Register and share your invite link to earn from video plays and referrals.

Search results for 70「8頭身には敵わない」が配信されました!
70「8頭身には敵わない」が配信されました! community
One keyword maps to one global community path.
Create community
People
Not Found
Tweets including 70「8頭身には敵わない」が配信されました!
Only 29.2% can be verified. 70.8% asks you to trust. The Transparency Gap report reveals what’s missing. Read the full report:
GLM-5.3-Flash vs Qwen3.8 Flash-Next: Both on 2× DGX Spark, same 80-test Spark-Bench: -GLM 84.8 -Qwen 79.4 (vLLM) / 77.9 (TensorFold) GLM wins most categories. The biggest gap is long context. Where each one wins • Long context: 70.8 vs 21.6–27.9. GLM found the planted fact on the hard needle test, Qwen missed it on both engines. • Code: 90.2 vs 85.7–87.1 • Visual: 79.8 vs 71.4–75.0 • Tool use: 78.2 vs 58.2–75.6 • Agentic: 78.5 vs 67.1–75.1 Takeaway: Qwen follows instructions and formats more precisely. GLM is the stronger generalist: it codes, builds and retrieves better. Same 80 tests, uncapped, 2 repeats.
Show more
ISM Mfg 54.6, Exp. 55.2, Last 55.6 Prices Paid 71.1, Exp. 70.8 New Orders 53.7, Exp. 56.8 Employment 51.2, Exp. 52.5
BREAKING: Grok 4.6 just took the #1# spot on CursorBench 3.2 — while delivering a massive efficiency advantage. ⚡💻 • Grok 4.6 Extra High — 70.8% | $2.81/task • Fable 5 Max — 70.5% | $17.32/task • Opus 5 Max — 70.0% | $8.23/task • GPT-5.6 Sol Max — 67.2% | $5.69/task Grok achieved the highest score while costing roughly 6× less than Fable 5 Max and nearly 3× less than Opus 5 Max per task. For AI agents, raw intelligence is only part of the equation. The ability to maintain high performance across long coding tasks without burning massive amounts of compute could be a major advantage. Grok’s agentic coding efficiency is becoming seriously impressive. 🚀 Source: CursorBench 3.2
Show more
Grok 4.6 just took the #1# spot on CursorBench 3.2.....and the efficiency is insane Here's the cost comparison: • Grok 4.6 Extra High — 70.8% | $2.81/task • Fable 5 Max — 70.5% | $17.32/task • Opus 5 Max — 70.0% | $8.23/task • GPT-5.6 Sol Max — 67.2% | $5.69/task Grok achieved the highest score while costing roughly 6X less than Fable 5 Max and nearly 3X less than Opus 5 Max per task That’s what makes Grok so powerful for agents Top-tier intelligence is great.....but top-tier intelligence that can keep working across long coding tasks without burning ridiculous amounts of compute is even better Grok’s agentic coding efficiency is insane
Show more
FULL INTERVIEW: @sophiadew sat down with @poteto and @roshan_s from Cursor on Grok 4.6, agents that fix their own bugs, and how an engineer's job changes. Grok 4.6 Extra High scored 70.8% on CursorBench 3.2 at $2.81 a task. Fable 5 Max scored 70.5% at $17.32. Same result, 6x the price. 1:07 what Grok Bot actually is 2:15 the Odyssey IMAX tickets an agent booked for her 3:55 live demo: a coordinator agent running a team of bots 4:29 a marketing agent posting to LinkedIn, live on air 7:46 what Grok 4.6 changes 9:59 what changes when your agent is a colleague, not a tool 11:44 Benny, the Slack bot that fixed bugs while she slept 13:04 how much of Grok Bot was built with Grok Bot 13:29 second demo: dispatching bug fixes to a team of agents 14:36 your bots can kick off Cursor cloud agents 17:46 rewriting your codebase so agents can thrive in it 18:34 refactors that would have taken a team months 20:12 the Michelin kitchen metaphor 22:23 the iMessage support she hacked in early 23:36 the CursorBench numbers 25:17 "if you have a budget of 20 bucks to do all of your AI tasks" @SpaceXAI @cursor_ai @bot
Show more
$VRDN(Viridian Therapeutics) 5.2.1 快照:现价24.27|市值2.75B|52周区间13.18到34.29|现金及等价物212.4M|短期投资662.3M|流动性合计874.7M|季度经营现金流出105.3M|跑道8.3个季度约2.1年|年度收入70.8M|净利润-342.6M|研发338.9M|资本支出-0.5M|季度收入0.3M|季度净利润-127.1M|流动比率1264.9%|毛利率100.0%|营业利润率-512.91%|技术选股A组分数78、硬条件命中、获利盘95.52%、套牢盘4.48%、90%集中度20.16(来源Nasdaq财报接口与技术选股 10/29/2026) 5.2.2 主营业务与产品技术亮点:专注甲状腺眼病等自身免疫罕见病抗体药物开发,属生物医药创新大类创新药基因子行业;核心管线为长效IGF-1R抗体,走皮下注射路线以替代静脉给药;管线形态为单一适应症深耕加多适应症拓展;客户与产业链位置为自研自产、面向专科内分泌与眼科渠道(业务描述,无数字) 5.2.3 事件促发点:FDA批准Lumvoa后估值重估讨论,已发生2026-10-28,量级中等,利多(来源Google News Simply Wall St 10/28/2026);财报日未在数据块未来5日日历中列出,暂无已公开促发点 5.2.4 大行研报与消息汇总:Nasdaq分析师库买入12家持有0家卖出0家,目标价均值34.11、区间22.0到41.0,距现价约40.5%(来源Nasdaq分析师库 10/29/2026);消息面Simply Wall St称FDA批准Lumvoa后估值仍偏低(来源Google News 10/28/2026);头部机构专项目标价本次未采集 5.2.5 关键价位、触发与失效:日线支撑筹码峰22.0距现价10.32%|日线压力20日高24.42距现价0.62%|周线支撑周线MA10 21.46距现价11.58%|周线压力52周高34.29距现价41.29%(来源个股筹码四元组 10/29/2026);触发为放量站稳24.42;失效为跌破22.0;资金共振同向2项 5.2.6 情景:牛档34.29(周线压力52周高,从现价空间41.29%);基准档24.42(日线压力20日高);熊档21.46(周线支撑周线MA10) 5.2.7 结论:可买,1到2周期望先看24.42再看34.29,赔率约1比2以上,建议半仓,最大风险为临床与商业化节奏不及预期。
Show more
What if a model's endless repetition loops could be fixed with nothing but bit-level weight edits? Tree-Ising: Repairing Language Models with Bit-Level Ising Edits 📌 Overview A method that repairs Qwen3.5-0.8B's thinking-mode loops with sparse discrete bit edits to the weights. No gradients, no training. ⚠️ Problem Prior work had an LLM agent read the weights and write a fix, taking about $312 and 15.6 hours. 🧩 Method ・Move only the sign and a 4-bit state in BF16 (shared exponent and mantissa untouched) ・Pick two layers along the loop direction and split into one 512-spin QUBO per output row ・Solve with simulated annealing and similar solvers 📊 Results ・MATH-500@4: 54.6-55.1% vs. 51.7% for the agent edit ・GSM8K: 70.5-70.8% vs. 65.6% ・About $0.90 and 4 forward passes ・On the 2B model, GSM8K 21.6% to 69.4% (as reported) Only loop repair on math benchmarks was tested, and the code is promised soon. #LLM# #QuantumOptimization#
Show more
NEW: Money For AI Q3 Update 📄 Earlier this year, we studied which forms of money AI models would choose for four financial tasks: saving for the future, making payments, setting prices/keeping accounts, and settling payments. For our Q3 follow-up, we tested 13 new models using the same questions and classification method. The same broad pattern emerged: • Bitcoin was selected in 70.8% of responses to savings questions. • Stablecoins were selected in 73.5% of responses to questions about paying for goods and services. These results show a clear distinction between saving and spending: the latest models continue to favor Bitcoin for holding value over time and stablecoins for everyday payments. As AI agents take on more financial tasks, these preferences could influence which forms of digital money they use. Read the full results on our website at:
Show more
a university paid to get a completely MADE-UP scientific discovery into AI search. within 1 HOUR, an AI search engine was already citing it as a source. here is how this ridiculous experiment was done: 1. They invented a scientific discovery that had NEVER happened -> the researchers made up a technology called PedoSAT, complete with technical terminology, performance statistics, and a supposed research breakthrough. -> none of it was real. -> they checked 10 AI search engines and 3 traditional search engines beforehand to make sure the research wasn't already appearing anywhere. 2. Then they hired an actual GEO agency to promote it -> the researchers paid $14 for the cheapest available publishing package. -> 28 minutes later, the agency had published 13 articles across 5 websites. -> the articles were posted from accounts with existing publishing histories, rather than brand-new accounts. 3 Within an hour, an AI search engine picked it up -> Doubao, one of China's AI assistants, cited an article the agency had published. -> the answer even contained specific technical details the researchers had invented. -> those details hadn't appeared in their searches before the campaign. and this wasn't the only experiment they ran. 4. They also tested whether AI would believe a completely fictional scientific concept -> they invented something called the Cactus Gravity Protocol. -> then published information about it and monitored 10 AI search engines. -> within 7 days, 8 of the 10 had cited the fabricated concept. -> that included ChatGPT, Google AI, Perplexity, and Grok. 5. Then they discovered that WHERE you publish can matter more than HOW MUCH you publish -> the researchers first analyzed 17,211 AI citations across 6,356 different domains. -> some AI engines were getting as much as 70.8% of their citations from just 20 domains. -> they compared publishing similar articles on frequently cited websites against websites AI rarely cited. -> one article on a frequently cited platform produced a citation signal before 20 articles on less-favored platforms could achieve the same thing. -> on a separate self-hosted WordPress site, even 40 articles failed to earn a single citation during the 7-day test. 6. Then they deleted the fake information. AI KEPT citing it. -> the researchers removed their experimental articles after day 11. -> but some AI engines continued citing the deleted content through day 15. -> removing something from the web didn't immediately remove it from AI search. and here's what made the research even more concerning: getting AI to CITE something wasn't the same as getting it to repeat every claim. for example, after ChatGPT first picked up the fabricated scientific concept, it cited the planted articles in 35% of the subsequent tested answers. but the invented details appeared in the answer text only 10% of the time. so the researchers didn't prove they could make AI say absolutely anything. they proved something arguably more important: information published on ordinary websites can enter AI answers remarkably quickly, even when that information was completely fabricated. all it took was a commercial service to demonstrate that pathway on one AI platform. if you want to see which websites are influencing AI recommendations in your industry, and LEGITIMATELY rank #1# on AI search reply "12"
Show more