Skip Navigation

leanleft

@ leanleft @lemmy.ml

Posts
4
Comments
75
Joined
7 yr. ago

  • GLM-5.3 Imminent drop

    Jump

  • Don't Contribute Anything Relevant in Web Forums Like Reddit

    Jump

  • personally, i would be cautious with discord too

  • How China gets better bang for its buck than America in AI. Its investment lags far behind America’s. Its models do not

    Jump

  • i distilled this article

    Summary of the article “How China gets better bang for its buck than America in AI” (Aug 3 2026)

    • U.S. AI spending is massive – Bloomberg Intelligence estimates U.S. data‑centre capital outlays could exceed $740 billion in 2026, with Nvidia alone negotiating a $250 billion financing deal for a $500 billion data‑centre run by OpenAI. Alphabet announced a $205 billion AI budget.
    • China spends far less – Chinese tech firms are projected to invest less than one‑tenth of the U.S. amount in data centres. Yet their models perform only slightly behind U.S. equivalents. For example:
      • K3 (Moonshot AI) scores ≈ 95 % of Anthropic’s Fable 5 on common benchmarks while being 70 % cheaper to run.
      • Alibaba’s newly released model ranks among the world’s best on certain metrics.

    • Why Chinese spending is efficient
      1. Lower input costs – Land, construction, equipment and labour are cheaper in China.
      2. Model distillation – Chinese labs often train models using outputs from expensive U.S. models, reducing the compute needed.
      3. Hidden spending – Some expenditures on high‑end chips are masked as “cost‑saving” techniques that make inferior hardware achieve higher performance (e.g., DeepSeek’s efficiency tricks).

    • Export restrictions limit Chinese capital use – U.S. bans on advanced AI chips (Nvidia designs, TSMC manufacturing) prevent China from buying the most powerful hardware.
      • Chinese firms are pushed toward domestic alternatives (Huawei, SMIC).
      • Sanctions also block access to cutting‑edge chip‑making equipment, forcing costly work‑arounds and capping production capacity.

    • Domestic demand constraints – Chinese enterprises spend < 10 % of what U.S. firms spend on IT, despite China’s GDP being two‑thirds of the U.S. (or a third larger in PPP terms). This throttles revenue prospects for AI providers, curbing their willingness to invest heavily.
    • Strategic focus differs – The Chinese Communist Party emphasizes diffusing AI across the economy, not pursuing a race toward artificial general intelligence (AGI). Fewer than ten Chinese firms target AGI, compared with dozens of U.S. players.
    • Investor attitudes – Chinese investors have historically punished over‑spending on AI, whereas U.S. investors once rewarded aggressive budgeting. This cultural difference keeps Chinese AI budgets modest.
    • Potential bottlenecks for China – Despite restraint, China may face compute shortages:
      • ByteDance experiences ten‑hour processing times for some videos.
      • Alibaba Cloud, Zhipu AI, and Moonshot’s K3 have long waiting lists or quickly sell out capacity.
      • Over‑restriction could stifle growth if AI services cannot meet user demand.

    Overall takeaway: China achieves comparable AI performance to the U.S. while spending a fraction of the capital by leveraging cheaper resources, model‑distillation techniques, and a strategic focus on wide‑scale diffusion rather than raw computational power. However, export bans, limited domestic chip capacity, modest corporate demand, and cautious investors together create both an efficiency advantage and a risk of under‑provisioned infrastructure.

  • How do you get better at communicating your thoughts in discussions?

    Jump

  • "Or is it really just keeping your mouth shut if you aren't knowledgeable about something."i often abreviate by saying "i know some stuff about X topic.." just leave it at that.if they really want to listen then they will request my expertise. im not going to beg to abused while im trying to do someone a favor.but first, i just dont conversate deeply with shitty people.if someone has a good heart, then maybe they might not be the best conversation partner. thats ok. as long as they deliver on all the essential value, that i need from them.complaining to the city clerk about various issues with the city. i think thats challenging and requires alot of carefully scripted+rehearsed fomation&responses.i dont have a good answer. but maybe act kinda like a carefully controlled, unemotional, plan-adhering robot. (add back in some scripted emotion so that you dont sound like a robot.)

  • GLM-5.3 Imminent drop

    Jump

  • glm 5.2 is already pretty good, soo...

  • [CBH] Released Lanemu P2P VPN 0.14.1 - Open-source alternative to Hamachi/Zerotier/Radmin/etc.

    Jump

  • theres also nebula and netbird

  • TIL you can get much better online prices using a public library computer

    Jump

  • go to https://www.bonappetit.com/ and watch the price suddenly jump! 

        
        
bonappetit.com 4d.condenastdigital.com * block
bonappetit.com ads-static.conde.digital * block
bonappetit.com allure.com * block
bonappetit.com apis.google.com * block
bonappetit.com architecturaldigest.com * block
bonappetit.com bd1cec50-00d1-4ce9-9572-785857419a1e.edge.permutive.app * block
bonappetit.com cdn.gladly.com * block
bonappetit.com cntraveler.com * block
bonappetit.com conde.digital * block
bonappetit.com condenastdigital.com * block
bonappetit.com edge.permutive.app * block
bonappetit.com epicurious.com * block
bonappetit.com gladly.com * block
bonappetit.com glamour.com * block
bonappetit.com globalservices.conde.digital * block
bonappetit.com google.com * block
bonappetit.com gp-prod.conde.digital * block
bonappetit.com gq.com * block
bonappetit.com id-mapper-service.gp-prod.conde.digital * block
bonappetit.com lil.condenastdigital.com * block
bonappetit.com martech.condenastdigital.com * block
bonappetit.com newyorker.com * block
bonappetit.com permutive.app * block
bonappetit.com pitchfork.com * block
bonappetit.com privacy.condenastdigital.com * block
bonappetit.com self.com * block
bonappetit.com teenvogue.com * block
bonappetit.com vanityfair.com * block
bonappetit.com variants.conde.digital * block
bonappetit.com vogue.com * block
bonappetit.com wired.com * block
bonappetit.com www.allure.com * block
bonappetit.com www.architecturaldigest.com * block
bonappetit.com www.cntraveler.com * block
bonappetit.com www.epicurious.com * block
bonappetit.com www.glamour.com * block
bonappetit.com www.gq.com * block
bonappetit.com www.newyorker.com * block
bonappetit.com www.self.com * block
bonappetit.com www.teenvogue.com * block
bonappetit.com www.vanityfair.com * block
bonappetit.com www.vogue.com * block
bonappetit.com www.wired.com *  block

    goddamnit. all i wanted was a muffin !why are you counting my pockets so aggressively ?

  • What is a life lesson you never seem to learn?

    Jump

  • guarana tea

  • LocalLLaMA @sh.itjust.works
    leanleft @lemmy.ml

    model: deepseek-moe-16b

    github.com /deepseek-ai/DeepSeek-MoE/tree/main

  • Don't Contribute Anything Relevant in Web Forums Like Reddit

    Jump

  • Analogy

    Imagine a restaurant where diners place orders through a discreet slot on the kitchen door.

    • The web user is the diner who slips a note into the slot asking for a specific dish.
    • The web‑service backend is the chef who receives the note, prepares the meal, and slides it back out the slot for the diner to collect.

    Now, the diner adds a bizarre request: “Please serve the food, but I never want you to know you ever cooked it.” In practice, the chef can’t fulfill this—once he’s touched the ingredients and used the stove, his hands are inevitably stained, and the kitchen’s heat tells the story. Likewise, when a web user asks a server to fetch data, the server must process the request, so it inevitably knows it handled that request, even if it can’t tie it back to the individual’s identity.

    Ad‑blockers, cookies, IP addresses, user‑agents, and browser signatures are like the various “fingerprints” a diner might leave on the restaurant’s doorstep:

    • Cookies are tiny crumbs the diner drops on the floor while walking to the slot. The chef (or later staff) can later sweep them up and infer that the same person visited before.
    • IP address is the diner’s shoeprints on the hallway carpet leading to the kitchen—an easy way to trace where the diner entered from.
    • User‑agent / browser signature is the unique style of the diner’s napkin (its color, fold, and logo). Even if the napkin is discarded, the pattern tells the chef which brand of napkin was used.
    • Ad‑blockers are like the diner wearing a cloak that blocks the kitchen staff from seeing the crumbs he drops, trying to keep the chef from noticing the usual “tipping” (ads) that would normally be left behind.

    Just as no chef can truly be blind to the fact that he cooked a dish, a web service inevitably knows it processed a request, and the “fingerprints” left behind (cookies, IP, user‑agent, etc.) let it—or any intermediary—recognize or track the diner unless the diner takes strong steps (like using a cloak or wiping the floor) to hide those traces.

    https://f-droid.org/packages/com.fauxx.full

  • Looking for selfhosted alternative to "Find my phone"

    Jump

  • if theft is a concern. record IMEI number and they(carriers etc) will probably blacklist the phone from mobile networks and maybe aid the tracking(investigation..if any). which, afterwards, lowers resale value

  • Is there are good alternative to IPFS?

    Jump

  • 256 KB is the chosen default. so this is the best choice. however.. i wonder..if you share all 3900 finetune versions of llama 3.1 8b or all 2900 finetune versions of qwen 2.5 7b instruct

    there is likely alot of matching chunks if you break the chunk size super small.

    so it may not be the most optimal chunk size for sharing.. but you could (potentially have waay more peers (because peeps are all hosting diff models with some shared matches between them)

    ** a more difficult question is wether to create ,efficient, highly compressed archive of specific popular sets of related models("most-popular, starter packs"). this requires communication,coordination,effort to get it EXACTLY right. IMO this is a trade for using centralized trackers and random uploaders.

  • Don't Contribute Anything Relevant in Web Forums Like Reddit

    Jump

  • great post. im glad it has so many upvotes.additionally:i think it ties back to many issues:

    • privacy
    • stance and dividend of LLMs
    • online culture. digital communities
    • citizen rights
    • and lots more!

  • Don't Contribute Anything Relevant in Web Forums Like Reddit

    Jump

  • thats good. but it doesnt cover behavioral fingerprints(indications).fyi i have seen some moderately effective tools that help to fight against this. but generally rare. people understimate how much their data is worth and the amount of development, that companies are willing to invest to break every privacy tool

  • Don't Contribute Anything Relevant in Web Forums Like Reddit

    Jump

  • i think the reson reddit deleted both my accounts, was so that i couldnt modify or delete my contributions.after the accounts are deleted, they probably still have copies saved somewhere. but it revokes my access to remove or tamper with my content.

  • Don't Contribute Anything Relevant in Web Forums Like Reddit

    Jump

  • how many models have you collected locally?

    Jump

  • Proton VPN's anti-censorship Stealth protocol finally lands on Linux

    Jump

  • opinion: title should be /on linux/for linux/

  • You Should Get to Own Your Media, Not Rent it Forever

    Jump

  • i think browsing webpages should be the same.load it once, then store it locally andor host a torrent for it

  • What is your take on romantic love? How do you experience it?

    Jump

  • to me, relationship(or love) entails ..

    <summary>

    making the most of bad situations

    </summary>

    another theory: being bold and available.

  • LocalLLaMA @sh.itjust.works
    leanleft @lemmy.ml

    how many models have you collected locally?

  • LocalLLaMA @sh.itjust.works
    leanleft @lemmy.ml

    alibaba-pai/DistillQwen-ThoughtY-4B

    huggingface.co /alibaba-pai/DistillQwen-ThoughtY-4B
  • Pleroma @lemmy.ml
    leanleft @lemmy.ml

    Tinmop client for Pleroma using the mastodon API and gemini.

    www.autistici.org /interzona/tinmop.html