UNWIRE.HK<p>DeepSeek V3.1 出現代碼污染問題 「极」字不斷出現於生成代碼中 暫時未知污染數據來源<br>DeepSeek 最新推出的 V3.1 語言模型被發現存在技術缺陷，在生成程式碼時會意外插入中文字符「极」，影響模型在編程應用上的可靠性。多名開發者在不同推理平台上都發現這個問題，顯示這並非個別事件。<br> <a href="https://mastodon.hongkongers.net/tags/%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>人工智能</span></a> <a href="https://mastodon.hongkongers.net/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> <a href="https://mastodon.hongkongers.net/tags/deepseek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>deepseek</span></a> v3.1 <a href="https://mastodon.hongkongers.net/tags/%E6%8A%80%E8%A1%93%E7%BC%BA%E9%99%B7" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>技術缺陷</span></a><br><a href="https://unwire.hk/2025/08/26/deepseek-v3-coding-bug-chinese-character-issue/ai/?utm_source=rss&utm_medium=rss&utm_campaign=deepseek-v3-coding-bug-chinese-character-issue" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">unwire.hk/2025/08/26/deepseek-</span><span class="invisible">v3-coding-bug-chinese-character-issue/ai/?utm_source=rss&utm_medium=rss&utm_campaign=deepseek-v3-coding-bug-chinese-character-issue</span></a></p>
101010.pl is one of the many independent Mastodon servers you can use to participate in the fediverse.
101010.pl czyli najstarszy polski serwer Mastodon. Posiadamy wpisy do 2048 znaków.
Administered by:
Server stats:
476active users
101010.pl: About · Status · Profiles directory · Privacy policy
Mastodon: About · Get the app · Keyboard shortcuts · View source code · v4.5.0-alpha.2
#deepseek
7 posts · 7 participants · 1 post today
PUPUWEB Blog<p>The new DeepSeek V3.1 AI is a budget powerhouse. It uses a smart hybrid-reasoning system to tackle complex tasks. While not yet dethroning GPT-5, its performance-to-cost ratio makes it a compelling choice for developers. <a href="https://mastodon.social/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AI</span></a> <a href="https://mastodon.social/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> <a href="https://mastodon.social/tags/Tech" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Tech</span></a></p><p><a href="https://pupuweb.com/why-deepseek-v3-1-fall-short-against-gpt-5-claude-opus/" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">pupuweb.com/why-deepseek-v3-1-</span><span class="invisible">fall-short-against-gpt-5-claude-opus/</span></a></p>
Geekland<p>Lo que DeepSeek acaba de lanzar no gustará ni a NVIDIA ni a las empresas de chips de Estados Unidos: su nombre es DeepSeek-V3.1 <a href="https://mastodon.social/tags/deepseek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>deepseek</span></a> <a href="https://mastodon.social/tags/noticias" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>noticias</span></a> <a href="https://mastodon.social/tags/china" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>china</span></a> <a href="https://www.genbeta.com/a-fondo/que-deepseek-acaba-lanzar-no-gustara-a-nvidia-a-empresas-chips-estados-unidos-su-nombre-deepseek-v3-1" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">genbeta.com/a-fondo/que-deepse</span><span class="invisible">ek-acaba-lanzar-no-gustara-a-nvidia-a-empresas-chips-estados-unidos-su-nombre-deepseek-v3-1</span></a></p>
Online-Magazin Telepolis<p>Auf chinesische Hardware optimiert: DeepSeek stellt neues KI-Modell vor</p><p>Das chinesische Start-up präsentiert eine verbesserte Version seines Flaggschiff-Modells. Die Entwicklung erfolgt vor dem Hintergrund verschärfter US-Exportbeschränkungen.</p><p><a href="https://www.heise.de/tp/features/Auf-chinesische-Hardware-optimiert-DeepSeek-stellt-neues-KI-Modell-vor-10578917.html?wt_mc=sm.red.ho.mastodon.mastodon.md_beitraege.md_beitraege&utm_source=mastodon" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">heise.de/tp/features/Auf-chine</span><span class="invisible">sische-Hardware-optimiert-DeepSeek-stellt-neues-KI-Modell-vor-10578917.html?wt_mc=sm.red.ho.mastodon.mastodon.md_beitraege.md_beitraege&utm_source=mastodon</span></a></p><p><a href="https://social.heise.de/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> <a href="https://social.heise.de/tags/K%C3%BCnstlicheIntelligenz" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>KünstlicheIntelligenz</span></a> <a href="https://social.heise.de/tags/news" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>news</span></a></p>
UNWIRE.HK<p>DeepSeek V3.1 悄然推出 時隔 5 個月推出混合推理架構升級 針對國產晶片運算最佳化<br>中國人工智能初創企業 DeepSeek 時隔近 5 個月再度對 V3 基座模型進行升級，新發佈的 DeepSeek V3.1 模型具備對中國製造晶片最佳化功能及更快處理速度。DeepSeek 在微信帖文中表示，V3.1 模型採用 UE8M0 FP8 精度格式，該格式專為「即將發佈的下一代國產晶片」而設計，但未具體說明新兼容功能支援哪些晶片型號或製造商。<br> <a href="https://mastodon.hongkongers.net/tags/%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>人工智能</span></a> <a href="https://mastodon.hongkongers.net/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> <a href="https://mastodon.hongkongers.net/tags/deepseek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>deepseek</span></a> v3.1 <a href="https://mastodon.hongkongers.net/tags/%E5%9C%8B%E7%94%A2%E6%99%B6%E7%89%87" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>國產晶片</span></a><br><a href="https://unwire.hk/2025/08/22/deepseek-v31-china-chip-optimization/ai/?utm_source=rss&utm_medium=rss&utm_campaign=deepseek-v31-china-chip-optimization" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">unwire.hk/2025/08/22/deepseek-</span><span class="invisible">v31-china-chip-optimization/ai/?utm_source=rss&utm_medium=rss&utm_campaign=deepseek-v31-china-chip-optimization</span></a></p>
Anonymous<p>DeepSeek otrzymał dużą aktualizację – co nowego?</p><p><a href="https://myecho.pl/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> <a href="https://myecho.pl/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AI</span></a> <a href="https://myecho.pl/tags/SztucznaInteligencja" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>SztucznaInteligencja</span></a> <a href="https://myecho.pl/tags/Technologia" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Technologia</span></a> <a href="https://myecho.pl/tags/News" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>News</span></a></p><p><a href="https://onetech.pl/deepseek-otrzymal-duza-aktualizacje-co-nowego/" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">onetech.pl/deepseek-otrzymal-d</span><span class="invisible">uza-aktualizacje-co-nowego/</span></a></p>
UNWIRE.HK<p>DeepSeek V3.1 靜靜發布免費開放 大幅提升程式編寫及推理能力<br>中國 AI 研發實驗室 DeepSeek 於 8 月 19 日透過官方微信群組悄然發布 V3.1 版本，這次更 […]<br> <a href="https://mastodon.hongkongers.net/tags/%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>人工智能</span></a> <a href="https://mastodon.hongkongers.net/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AI</span></a> <a href="https://mastodon.hongkongers.net/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> <a href="https://mastodon.hongkongers.net/tags/deepseek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>deepseek</span></a> v3.1<br><a href="https://unwire.hk/2025/08/20/deepseek-v31-free-ai-cost-analysis/ai/?utm_source=rss&utm_medium=rss&utm_campaign=deepseek-v31-free-ai-cost-analysis" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">unwire.hk/2025/08/20/deepseek-</span><span class="invisible">v31-free-ai-cost-analysis/ai/?utm_source=rss&utm_medium=rss&utm_campaign=deepseek-v31-free-ai-cost-analysis</span></a></p>
Sunflower Björnskalle 🌻<p><a href="https://mastodon.online/tags/ChatGPT" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>ChatGPT</span></a> and <a href="https://mastodon.online/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> are so incredibly wrong when you ask it even basic questions about <a href="https://mastodon.online/tags/typetheory" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>typetheory</span></a>. I asked for the top 10 simplest recursive types. Only 4 of them even were for DS and ChatGPT wasn't much better but a bunch of its formulations were even wrong. One of the things it gave wasn't even semantically let alone syntactically valid. Also the ways they were wrong were braindead: the unit type and empty types aren't even close.</p><p><a href="https://mastodon.online/tags/LLM" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>LLM</span></a> <a href="https://mastodon.online/tags/ai" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>ai</span></a></p>
Geekland<p>DeepSeek R2: El fallido intento de China por sustituir a Nvidia con chips de Huawei <a href="https://mastodon.social/tags/ia" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>ia</span></a> <a href="https://mastodon.social/tags/deepseek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>deepseek</span></a> <a href="https://wwwhatsnew.com/2025/08/16/deepseek-r2-el-fallido-intento-de-china-por-sustituir-a-nvidia-con-chips-de-huawei/" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">wwwhatsnew.com/2025/08/16/deep</span><span class="invisible">seek-r2-el-fallido-intento-de-china-por-sustituir-a-nvidia-con-chips-de-huawei/</span></a></p>
Andrii Kuznietsov<p>🤖🇨🇳 Провідний китайський стартап у сфері ШІ <a href="https://social.kyiv.dcomm.net.ua/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> був змушений відкласти випуск своєї майбутньої моделі R2 через труднощі з її навчанням. Проблеми почались після втручання політиків у робочі процеси - китайські чиновники наполягли на тому, щоб DeepSeek відмовилась від американських чіпів <a href="https://social.kyiv.dcomm.net.ua/tags/Nvidia" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Nvidia</span></a> на користь продукції місцевої компанії <a href="https://social.kyiv.dcomm.net.ua/tags/Huawei" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Huawei</span></a>. Однак навчання R2 зіткнулося з постійними збоями обладнання Huawei, що затримало випуск моделі.</p><p><a href="https://www.tomshardware.com/tech-industry/artificial-intelligence/deepseek-reportedly-urged-by-chinese-authorities-to-train-new-model-on-huawei-hardware-after-multiple-failures-r2-training-to-switch-back-to-nvidia-hardware-while-ascend-gpus-handle-inference" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">tomshardware.com/tech-industry</span><span class="invisible">/artificial-intelligence/deepseek-reportedly-urged-by-chinese-authorities-to-train-new-model-on-huawei-hardware-after-multiple-failures-r2-training-to-switch-back-to-nvidia-hardware-while-ascend-gpus-handle-inference</span></a></p>
Linux Magazine<p>Newelle, a Linux AI assistant, works with different LLMs and includes document parsing and profiles<br><a href="https://www.linux-magazine.com/Online/News/A-New-Linux-AI-Assistant-in-Town?utm_source=mlm" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">linux-magazine.com/Online/News</span><span class="invisible">/A-New-Linux-AI-Assistant-in-Town?utm_source=mlm</span></a><br><a href="https://fosstodon.org/tags/Newelle" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Newelle</span></a> <a href="https://fosstodon.org/tags/Linux" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Linux</span></a> <a href="https://fosstodon.org/tags/ArtificialIntelligence" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>ArtificialIntelligence</span></a> <a href="https://fosstodon.org/tags/LLM" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>LLM</span></a> <a href="https://fosstodon.org/tags/MachineLearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>MachineLearning</span></a> <a href="https://fosstodon.org/tags/Ollama" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Ollama</span></a> <a href="https://fosstodon.org/tags/Groq" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Groq</span></a> <a href="https://fosstodon.org/tags/Google" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Google</span></a> <a href="https://fosstodon.org/tags/Gemini" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Gemini</span></a> <a href="https://fosstodon.org/tags/OpenAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>OpenAI</span></a> <a href="https://fosstodon.org/tags/gpt4free" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>gpt4free</span></a> <a href="https://fosstodon.org/tags/Claude" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Claude</span></a> <a href="https://fosstodon.org/tags/Mistral" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Mistral</span></a> <a href="https://fosstodon.org/tags/OpenRouter" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>OpenRouter</span></a> <a href="https://fosstodon.org/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a></p>
ROOTBLOG<p>🆕Fiasko treningu AI DeepSeek na chipach Huaweia: chińskie ambicje napotykają na bariery technologiczne<br />➡️<a href="https://rootblog.pl/fiasko-treningu-ai-deepseek-na-chipach-huaweia-chinskie-ambicje-napotykaja-na-bariery-technologiczne/" target="_blank" rel="nofollow noopener" translate="no"><span class="invisible">https://</span><span class="ellipsis">rootblog.pl/fiasko-treningu-ai</span><span class="invisible">-deepseek-na-chipach-huaweia-chinskie-ambicje-napotykaja-na-bariery-technologiczne/</span></a></p><p><a href="https://101010.pl/tags/ai" class="mention hashtag" rel="tag">#<span>ai</span></a> <a href="https://101010.pl/tags/Chiny" class="mention hashtag" rel="tag">#<span>Chiny</span></a> <a href="https://101010.pl/tags/deepseek" class="mention hashtag" rel="tag">#<span>deepseek</span></a> <a href="https://101010.pl/tags/huawei" class="mention hashtag" rel="tag">#<span>huawei</span></a> <a href="https://101010.pl/tags/pprzewodniki" class="mention hashtag" rel="tag">#<span>pprzewodniki</span></a> <a href="https://101010.pl/tags/technologia" class="mention hashtag" rel="tag">#<span>technologia</span></a> <a href="https://101010.pl/tags/USA" class="mention hashtag" rel="tag">#<span>USA</span></a></p>
Catharina Bethlehem ☑️ Cath<p>Heel interessante <span class="h-card" translate="no"><a href="https://mastodon.green/@tegenlicht" class="u-url mention" rel="nofollow noopener" target="_blank">@<span>tegenlicht</span></a></span> met de Chinese econoom Andy Xie over de AI-bubbel.</p><p>Het gaat o.a. ook over de Chinese AI Deepseek, die in tegenstelling tot OpenAI wel open source is, hoe en waarom er in de VS zoveel geld rondgepompt wordt en over de enorme ongelijkheid daar.</p><p>Artikel en video:<br><a href="https://tegenlicht.vpro.nl/artikelen/wanneer-barst-de-ai-bubbel" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">tegenlicht.vpro.nl/artikelen/w</span><span class="invisible">anneer-barst-de-ai-bubbel</span></a></p><p><a href="https://mastodon.social/tags/Tegenlicht" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Tegenlicht</span></a> <a href="https://mastodon.social/tags/Deepseek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Deepseek</span></a> <a href="https://mastodon.social/tags/OpenAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>OpenAI</span></a> <a href="https://mastodon.social/tags/AIBubbel" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AIBubbel</span></a> <a href="https://mastodon.social/tags/China" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>China</span></a> <a href="https://mastodon.social/tags/VS" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>VS</span></a> <a href="https://mastodon.social/tags/VPRO" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>VPRO</span></a> </p><p><a href="https://mastodon.social/tags/TegenlichtMoetBlijven" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>TegenlichtMoetBlijven</span></a></p>
UNWIRE.HK<p>DeepSeek 母公司高層涉貪 行銷總監私吞 2,160 萬港元非法佣金遭拘留<br>中國 AI 初創公司 DeepSeek 母公司幻方量化（High-Flyer Quant）行銷總監李橙，涉嫌透 […]<br> <a href="https://mastodon.hongkongers.net/tags/%E7%A7%91%E6%8A%80%E6%96%B0%E8%81%9E" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>科技新聞</span></a> <a href="https://mastodon.hongkongers.net/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a><br><a href="https://unwire.hk/2025/08/10/deepseek-parent-company-illegal-money/fun-tech/?utm_source=rss&utm_medium=rss&utm_campaign=deepseek-parent-company-illegal-money" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">unwire.hk/2025/08/10/deepseek-</span><span class="invisible">parent-company-illegal-money/fun-tech/?utm_source=rss&utm_medium=rss&utm_campaign=deepseek-parent-company-illegal-money</span></a></p>
𝙹𝚘𝚑𝚊𝚗<p><a href="https://cr8r.gg/tags/NeuralNetwork" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>NeuralNetwork</span></a> <a href="https://cr8r.gg/tags/%D0%B2%D0%BE%D0%BF%D1%80%D0%BE%D1%81" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>вопрос</span></a> </p><p>А поделитесь в каментах, какая ваша любимая нейронка в качестве собеседника? <a href="https://cr8r.gg/tags/%D0%9A%D0%BB%D0%B0%D0%B2%D0%B4%D0%B8%D1%8F" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Клавдия</span></a> меня забанила (я хз, за что, возможно, за «не ту страну», и усекла, что ее дурит <a href="https://cr8r.gg/tags/comss" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>comss</span></a> 😔) <a href="https://cr8r.gg/tags/Mistral" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Mistral</span></a> — неплоха, потому что умеет в длинные контексты, перевод, этимологию — юзаю для <a href="https://cr8r.gg/tags/%D1%8F%D0%B7%D1%8B%D0%BA%D0%BE%D0%B7%D0%BD%D0%B0%D0%BD%D0%B8%D0%B5" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>языкознание</span></a>, но не только ее. <a href="https://cr8r.gg/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> — какие-то сухие ответы. Вот <a href="https://cr8r.gg/tags/Qwen" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Qwen</span></a> вроде с потенициалом, но надо тестить и тестить. <a href="https://cr8r.gg/tags/Grok" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Grok</span></a> — пишет статьи c пруфами, приводит в нормальный вид автосубтитры ютуб-роликов, причем можно просить дословно, можо — переработать полтора часа бормотания подкастеров в аккуратную статью. Но тот факт, что нужно всё равно потом вычитывать — рушит всю идею использования «для себя», это только если хочешь пост интересный сделать.</p><p><a href="https://cr8r.gg/tags/Perplexity" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Perplexity</span></a>, <a href="https://cr8r.gg/tags/phind" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>phind</span></a>, <a href="https://you.com" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="">you.com</span><span class="invisible"></span></a> — такие уже более нишевые про кодинг, рисуют блок-схемы и вот это всио.</p><p>А вот <a href="https://cr8r.gg/tags/ChatGPT" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>ChatGPT</span></a> — она как-то <em>личность</em> выстраивает. И память между диалогами какую-то сохраняет, и общение получается более «человечное». Вот, кота нарисовала. Код, однако, более <em>избыточный</em> по сравнению с <a href="https://cr8r.gg/tags/Perplexity" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Perplexity</span></a> и <a href="https://cr8r.gg/tags/phind" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>phind</span></a>, но человечность в общении однако подкупает.</p><p>А у вас как? Может, посоветуете конкретные юзкейсы для конкретных нейросеток?</p><p><span class="h-card" translate="no"><a href="https://mastodon.ml/@rf" class="u-url mention" rel="nofollow noopener" target="_blank">@<span>rf</span></a></span> <br><span class="h-card" translate="no"><a href="https://3zi.ru/@Russia" class="u-url mention" rel="nofollow noopener" target="_blank">@<span>Russia</span></a></span> <br><span class="h-card" translate="no"><a href="https://mastodon.social/@russian_mastodon" class="u-url mention" rel="nofollow noopener" target="_blank">@<span>russian_mastodon</span></a></span></p>
heise online English<p>OpenAI publishes new open models</p><p>With two open-weight reasoning models, OpenAI apparently wants to stand up to the competition from China in particular. One challenge was security.</p><p><a href="https://www.heise.de/en/news/OpenAI-publishes-new-open-models-10510608.html?wt_mc=sm.red.ho.mastodon.mastodon.md_beitraege.md_beitraege&utm_source=mastodon" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">heise.de/en/news/OpenAI-publis</span><span class="invisible">hes-new-open-models-10510608.html?wt_mc=sm.red.ho.mastodon.mastodon.md_beitraege.md_beitraege&utm_source=mastodon</span></a></p><p><a href="https://social.heise.de/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> <a href="https://social.heise.de/tags/IT" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>IT</span></a> <a href="https://social.heise.de/tags/K%C3%BCnstlicheIntelligenz" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>KünstlicheIntelligenz</span></a> <a href="https://social.heise.de/tags/OpenAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>OpenAI</span></a> <a href="https://social.heise.de/tags/news" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>news</span></a></p>
heise online<p>OpenAI veröffentlicht neue offene Modelle</p><p>Mit zwei Open-Weight-Reasoning-Modellen will OpenAI offenbar vor allem der Konkurrenz aus China die Stirn bieten. Eine Herausforderung war dabei die Sicherheit.</p><p><a href="https://www.heise.de/news/OpenAI-veroeffentlicht-neue-offene-Modelle-10510550.html?wt_mc=sm.red.ho.mastodon.mastodon.md_beitraege.md_beitraege&utm_source=mastodon" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">heise.de/news/OpenAI-veroeffen</span><span class="invisible">tlicht-neue-offene-Modelle-10510550.html?wt_mc=sm.red.ho.mastodon.mastodon.md_beitraege.md_beitraege&utm_source=mastodon</span></a></p><p><a href="https://social.heise.de/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> <a href="https://social.heise.de/tags/IT" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>IT</span></a> <a href="https://social.heise.de/tags/K%C3%BCnstlicheIntelligenz" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>KünstlicheIntelligenz</span></a> <a href="https://social.heise.de/tags/OpenAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>OpenAI</span></a> <a href="https://social.heise.de/tags/news" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>news</span></a></p>
Agnieszka Serafinowicz<p><strong>Chiński Z.ai rzuca rękawicę gigantom. Nowy model AI GLM-4.5 ma być otwarty, tani i rekordowo wydajny</strong></p><p>I lepszy od modelu DeepSeek. Startup Z.ai zaprezentował nowe rozwiązanie – GLM-4.5. Jest to model AI udostępniony w formule open source, który ma być jeszcze tańszy, wydajniejszy i „bystrzejszy” od swojego głośnego, chińskiego poprzednika.<br></p><p>Nowy model, GLM-4.5, wyróżnia się tak zwaną „agentową” architekturą, co oznacza, że potrafi automatycznie dzielić złożone zadania na mniejsze etapy, by wykonać je precyzyjniej. Co więcej, jest o połowę mniejszy od DeepSeeka i do działania potrzebuje podobno zaledwie ośmiu specjalistycznych chipów Nvidia H20, stworzonych na potrzeby chińskiego rynku z uwzględnieniem amerykańskich restrykcji eksportowych.</p><p>Nowy model GLM-4.5 to dziś najbardziej zaawansowana (publicznie znana) chińska konstrukcja oparta na architekturze MoE (Mixture of Experts), dostępna w dwóch wariantach: flagowym z 355 miliardami parametrów oraz lżejszej wersji Air ze 106 miliardami parametrów. Architektura „agentowa” pozwala mu na autonomiczne planowanie i wykonywanie złożonych, wieloetapowych zadań. Według testów producenta, model Z.ai osiągnął trzeci wynik na świecie w branżowych benchmarkach, plasując się jednocześnie na pierwszym miejscu wśród wszystkich modeli typu open-source. Co niezwykle istotne, jego wysoka wydajność została osiągnięta przy relatywnie niskich wymaganiach sprzętowych.</p><p>Z.ai pozycjonuje swój produkt jako „prawdziwie otwartą alternatywę” dla zamkniętych, autorskich systemów, które dominują na rynku. Model jest dostępny na otwartej licencji, co daje firmom większą kontrolę i transparentność. Jednak to właśnie koszty stanowią jego największą przewagę. Według oficjalnego cennika, koszt przetwarzania miliona tokenów (fragmentów słów) to zaledwie 11 centów za dane wejściowe i 28 centów za dane wyjściowe. Dla porównania, w przypadku konkurencyjnego modelu DeepSeek R1 koszt tokenów wyjściowych to 2,19 dolara, co pokazuje ogromny (wręcz niewiarygodny) skok w optymalizacji kosztowej.</p><p>Dynamiczny rozwój chińskich firm AI ma coraz większe znaczenie geopolityczne. Gdy na początku roku DeepSeek zaprezentował swoją wydajność, wywołało to chwilowe załamanie kursów akcji amerykańskich gigantów technologicznych. Sukcesy Z.ai, firmy założonej w 2019 roku, która zebrała już ponad 1,5 mld dolarów od inwestorów takich jak Alibaba i Tencent, również nie pozostały niezauważone. Firma została wymieniona przez OpenAI jako jeden z nielicznych na świecie konkurentów zdolnych do budowy rywalizujących modeli, a także trafiła do prestiżowego raportu Stanford University „AI Index Report 2025”. Jednocześnie ten szybki postęp wzbudził niepokój w Waszyngtonie – startup Z.ai został wpisany na listę podmiotów objętych amerykańskimi restrykcjami handlowymi.</p><blockquote><p><a href="https://imagazine.pl/2025/02/07/trening-modelu-deepseek-nie-kosztowal-6-mln-dolarow-lecz-13-miliarda-dolarow-raport-semianalysis/" rel="nofollow noopener" target="_blank">Trening modelu DeepSeek nie kosztował 6 mln dolarów, lecz 1,3 miliarda dolarów – raport SemiAnalysis</a></p></blockquote><p></p><p><a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/ai-si/" target="_blank">#AI</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/chiny/" target="_blank">#chiny</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/deepseek/" target="_blank">#DeepSeek</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/geopolityka/" target="_blank">#geopolityka</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/glm-4-5/" target="_blank">#GLM45</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/mixture-of-experts/" target="_blank">#MixtureOfExperts</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/model-jezykowy/" target="_blank">#modelJęzykowy</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/modele-jezykowe/" target="_blank">#modeleJęzykowe</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/news/" target="_blank">#news</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/open-source/" target="_blank">#openSource</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/sztuczna-inteligencja/" target="_blank">#sztucznaInteligencja</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/technologia/" target="_blank">#technologia</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/usa/" target="_blank">#usa</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/z-ai/" target="_blank">#ZAi</a> <a rel="nofollow noopener" class="hashtag u-tag u-category" href="https://imagazine.pl/tag/zhipu/" target="_blank">#Zhipu</a></p>
UNWIRE.HK<p>中國新開源 AI 模型稱成本勝 DeepSeek Z.ai GLM-4.5 輸出每百萬 token 收 US$0.28 平 87%<br>中國初創企業 Z.ai（前稱智譜）週一宣布推出全新 GLM-4.5 AI 模型，聲稱使用成本比 DeepSeek 更低，標誌著中國企業在 AI 價格戰中再下一城。該模型基於「代理式」AI 技術構建，能自動將複雜任務分解為子任務以提高準確性，有別於傳統同時處理所有資訊的做法。<br> <a href="https://mastodon.hongkongers.net/tags/%E4%BA%BA%E5%B7%A5%E6%99%BA%E8%83%BD" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>人工智能</span></a> <a href="https://mastodon.hongkongers.net/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a> <a href="https://mastodon.hongkongers.net/tags/GLM" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>GLM</span></a>-4.5 <a href="https://mastodon.hongkongers.net/tags/Z" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>Z</span></a>.ai<br><a href="https://unwire.hk/2025/07/29/china-ai-model-cheaper/ai/?utm_source=rss&utm_medium=rss&utm_campaign=china-ai-model-cheaper" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">unwire.hk/2025/07/29/china-ai-</span><span class="invisible">model-cheaper/ai/?utm_source=rss&utm_medium=rss&utm_campaign=china-ai-model-cheaper</span></a></p>
Ian Brown 👨🏻💻<p><a href="https://eupolicy.social/tags/DeepSeek" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>DeepSeek</span></a>, ‘the homegrown <a href="https://eupolicy.social/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>AI</span></a> model whose sympathetic use of the Chinese language has proved surprisingly adept at <a href="https://eupolicy.social/tags/BaZi" class="mention hashtag" rel="nofollow noopener" target="_blank">#<span>BaZi</span></a> — a form of astrology that uses a person’s birth time and other details to determine the balance of five elements and other aspects of their destiny.’ 🔮<a href="https://on.ft.com/41aiKCe" rel="nofollow noopener" translate="no" target="_blank"><span class="invisible">https://</span><span class="">on.ft.com/41aiKCe</span><span class="invisible"></span></a></p>
TrendingLive feeds
Mastodon is the best way to keep up with what's happening.
Follow anyone across the fediverse and see it all in chronological order. No algorithms, ads, or clickbait in sight.
Create accountLoginDrag & drop to upload