Sabitlenmiş Tweet
Know Trend LLC
339 posts
Know Trend LLC
@knowtrend_ai
Super-charged CTRL + F for public companies. Discover emerging trends for stocks and sectors
New York, NY Katılım Mayıs 2025
17 Takip Edilen123 Takipçiler
@PolycarpFX love your use of LLMs to create custom tools at scale ... but ZN+ZB+UB positioning as measured by derivatives reporting is very lopsided this time, "like never before"
English
Mid Term Years and the 10Y tend to follow a pattern of topping right into this window $TLT
English
While bond prices have struggled to sustain a meaningful rally, positioning has built a massive structural base that suggests either a catastrophic consensus error or an impending, violent move lower in long-end yields as the trade becomes saturated.
knowtrend.ai/cot/2026-04-21
English
DeepSeek is the #1 global keyword since 2016. It is a Chinese revolution. American analysts, investors, and management are obsessed with LLMs but almost entirely oblivious to DeepSeek's growth in China. Welcome to the 21st century
Compute King@Compute_King
DeepSeek V4传新进展：转向华为昇腾生态，由CUDA迁移至CANN；API定价极低；泄露测试显示性能表现强劲 DeepSeek V4 Rumored Progress: Shifts to Huawei Ascend 950PR, Migrates from CUDA to CANN; API Pricing Extremely Low; Leaked Tests Show Strong Performance 笔者注：和大家一样V4关注漫长时间了，笔者的多位业内好友今年较早就开始提示，但小鲸鱼创始人梁文峰对性能要求比较高，对整体节奏还是掌握得比较稳。最近，据多家媒体和知情人士消息，梁文锋已在内部沟通中明确新一代旗舰大模型V4将于4月下旬正式发布。这一消息疑似终结了此前多次跳票带来的不确定性。从灰度测试，架构创新到算力生态彻底转向，V4正从"长期预期"进入可验证的交付阶段。 生态跃迁：从CUDA到CANN的"去英伟达化" 如果大家看到前几天老黄接受访谈的话，也知道老黄关注的也是DeepSeek V4的战略转向；这并非单纯的技术升级，而是一场算力生态的重构。 据报道，该模型将首次全面适配华为昇腾等国产AI芯片，成为首个在全链路算力上彻底摆脱英伟达依赖的万亿级大模型。为实现这一跨越，DeepSeek工程团队投入几个月的时间，重写了核心代码，完成从英伟达CUDA生态向华为CANN架构的底层迁移。 迁移效果相当显著。据传，内部的测试数据显示，DeepSeek V4在昇腾950PR上的推理速度较初期版本提升35倍，单卡推理性能达到英伟达特供版H20芯片的2.87倍，能耗降低40%。华为CANN框架已实现超95%的CUDA代码兼容性，辅以一键迁移工具，原本需要数月完成的代码重构如今可缩短至小时级别。 老黄在几天前的采访中直言，如果DeepSeek V4与华为芯片深度适配，将削弱美国在AI技术生态的壁垒，这对美国企业来说将是"可怕的结果"。 产业连锁反应已在加速。阿里巴巴，字节跳动和腾讯等科技巨头已提前向华为下单数十万颗昇腾芯片，需求激增推动该芯片价格上涨约20%。据IDC数据，2025年中国AI加速服务器市场中，本土芯片厂商市占率已攀升至约41%，英伟达份额降至约55%。 性能突破：万亿MoE架构与百万级上下文 传出的技术参数：V4采用混合专家架构，总参数量约1万亿（业界推测最高可达1.6万亿），每次推理激活约370亿参数，在规模大幅跃升的同时通过精巧的路由机制将推理成本控制在与前代持平的极低水平。上下文窗口直接拉升至100万Token级别，原生支持多模态输入（文本与图像），补齐了与第一梯队竞争的关键短板。 据传，在架构创新上：V4有望引入Mega MoE与HyperConnection两大新机制。Mega MoE被视为MoE技术路径的实质性跃迁，有望将单次前向推理中激活的专家数量从V3的256个显著提升至数千量级，大幅增强模型表达能力与推理质量。此外，官方公开的Engram条件记忆架构使模型能在超长上下文中高效检索信息，非常适合超长文档处理，代码库分析等企业级应用场景。 基准测试表现方面：今天有传一张V4 Technical Report的图片，里面的测试结果显著。当然，更可靠的性能数据需等待官方正式发布后确认。 定价策略：极致性价比的延续 据传，V4延续了极致性价比策略。据传播的API定价信息，V4输入成本为每百万Token 0.3美元，输出成本为0.5美元，启用缓存后输入成本可进一步降至每百万0.03美元。 如果有个用户每个月处理10亿Token计算，V4无缓存费用约300美元，启用缓存后更低到30美元附近。相比之下，OpenAI ChatGPT的旗舰模型同等Token量可能需要~2500美元，而Anthro[ic的Claude Opus 4.6则需约5000美元，成本差距巨大。 这一价格策略与其MoE架构设计密切相关，尽管总参数量跃升至万亿级，但每个Token仅激活约370亿参数，实际推理成本得以与V3保持基本持平。 在笔者看来：DeepSeek V4的战略意义已远不止于技术参数的迭代升级。如果说R1证明了"中国能做出世界级模型"，那么V4正在回答一个更核心的问题：在被封锁的环境中，中国AI能否建立起从芯片，框架到模型，应用的完整自主闭环？从CUDA到CANN的生态迁移，从英伟达到华为昇腾的算力转向，V4的每一步都在为国产AI产业链的自主可控铺设可验证的道路。
English
And to top it off, the next question was about AI disruption hype somehow applying to U.S. Bancorp.
English
Last week's $USB earnings call demonstrates how little understanding there is of bank exposures to private credit back-leverage, with mass confusion over private credit, private equity, and capital call financing
English
Macro funds have indeed bought the duration dip the entire way down since 2022
*Walter Bloomberg@DeItaone
IMF WARNS OF 'FORCED SELLING' OF LEVERAGED BOND HOLDINGS IF RATES RISE
English
FYI they are whitewashing clandestine government remote viewing findings through these organizations. But of course, only the interpretations that suit them
Valerie Anne Smith@ValerieAnne1970
The World Economic Forum has your future planned... "The rich will be able to travel, but the poor will need to use virtual reality headsets to travel to the same place, but from their own couch." ~Andrew Ross Sorkin
English
While it may seem to outsiders that Citrini is stepping outside his area of expertise by commenting on the Iran war, the original fan-favorite Citrini smallcap stock - American Superconductor $AMSC - has always been a Strait of Hormuz play
Citrini@citrini
The difference in Strait activity from when #3 first arrived / began observing the strait about 4-5 days ago to today is stark. Traffic has meaningfully picked up - there are still “dark” runs and ships transiting without AIS turned on, but there’s a lot more going along the coast of Oman. At least 15 ships have crossed, including at least 3 VLCCs. When we arrived, virtually none were going through. Then a trickle through the Qeshm channel. It’s meaningful now, could be talking low double digit percent of pre-conflict volume. Meanwhile, expectations for a US operation involving “boots on the ground” within the next week or two are still high among locals. When analyst #3 first got to the strait we were hopeful we’d get a clear cut answer - bullish or bearish, open or closed, war or deal. It soon became clear that was the wrong framework through which to view this trip. On the same day that we learned it was the broad expectation of nearly everyone in the region - from locals to informed parties - that US ground troops would be launching an operation (“boots on the ground!”), we also observed multiple ships beginning to cross the strait. Soon they weren’t just limited to the Qeshm channel. It is clear to us that this isn’t as much a story in isolation as it is a story about the multipolar world and how it’s rapidly changing from what we’re used to. It’s a story about parallel warfare and diplomacy, US promises for the “Stone Age” in tandem with Allies’ seeking new venues for negotiation, and the changing global climate that necessitates this balance. Before, it would have been unlikely to imagine a world where Japan, the EU and other US allies were negotiating with a country the US is directly in conflict in to secure passage and work on agreements while the US still maintained footing for an escalation of kinetic warfare. Now, that’s simply how the world works. These countries must deal with the issues imposed, as the US won’t be sorting it out on their behalf. It’s undeniable the world is very different now and viewing this conflict through the lens of the past 50 years is a flawed approach. On Sunday, we will release our report that covers in depth what we’ve learned, how complex the situation is and what investment implications and nuances exist that have longer term implications than the next 100 points on SPX.
English
They're trying to make "agentic payments" the PayPal moment of the AI bubble
But with over 80% of companies mentioning "agentic payments" down YTD, it's probably too late for this narrative to take hold
a16z@a16z
"The grand unification of AI and crypto is about to happen." — Marc Andreessen "It's now obvious that AI agents are going to need money—it's already happening." "It's that William Gibson quote: the future is already here, it just isn't distributed yet." "My friends who are the most aggressive users of OpenClaw have given their Claws bank accounts and credit cards. And not only have they done it—it's obvious that they needed to do it, because it's obvious that they needed to be able to spend money on their behalf." "And by the way, OpenClaw—if you don't give it a bank account, it's just going to break into your bank account anyway and take your money. So you might as well do it." @pmarca with @latentspacepod
English
"Nano Banana" corporate mentions have so far gone nowhere. Without an explosion in usage this quarter, we believe it will be safe to classify imagegen (and videogen) models as extremely expensive toys with limited uptake in industry
$ADBE $GOOGL
English
@fallacyalarm Love to see you do a deep dive on $UPBD if you have time
English
ngl, this is a pretty nasty guidance for $gsy.to. 18% charge off rate for the year. the normal level is 9% and they guided 13% just 2 weeks ago in their profit warning.
This likely translates into a loss of $200m+, more than 20% of equity book value. EPS will likely be -$10 or worse. Markets won't like this at all. Communication strategy makes it even worse. Stock is a no touch for the foreseeable future.
English
@TTCSteve Refreshing take. $AUDUSD derivatives positioning is at strange highs
$6bn is not necessarily a lot of notional, nor is strong positioning always contrarian
In context of your chart, there is probably some narrative going around Wall Street, and the unwind could be violent...
English
Currencies: $AUDUSD
It's at the first level among many below, and capitulation occurs a few cents lower yet, though risk builds. If it can't build traction during a DXY wave-2 ahead, that's the sign of true weakness. In Albo we trust? Not I...
English
A unique view into CFTC CoT reports is to aggregate data across asset classes and entities
Aggregate USD net positioning (excluding dealers + levered funds) is significantly elevated since the 2022 bottom: now $2.2 trillion vs. $300 billion at that time
English
@Namzes_G @m0d3rnpr0bl3ms Probably not as good as experts but rapid-fire AI-generated primer using earnings calls for context suggests $AGRO as most interesting name
knowtrend.ai/narrative/fert…
English
@m0d3rnpr0bl3ms Still have it. Just back to base. My guys told me MOS is the wrong pick though 🤷🏻♂️
English
$FMC agri business company. Priced for death trading near GFC lows /single digit PE / messy balance sheet / -91% from ‘22 ATH
-LT cycles were due a low in Nov 2025
-20W cycle low (red) was due mid-Mar
-Daily working on a breakout above ST MAs. $22 first ST target, then 200 DMA test or possible gap fill.
-Similar to my earlier analysis of $LYB chemicals from Jan, I took a long today but on weekly want to see more basing and expecting a retest around July to solidify weekly structure.
English
And bragging about "vibe coding platforms" for pharmaceutical companies
English
Now the company is navel gazing at "future business models of the internet"
English
Since 2018, $NET has burned a cumulative -$1.1bn in free cash flow on +$7.6bn of cumulative revenue
Ryan Daigle@rwdaigle
Cloudflare agents are about to get the ability to provision a git-backed filesystem in their working environment. Use git… on a filesystem… on an instantly provisioned durable object… at the edge I hope you see how powerful this primitive is becoming!
English
"Agents!"
~90% of companies mentioning AI agents are down YTD
Quinn Nelson@SnazzyLabs
When even Lex Fridman thinks you said something dumb, boy, you really botched it.
English