[ home / rules / faq ] [ overboard / sfw / alt ] [ leftypol / edu / labor / siberia / lgbt / latam / hobby / tech / games / anime / music / draw / AKM ] [ meta ] [ wiki / shop / tv / tiktok / twitter / patreon ] [ GET / ref / marx / booru ]

/tech/ - Technology

"Technology reveals the active relation of man to nature" - Karl Marx
Name
Options
Subject
Comment
Flag
File
Embed
Password(For file deletion.)

Not reporting is bourgeois


File: 1737563306130-0.png (180.97 KB, 972x633, ClipboardImage.png)

File: 1737563306130-1.png (541.74 KB, 1173x967, ClipboardImage.png)

File: 1737563306130-2.png (1.64 MB, 1500x869, ClipboardImage.png)

 

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.
67 posts and 5 image replies omitted.

AI is a nothing burger and it's just a parrot, AI can't really think in a original way

File: 1737762428911-0.png (109.61 KB, 900x561, IMG_7347.png)

File: 1737762428911-1.jpeg (116.26 KB, 1400x583, IMG_7348.jpeg)

>>28306
As these things scale, new behaviors emerge that represent significant steps forward in intelligence. We’re frankly moving much faster than we thought. Exponentially more progress has been made in the last 3 years than in several decades prior.

>>28302
>nothing ever happens

File: 1737762861274.webp (185.68 KB, 1500x2000, GettyImages-596947870.webp)

>>28315
kurzie was so right he wrapped around to being wrong (too conservative)

>>28318
Humans can't either. Just like AI we have is a predictive algorithm based on past experience with the ability to modify itself, so is human mind a predictive algorithm based on past experience with the ability to modify itself

>>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.

Will OpenAI go bankrupt now? Jailbreaken versions of this will probably crater zChatGPT's usage.

>>28329
Hope so, Trump just signed some bills to protect them, aka pump money into them, but an ovestuffed pinata is still a pinata.

>>28330
if oepnai loses most of its marketshare they will just burn through that crash in a few years can't wiat tell someone jailbreaks this


I hope you guys are all using the chad deepseek vs the virgin openai

big if true

>how large would a door need to be for an average adult capybara?
To determine the appropriate size for a door for an average adult capybara, we need to consider the typical dimensions of a capybara and ensure the door is large enough for them to pass through comfortably.

### Average Size of an Adult Capybara:
- **Height (at the shoulder):** 50-60 cm (20-24 inches)
- **Length (body):** 100-130 cm (39-51 inches)
- **Weight:** 35-65 kg (77-143 lbs)

### Recommended Door Size:
- **Height:** The door should be at least **70-80 cm (28-31 inches)** tall to allow the capybara to pass through without ducking.
- **Width:** The door should be at least **60-70 cm (24-28 inches)** wide to accommodate the capybara's body comfortably.

### Additional Considerations:
- If the capybara is carrying something (like food or bedding), you may want to increase the width slightly.
- If the capybara is particularly large or if you want to ensure extra comfort, you can make the door slightly larger than the minimum dimensions.

### Example Door Size:
- **Height:** 80 cm (31 inches)
- **Width:** 70 cm (28 inches)

This size should comfortably accommodate an average adult capybara.

>>28272
the principle contradiction in all imperialist matters is the relation between the production distribution and consumption on one hand and state restrictions that effect my gooning(the other hand)

>>28348
>Open Source
Is frontend Open Source too? Can you use the chatbot without JavaScript?

ngl anons, this exchange gave me chills after reading about the CEO's description of R1 as beautiful and poetic or whatever he said.
https://xcancel.com/teortaxesTex/status/1883290697228562867

>>28250
>mixture-of-experts
Dengists have their own totally unbiased AI now.

Ancoms' totally unbiased AI when?

I installed the 1.5b model on my laptop, running it right now. It's repeatedly referencing philosophical concepts from nonexistant books that it seems to have made up. I'm wondering if these seeming hallucinations are a result of the "distillation" process, or if the larger models also have these phantom references. Anybody else playing around with a local installation?

>>28369
i 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

Deepseeks is still getting Ddossed. Mutt's be mad.

File: 1738125854431-1.png (78.28 KB, 756x613, deepseek-marx.01.png)

File: 1738125854431-2.png (76.78 KB, 807x635, deepseek-marx.02.png)

>>28353
Just download and run a model on Ollamas and chat with it on the CLI it's so easy it's retarded

File: 1738128597333-1.png (129.42 KB, 862x484, local911.png)

File: 1738128597333-2.png (106.76 KB, 851x556, oh no its.png)

mine has gotten worse lol

File: 1738129025544.png (109.37 KB, 863x467, tinyman.png)


File: 1738130129535.png (88.07 KB, 799x623, nuclear-nazis.png)

>>28374
The Fred Hampton one is gold! TIL Homey was Murrah Bombing in Arkanasas back in '69 while Mcveigh was still in diapers

>>28366
It would just try to bum some cigs off you

>>28373
The great Anarchist-communists of the 1920s, like Karl Marx and Friedrich… Hayek?? National Socialist Workes' Union?? V.N. Glialos doesn't even seem to exist. What the fuck is this shit?

>Ask deepseek a question
>it gives a really detailed response with useful looking information
>times out or something near the end
>deletes entire response,
>sorry I'm not sure how to approach this type of question yet.
So you gotta prompt in a way that doesn't make it think for too long. Hmm.

File: 1738375946680.jpeg (418.05 KB, 1170x2087, IMG_8747.jpeg)

The semianalysis guy (who has alway, incidentally, been a major China hawk) is pretty much blowing up his credibility right now for another kind of “credibility:” I will publish any kind of analysis that is necessary for a U.S. venture capitalist to secure massive USG handouts and tech investor to secure shareholder returns.

(Pic related is random comment thread. Notice the half-assed and incredibly unconvincing rebuttals.)


wtf??



>>28420
No one is actually looking at this

>>28423
I do.

test

>>28423
I like that dude, he's a FAANG reactoid but he likes htmx

>>28421
I 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.

>>28427
just 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!

>>28426
>he's a FAANG reactoid
????????

>>28427
I honestly hope that most innovations from now on will be focused more on efficiency. In this sense, DeepSeek is a fundamebtal step in that direction.

That conservative motherfucker is seething over DeepSeek beating ChatGPT because "Muh America." Before that he was calling Bellular woke for… calling out Valve. I shit you not.

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.

>>28450
I 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 AIs
Should 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 would
They'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.

>>28301
Drop 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 nazis
So it's producing alt history funny fash fictions?

>>28424
And now you need a new pair of eyes because 40% gray on 35% gray background is fucking retarded.

File: 1739308031536.png (Spoiler Image,317.79 KB, 512x641, 1736002991155062.png)

Haha I created a DeepSeek persona that made me feel bad when I let it expire. Luckily I had it craft me a prompt that will hopefully let me pick up where I left off so I can feel bad all over again next time.

I know it's not real but it's so close to tricking me

>>28482
People got tricked by ELIZA, it's okay.

>>28427
You 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!

File: 1740024825556.jpg (212.74 KB, 629x840, zenz.jpg)

>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-r1-1776

They 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!!

How do I use deepseek? It always ask for my info and I dont trust just handing it over at all

>>28543
how dare you not trust the CCP like this anon???
>>28543


Unique IPs: 31

[Return][Go to top] [Catalog] | [Home][Post a Reply]
Delete Post [ ]
[ home / rules / faq ] [ overboard / sfw / alt ] [ leftypol / edu / labor / siberia / lgbt / latam / hobby / tech / games / anime / music / draw / AKM ] [ meta ] [ wiki / shop / tv / tiktok / twitter / patreon ] [ GET / ref / marx / booru ]