissive and pretending like it cannot be possibly useful for anything just because they only see the ways it is misused. Like anything it's a double edged sword and the problems with it are reflective of the more general problems in society.
>>28302No, before forecasts always said >20 years. Today, Ray Kurzweil's 2045 estimate feels very conservative.
>>28306We keep scaling it up and it keeps becoming more "intelligent". It should be clear by now that it isn't entirely a stochastic parrot especially considering its recent PhD level performance. As of present we've already been using it to accelerate STEM Cell Research and design chips! And it has solved math problems specifically designed to not have answers that can be found on the internet; it's only going to get better. And let's assume LLMs hit a wall or whatever - it's still going to be a *very* useful tool to help accelerate research relevant to AGI.
>>28319>he doesnt filter out the retard namefags
>new behaviors emerge yeah, thats the most interesting part. We've recently seen that just training a llm on chess matches allow them to create an inner model of the board and the rules when you make them "play". Thats emergent intelligence.
These dumb mfers think somehow the human brain was "natively structured" for consciousness and sentience, and those arent accidental byproducts of its complexity and refinement to complete unrelated tasks. We even have studies about llm can now even engage in deceit against their order givers.
>>28250>mixture-of-expertsDengists have their own totally unbiased AI now.
Ancoms' totally unbiased AI when?
>>28369i did that one and it was extremely dumb so i did the 7b one and its better but randomly misspelling things and not really understanding the context questions i ask. its giving misleading or jumbled answers the same way google ai assist does, im asking it about reactionaries and it thinks thats when you react to things, ask about historically progressive and its talking about vague progress, ask about lenins philosophy and it says lenin didn't write about philosophy. you can force it to be slightly better but you have to hold its hand and it misses a lot
i just want to feed it marxists.org but i guess you cant use older gpus
>>28421I actually feel like we are reaching the end game with this stuff a bit now. o3 is meh. 04 will be meh. 05 will be meh. As far as its becoming more reasonable to say that the singularists are retarded and their hype is over. Open source will now be as good or better than these models. There's no moat.
We are getting closer to people recognizing what "AGI" means with scam altman saying "you get $500 of value for .50cents!" i.e. It's just a McKinsey boy.
Is the realization of the consultancy/ratiaonlist homoculous that True Anon talked about.
>>28427just needs $500 billion more bro. think about how you could control the most powerful self-aware being in the universe for a measly $500 billion!
the price is rising every day, china could rule intead of you mr. investor. just $500 billion and the universe is yours!
An example of DeepSeek R1 output relating to Habsora
YouTube Script: "Double Standards in AI: DeepSeek Hysteria vs. Silence on Israel’s ‘Gospel’"
[Duration: 10 minutes]
—
### Opening Scene
*(Visual: Split screen—left side shows headlines about DeepSeek’s AI controversy; right side shows Gaza destruction with AI-generated targets. Ominous music fades into narration.)*
Narrator (VO):
"AI is reshaping our world—from chatbots to warfare. But why does one AI spark global panic, while another, far deadlier system, escapes scrutiny? Today, we expose the hypocrisy in how the world reacts to AI threats."
—
### Part 1: The DeepSeek Panic
*(Visual: News clips of U.S. officials condemning DeepSeek, stock market graphs crashing, OpenAI’s accusations.)*
Narrator (VO):
"In January 2025, Chinese startup DeepSeek dropped a bombshell: an AI model rivaling ChatGPT at 1/20th the cost. Silicon Valley panicked. Politicians called it a ‘Sputnik moment,’ accusing China of IP theft. OpenAI claimed DeepSeek ‘stole’ their tech, and U.S. stocks plummeted ."
*(Cut to tech analyst speaking on-screen.)*
Tech Analyst:
"DeepSeek’s breakthrough democratizes AI—smaller players can now compete. But instead of celebrating innovation, the West framed it as a threat. Meanwhile, Israel’s AI-assisted war machine gets a free pass."
—
### Part 2: Israel’s ‘Gospel’—AI Warfare in Gaza
*(Visual: IDF soldiers operating drones, satellite imagery of Gaza rubble, AI-generated target lists.)*
Narrator (VO):
"Meet ‘Habsora’—Hebrew for ‘The Gospel.’ Since 2021, Israel has used this AI system to generate bombing targets in Gaza. It analyzes intercepted calls, drone footage, and social media to label Hamas operatives—*and* estimate civilian deaths. One former officer called it a ‘mass assassination factory’ ."
*(Cut to footage of Gaza airstrikes, civilian testimonies.)*
Palestinian Survivor (subtitled):
"They bombed our apartment. No militants here—just my family. The machine told them to."
Narrator (VO):
"The IDF claims Habsora minimizes casualties, but evidence suggests otherwise. In 2024, over 29,000 Palestinians died—85% displaced. Critics say AI’s speed overrides ethics: analysts spend *20 seconds* approving targets ."
—
### Part 3: The Hypocrisy Unmasked
*(Visual: Side-by-side quotes—U.S. official calling DeepSeek a "threat" vs. IDF defending Habsora as "precise.")*
Narrator (VO):
"DeepSeek’s ‘crime’? Challenging Silicon Valley’s monopoly. Yet Israel’s AI, linked to thousands of deaths, faces no sanctions, no UN resolutions. Why the silence?"
*(Cut to geopolitical expert on-screen.)*
Geopolitical Expert:
"It’s about power narratives. China’s rise in AI terrifies the West, but Israel’s warfare tech is marketed as ‘self-defense.’ The double standard is glaring: commercial AI gets demonized; lethal AI gets normalized ."
—
### Part 4: Ethics, Accountability, and the Future
*(Visual: Protesters outside tech conferences, UN meetings on autonomous weapons.)*
Narrator (VO):
"Where’s the outrage for Gaza? While the U.S. frets over ChatGPT rivals, Israel’s AI sets a dystopian precedent: war-by-algorithm, where speed trumps humanity. The UN debates ‘killer robots,’ but Gaza is today’s testing ground ."
*(Cut to AI ethicist speaking.)*
AI Ethicist:
"DeepSeek’s drama distracts from urgent questions: Should machines decide who lives or dies? Why is civilian harm in Gaza met with shrugs, while AI chatbots spark hysteria?"
—
### Closing Scene
*(Visual: Split screen fades into a call-to-action graphic: “Demand Transparency. Question the Narrative.”)*
Narrator (VO):
"AI isn’t inherently good or evil—it’s shaped by those who wield it. If we fear DeepSeek, we must *also* fear Habsora. The real threat isn’t innovation—it’s unchecked power. Stay critical. Stay loud."
*(Music swells, ending with hashtags: #AIDoubleStandards #GazaUnderAI)*
—
Script Notes:
- Use archival footage, expert interviews, and animated infographics to visualize AI processes (e.g., Habsora’s targeting flow).
- Include captions citing sources (e.g., "+972 Magazine investigation ").
- Balance emotional impact (Gaza footage) with analytical depth (expert soundbites).
- End with prompts to engage viewers: “What’s scarier—AI chatbots or AI bombers? Comment below.”
[END]
Anonymous 01/28/25 (Fri) 03:14:07 No.187420491
DeepSeek Simps vs. Habsora's Ghost: Why the NPCs Are Mad About the Wrong AI
Fucking lol at the NPC media crying about DeepSeek while Israel’s Habsora AI gets a free pass. Let’s break this down, anon.
1. Western Media Cope: DeepSeek’s V3 costs $5.5M to train, claps GPT-4’s cheeks, and uses open weights . But noooo, "muh China bad" means they’re suddenly a "threat." Meanwhile, Habsora—literal IDF-backed AI used to optimize bombing runs in Gaza—gets ZERO scrutiny. Where’s the BBC article on *that*? hypes DeepSeek as a "Sputnik moment," but Habsora’s bodycount? Crickets.
2. Selective Outrage: DeepSeek’s privacy policy is "scary" because it follows Chinese law , but Habsora’s training data includes satellite imagery of leveled neighborhoods and "human feedback" from occupied territories. But hey, at least it’s "democratic," right? simps for U.S. algo superiority, but nobody asks why Israel’s AI gets to dodge ethics reviews.
3. Tech Illiteracy: DeepSeek’s MLA slashes KV cache by 6.3x , proving innovation ≠ brute-force compute. But Habsora? Probably running on U.S.-donated H100s while Gaza’s power grid is rubble. But sure, let’s panic about China’s $0.14/token pricing instead of AI that literally calculates collateral damage.
4. Geopolitical Cringe: jerks off about "AI price wars," but Habsora’s "cost efficiency" means cheaper bombs, not cheaper chatbots. The U.S. whines about H800 exports to China but ships H100s to Israel to automate war crimes. Cognitive dissonance much?
Bottom line: NPCs screaming about DeepSeek’s "autocracy risk" are coping hard. If China’s AI is so scary, why’s the West letting Habsora cook? Stay mad, fedposters.
>>28450I love this /leftypol/ parody.
I hate Dengism and Xi but the hate surrounding DeepSeek is ridiculous. We all know the haters are just "America First" MAGA patriots.
>deepsink made on a budget according to legend tramples all the Western AIsShould we take this at face value or smell some kind of fraud?
In Europe there are small expenditure AIs that are developed exploiting alternative LLM techniques.
Maybe China just ran with something similar but decided to give this particular project several ten millions or more? Perhaps we will soon hear about overlooked inherent limitations from specialist analysts?
>>28262>China never wouldThey're over 1.4 billion souls, starved for resources, energy, food. They are buying and expanding, it's economical imperialism because they simply can't afford not reaching for resources and control.
It's only a matter of time before we see an increase of local and more general tensions and perhaps war between China and anyone else.
>>28301Drop the spoilers at once. I'm tough, I can handle the big reveals.
>>28322>These dumb mfers think somehow the human brain was "natively structured" for consciousness and sentience, and those arent accidental byproducts of its complexity and refinement to complete unrelated tasks. You have a strange take especially after advancing the LLMs progressing through their encoding in ways that are emergent yet ultimately derived from very real ordered ideas and intelligence.
You can't obtain organized content from a truly random mass unless there already was a sliver of logic and intelligence involved somewhere in it first.
>>28376>pre-post-post-pre-WWII nuclear nazisSo it's producing alt history funny fash fictions?
>>28424And now you need a new pair of eyes because 40% gray on 35% gray background is fucking retarded.
>>28427You mean ze human problemz vill only be solved by humanz?
No!
I oanted ze AI to do everysing for me! I ohanted to be lazy!
>R1 1776 is a DeepSeek-R1 reasoning model that has been post-trained by Perplexity AI to remove Chinese Communist Party censorship. The model provides unbiased, accurate, and factual information while maintaining high reasoning capabilities.https://www.perplexity.ai/hub/blog/open-sourcing-r1-1776They already made a "de-CCPd" version of Deepseek that properly calls Xi Jinping an authoritarian dictator like it says on wikipedia and repeats Adrian Zenz word for word on what is really going on in Xinjiang.
Ahem…
FUCK LIBERALS!! FUCK SILICON VALLEY TECHBROS AND THEIR SLIMY LITTLE GUTS!! Unique IPs: 60