今日科技推文精选 - 2026年4月14日

2026年4月14日科技每日简报

Today's top tech conversations are led by @eladgil, whose post about 'Amazing. Perplexity hits $500M...' garnered the highest engagement. Key themes trending across the top stories include https, reasoning, claude, every, model. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. eladgil (Group Score: 217.2 | Individual: 35.3)

Cluster: 7 tweets | Engagement: 187 (Avg: 124) | Type: Tech

Amazing. Perplexity hits $500M in revenue (up from$ 100M) 🔥\n\nQT @AravSrinivas: Perplexity started as a small business tool for ourselves. We had 4 people and no revenue with AI at our fingertips.

The pivot to Computer is actually a full circle. Founders are using it to grow companies that matter to the economy and their communities.

It’s rewarding to see it now powering small businesses and startups in big ways.

Perplexity is still a startup. We just 5X’ed revenue from $100M to$ 500M with only 34% growth in team size. 2x revenue growth in 2026 with same small team. And we’re just warming up.

Everyone here works at a small business, and everything we build is for people who build.

See 6 related tweets

@testingcatalog: Perplexity grew from $100M to$ 500M with only 34% growth in team size and fully pivoted to Perplexit...
@rohanpaul_ai: We are entering a new era of impactful small businesses and entrepreneurship where revenue rises wa...
@jason: Super impressive product\n\nQT @AravSrinivas: Perplexity started as a small business tool for oursel...
@cryptopunk7213: damn so perplexity just 5x'd revenue to $500M, doubling 2026 revenue growth and its only april...

e...

@Parul_Gautam7: RT @AravSrinivas: Perplexity started as a small business tool for ourselves. We had 4 people and no ...

2. rohanpaul_ai (Group Score: 145.1 | Individual: 25.8)

Cluster: 8 tweets | Engagement: 28 (Avg: 56) | Type: Tech

Microsoft is building its own OpenClaw-like features into its existing Microsoft 365 Copilot tool. This is called Build, and that's in June It would gear toward enterprise customers, with better security controls. ~ The Information https://t.co/ciBcAd7yCf

See 7 related tweets

@kimmonismus: Microsoft is testing OpenClaw-style AI agents to evolve Microsoft 365 Copilot into an always-on assi...
@wallstengine: Microsoft $MSFT is developing new Copilot features inspired by OpenClaw, including always-on AI agen...
@StockSavvyShay: $MSFT is developing new Copilot features inspired by OpenClaw including always-on AI agents that can...
@aaronpholmes: New: Inside Microsoft's plan to capitalize on the OpenClaw craze as it aims to improve Copilot: http...
@Techmeme: Microsoft says it is "exploring the potential of technologies like OpenClaw in an enterprise context...

3. ohxiyu (Group Score: 144.4 | Individual: 35.6)

Cluster: 6 tweets | Engagement: 89 (Avg: 28) | Type: Tech

Claude Certified Architect 认证，考过的人都说"这不是看看文档就能过的"。

考试结构：60 道选择题，120 分钟，闭卷，不能用 AI 辅助。满分 1000，720 分过。题目不是背概念，是随机抽 4 个生产场景丢给你做架构决策。

五个考试领域和权重：

Agentic 架构与多智能体编排：27%
Tool 设计与 MCP 集成：18%
Claude Code 配置与工作流：20%
Prompt 工程与结构化输出：20%
上下文管理与可靠性：15%

难度大概 301 级别，官方建议至少 6 个月 Claude API + Claude Code 的实操经验。有人考了 985/1000，社区评价是"奖励真正在生产环境里用 Claude 搭过东西的人"。

备考资源全免费：https://t.co/23tOuJiBK4 上有完整学习指南、25 道练习题、12 周学习计划、反模式速查表。 Anthropic Academy（https://t.co/KEd9NIbGES）13 门课也免费。GitHub 上已经有好几个社区整理的备考仓库。

费用：正常考 $99 一次。Claude Partner Network 成员前 5000 人免费（Partner Network 本身免费加入）。

Anthropic 花 1 亿美元搞这套体系，不是为了卖考试费——是要在企业端建一个"会用 Claude 的人"的人才池。认证免费 + 备考免费 + 生态开放，跟 AWS 早年推 SA 认证抢市场一个路数。

See 5 related tweets

@ohxiyu: Anthropic 的 Claude Partner Network 权益

入门就给： Partner Portal 门户，里面有现成的销售 playbook、联合营销素材、Anthropic Ac...

@ohxiyu: 我拉了个小群，目前还有几个名额。企业合作伙伴能给的权益都给群里的小伙伴。没事大家多交流交流ai。

需要你 1.在anthropic academy确认完成这四节 Introduction to ag...

@GitHub_Daily: 有不少开发者，还在使用 IntelliJ IDEA 编辑器，想要用 Claude Code 或 Codex 模型辅助编码。

可以安装 CC GUI 这个开源插件，直接提供 Claude Code 和...

@jesselaunz: 我的django 项目体感opus不但没有降智，反而速度更快了

5个projects 10来分钟就搞定了\n\nQT @ShenHuang_: 说句得罪人的：

觉得 Opus 降智的人，大部分是把...

@leewaytor: AI 自进化的第一步不是换模型 — 是让它自己写 harness。Stanford + DSPy 作者用数据证了。

Meta-Harness 这篇论文做了一件事：不动模型权重，让 AI 自己迭代"该...

4. FT (Group Score: 137.8 | Individual: 27.8)

Cluster: 7 tweets | Engagement: 241 (Avg: 241) | Type: Tech

FT Exclusive: Meta is building an artificial intelligence version of Mark Zuckerberg that can engage with employees in his stead, as part of a broader push to remake the Big Tech company around AI. https://t.co/kE6SUyJWQm https://t.co/gH061u30Do

See 6 related tweets

@Techmeme: Sources: Meta is building photorealistic, AI-powered 3D characters; Zuckerberg helped train and test...
@Cointelegraph: ⚡️ JUST IN: Meta is building an AI version of Mark Zuckerberg, trained on his mannerisms and tone to...
@wallstengine: FT: $META is developing photorealistic AI-powered 3D characters, including an AI version of Mark Zuc...
@StockSavvyShay: $META is developing photorealistic AI-powered 3D characters including an AI version of Mark Zuckerbe...
@FT: Meta builds AI version of Mark Zuckerberg to interact with staff https://t.co/j6ChrXUtvC...

5. swyx (Group Score: 116.6 | Individual: 33.5)

Cluster: 4 tweets | Engagement: 164 (Avg: 19) | Type: Tech

RT @nicoalbanese10: 3 months ago I started building a coding agent that runs in the cloud.

It's since written every line of code I've shipped, including itself.

Today, I'm open sourcing it. Introducing Open Agents. https://t.co/puuqWfMKej

See 3 related tweets

@badlogicgames: neat!\n\nQT @nicoalbanese10: 3 months ago I started building a coding agent that runs in the cloud. ...
@waghnakh_21: never a single boring day on X, cloud coding agent by @nicoalbanese10\n\nQT @nicoalbanese10: 3 month...
@jiayuan_jy: Very impressive work!\n\nQT @nicoalbanese10: 3 months ago I started building a coding agent that run...

6. tryramp (Group Score: 115.8 | Individual: 37.3)

Cluster: 5 tweets | Engagement: 278 (Avg: 56) | Type: Tech

RT @eglyman: 99% of Ramp uses ai daily. but we noticed most people were stuck — not because the models weren't good enough, but because the setup was too painful and unintuitive for most. terminal configs, mcp servers, everyone figuring it out alone.

so we built Glass. every employee gets a fully configured ai workspace on day one — integrations connected via sso, a marketplace of 350+ reusable skills built by colleagues, persistent memory, scheduled automations. when one person on a team figures out a better workflow, everyone on that team gets it and gets more productive.

the companies that make every employee effective with ai will compound advantages their competitors can't match. most are waiting for vendors to solve this. we decided to own it.

See 4 related tweets

@noahzweben: Really cool use case of Claude Agent SDK\n\nQT @eglyman: 99% of Ramp uses ai daily. but we noticed m...
@ShanuMathew93: This is how the future will look, onboard employees with AI and make it frictionless.\n\nQT @eglyman...
@ashugarg: 27% of employees say AI is replacing parts of their job. 21% say it’s enabling entirely new tasks.

...

@BusinessInsider: From entrepreneurs to software developers, usage limits on AI tools are reshaping how some workers s...

7. heyshrutimishra (Group Score: 106.4 | Individual: 38.7)

Cluster: 3 tweets | Engagement: 51 (Avg: 38) | Type: Tech

CHINA JUST SOLVED THE PROBLEM THAT'S BEEN BREAKING ROBOT AI FOR A DECADE.

and the fix wasn't a smarter model.

for years, every robot AI failure got the same diagnosis.

the model isn't smart enough.

so everyone scaled intelligence. bigger models. more parameters. better reasoning.

@AGIBOTofficial asked a different question: what if the reasoning was never the problem?

there's a gap that runs through every traditional robot AI system.

reasoning on one side & motor commands on the other.

the brain decides but the body executes something different, because thinking and moving were never actually connected.

GO-2 fixes this by reasoning INSIDE the action space, not above it.

before moving, it runs a complete mental simulation of every step - like a basketball player mentally tracing the arc of a shot before releasing the ball.

watch the demo and you'll see exactly what this means.

the robot works through a task queue autonomously. classify toiletries. upright the drink bottle. place headphones in the leather box.

mid-execution, a new instruction drops: "my phone's missing. help me find it."

it doesn't pause. doesn't reset. it processes the new task and keeps moving.

that's not a scripted sequence. that's real-time instruction following on top of an active task queue.

that one architectural change is where the numbers come from.

#1 on LIBERO across Spatial, Object, Goal, and Long tasks → 98.5% average success 86.6% zero-shot accuracy in active disturbance environments 47.4 on VLABench → best-in-class on objects and textures it's never seen before 82.9% success trained on simulation only, tested on real hardware

sim-to-real is the graveyard of robotics research. models trained in simulation collapse the moment they touch the real world.

82.9% means that graveyard just got a lot smaller.

it holds because of how GO-2 trains. deliberately fed imperfect reasoning conditions, then trained to execute robustly anyway. not a researcher assumption. a design decision from a team that ships hardware and knows exactly what breaks.

then there's the infrastructure layer.

Genie Studio. fleet-wide data collection. cloud training. online post-training in live environments.

10x improvement in training efficiency. task startup reduced to minutes. 2-4x better success rates with 50%+ less data.

the model gets smarter every time a robot fails in the field.

this isn't a benchmark story. it's a compounding moat.

dual CVPR 2026 + ACL 2026 acceptance. computer vision AND natural language processing. top conferences. simultaneously.

that doesn't happen with incremental research.

the US-China robotics race has been framed as a compute race. a model quality race.

it was always an execution race.

the robot that wins won't be the smartest one in the lab.

it'll be the most reliable one on the floor.

full breakdown: https://t.co/4NfjeWMkyN

is execution reliability the real bottleneck, or are we still underestimating how far reasoning needs to go?\n\nQT @AGIBOTofficial: #AGIBOTAIWeek Day3: Introducing GO-2 - our next-generation foundation model for embodied AI, built to unify reasoning and action. To truly bridge “thinking” and “doing,” embodied AI must solve two challenges at once: • generate executable action plans through deep spatial reasoning • deliver stable execution in real-world environments GO-2 tackles both with a comprehensive architecture: Action Chain-of-Thought for action reasoning, and an Asynchronous Dual-System for robust execution.
#AGIBOT #AGIBOTAIWeek #Foundation #model #EmbodiedAI

See 2 related tweets

@heyshrutimishra: Every American robotics lab has been trying to build a smarter robot.

China just won by building a ...

@heyshrutimishra: RT @heyshrutimishra: CHINA JUST SOLVED THE PROBLEM THAT'S BEEN BREAKING ROBOT AI FOR A DECADE.

and ...

8. StockSavvyShay (Group Score: 102.4 | Individual: 36.6)

Cluster: 3 tweets | Engagement: 743 (Avg: 468) | Type: Tech

If you invested $10K in$ NBIS a year ago, you would have ~$76K 😳 https://t.co/TRLdolXwvb\n\nQT @StockSavvyShay: The neocloud category may be the most misunderstood corner of the AI trade because the market still treats these names as one uniform GPU-hours bet when they are actually very different business models:

$NBIS (Cloud Utility for the Agentic AI Age)

$NVDA just chose Nebius as an architecture partner for the agentic AI era by co-designing AI factories with them, and the Rubin GPU access that comes with this partnership means Nebius gets the next-generation inference stack before almost anyone else in the market. At a$ 28B market cap, a 5GW power target and Nvidia’s engineering team embedded in the stack.. this is my favorite name in the neocloud category.

$IREN (Energy-to-Compute Engine of the AI Era)

The dilution fear is real but the market is misreading it. IREN is not diluting to survive but diluting to scale into a $3.7B ARR target and the$ 9.3B in funding already secured through customer prepayments and GPU financing means the $6B ATM is optionality capital. The real bottleneck in AI infrastructure right now is power and IREN controls ~4.5GW of secured capacity while needing only ~500MW to support its ARR target by year-end. That 10x ratio of power capacity to near-term need is something no competitor can replicate quickly.

$CIFR (Landlord of the AI Utility Era)

Cipher is not a pure neocloud but is a hyperscale infrastructure landlord signing decade-long leases to $AMZN AWS and$ GOOGL while they fill the shells with compute. The AWS lease alone is expected to generate ~$700M in average annualized NOI for the next decade at nearly 100% NOI margins. Power-rich land is the scarcest resource in AI infrastructure and Cipher controls it with 600MW fully contracted, both facilities fully funded through non-recourse fixed-rate project debt and a 3.4GW development pipeline.

$CRWV (The Fragile Giant)

CoreWeave’s demand backlog and revenue growth are very real but none of that matters if the capital markets close for even one quarter. Interest expense hit $388M in Q4 and management guided Q1 2026 interest expense to ~$ 550M which implies an annualized run rate above $2B before a single new data center comes online. The bull case requires capital markets to stay open, rates to cooperate, hyperscalers to honor take-or-pay contracts in full and construction to stay on time. That is a lot of dependencies in a macro environment where oil is approaching$ 100 and private credit is already showing signs of stress.

See 2 related tweets

@ttunguz: The Beginning of Scarcity in AI

For the first time since the 2000s, technology companies are confro...

@HedgieMarkets: 🦔The AI compute crunch is becoming a real constraint on the industry. Anthropic's Claude API has had...

9. KarthiDreamr (Group Score: 101.6 | Individual: 29.5)

Cluster: 5 tweets | Engagement: 608 (Avg: 118) | Type: Tech

RT @AISecurityInst: We conducted cyber evaluations of Claude Mythos Preview and found that it is the first model to complete an AISI cyber range end-to-end. 🧵 https://t.co/gd9hi0Ve55

See 4 related tweets

@AISafetyMemes: UK government's AISI: "Our results show Claude Mythos represents a step up over previous frontier mo...
@scaling01: Mythos just one-shotted this cyber eval that takes humans ~20 hours to complete https://t.co/ArJkukj...
@scaling01: after ~10 million tokens Mythos is much more efficient than other models

it reaches the same perfor...

@wallstengine: AISI: Claude Mythos Preview posted a 73% success rate on expert-level capture-the-flag cybersecurity...

10. cryptopunk7213 (Group Score: 96.0 | Individual: 32.9)

Cluster: 4 tweets | Engagement: 621 (Avg: 307) | Type: Tech

biggest takeaway from the leaked openai memo is they're convinced anthropic fucked up. also amazon is killing it, they now own major stakes in the top 2 ai labs, from the memo:

Anthropic made a "strategic misstep to not acquire enough compute" and are "operating on a meaningfully smaller curve." oof

they also dunked on microsoft: they've "limited our ability to meet enterprises where they are"

openai is now choosing amazon as their main compute partner instead of microsoft
amazon just invested $50B in openai, the aim is to serve all AWS partners chatgpt via Bedrock platform
openai is targeting 30GW by 2030 vs. anthropics 8GW

the thing no ones talking about tho is amazon owns 20% of anthropic. its literally in their interest to serve them compute.

make no mistake, the next few months will be determined by whoever has the most operational compute

both anthropic and oai are racing to sign major deals.

See 3 related tweets

@wallstengine: OpenAI is leaning further into AWS for enterprise distribution, with CRO Denise Dresser reportedly t...
@rohanpaul_ai: Theverge: OpenAI just told employees in an internal memo that winning the next phase of AI will dep...
@StockSavvyShay: OpenAI is reportedly leaning further into $AMZN for enterprise distribution with its CRO saying$ MSF...

11. SawyerMerritt (Group Score: 95.9 | Individual: 35.2)

Cluster: 3 tweets | Engagement: 3241 (Avg: 2518) | Type: Tech

Tesla's Spring software update has officially been unveiled.

Here's what's new: • New Self-Driving App: Lets you subscribe with a single tap & learn about how to activate and use the feature. You can also view ongoing stats. AI4 hardware cars only. • Blind Spot Warning Accent Lights: Accent lights now turn red when an object is in your blind spot and your turn signal is engaged, or when an approaching object is detected while parked. • Launch Grok by saying "Hey Grok." You can also set location-based reminders, like "remind me to pick up milk when I'm near home." Say “goodbye” to dismiss Grok. • Pet Mode: Choose from dog, cat or hedgehog to display when Pet Mode is active. Also optionally customize the screen with your pet's name by going to Controls > Display > Customize Pet Mode • Car Visualization (New Model 3 & Y): A new park scene environment & higher quality car visualization. • More Trips: Create multiple trips to track energy stats across drives. Quickly access your trip’s consumption by swiping left on the media player. • Sketchpad: Now supports stickers & emojis. You can also save your sketches to access them in the Mobile App & share. • Model S & X owners can now personalize their Tesla avatar with window tints, custom wraps & license plates. • Immersive Sound (New Model 3 & Y w/ premium audio): Premium Immersive Sound uses advanced sound extraction to place the listener in front of a detailed soundstage within an Immersive space. This enhancement works with all streaming sources. • Your Tesla can now automatically install downloaded software updates overnight. When enabled, updates will install while your Tesla is parked and not in use. • Weather maps now show snow and rain with improved colors that make it easier to distinguish between precipitation types. You can also view the past hour of weather data to see how conditions have been changing along your route. • Recent Dashcam footage can now extend up to 24 hours & you can save any clip for permanent storage on your Tesla

Other Improvements • Swipe right on tracks in Apple Music or Spotify to add them to the queue. • Passengers in rear seats can now view and interact with maps on the rear display while your Tesla navigates a route\n\nQT @Tesla: https://t.co/FrWpKzTkjj

See 2 related tweets

@niccruzpatane: Tesla 2026 Spring Update has arrived! There are some cool new features:

• New, more comprehensive s...

@SawyerMerritt: Here is Tesla's new Self-Driving app that will be rolling out shortly to all AI4 Teslas where FSD is...

12. chhddavid (Group Score: 93.3 | Individual: 32.1)

Cluster: 3 tweets | Engagement: 11 (Avg: 54) | Type: Tech

BREAKING: As of today, @claudeai can build full-stack apps.

Claude got a huge update today inside Shipper, and I'm proud to say it changes vibe coding forever. Claude Code Opus 4.6 can now turn ideas into full-stack apps in minutes.

We just launched Shipper a package for Claude to:

→ Build anything for you: website, mobile app etc → Code, design, launch, monetize → Do email marketing → Auto-translate the app → Self-run it in the long run

Claude's most powerful models can now do all of that from a <10 word prompt, for as low as $0.28/app... And it takes minutes!

Simply go to shipper, then ask Claude to "create a language-learning app" or "build a SaaS that charges $29/mo"!

To celebrate the launch, we're giving away free credits randomly. Repost and comment "shipper" and we'll pick the winners.

See 2 related tweets

@chhddavid: so you're saying Claude Opus 4.6 just killed Lovable...
builds fullstack apps
has a chatting UI...
@chhddavid: this is scary...\n\nQT @chddaniel: Introducing Shipper

From today on, Claude Code Opus 4.6 turns id...

13. rohanpaul_ai (Group Score: 88.0 | Individual: 52.6)

Cluster: 2 tweets | Engagement: 512 (Avg: 56) | Type: Tech

RT @rohanpaul_ai: India is quietly becoming a training floor for humanoid robots, with workers filming thousands of first-person hand tasks so AI systems can learn grasping, folding, sorting, and tool use.

This story is really about how the humanoid robot boom still depends on cheap, repetitive human labor to teach machines basic physical skill.

The problem is that robots do not fail on big plans first; they fail on tiny physical details like grip angle, finger timing, slip correction, and object contact.

That kind of knowledge is hard to code and expensive to collect.

These labs capture that missing layer by putting cameras or sensors on people and recording ordinary actions as machine-readable motion examples.

The useful part is not the towel or box itself but the sequence: where the hand starts, how force changes, when fingers adjust, and how the body recovers from small mistakes.

That gives robotics teams supervised data for models that map visual input to physical actions, which is much easier than hand-coding every movement rule.

This is a story about how physical intelligence gets extracted before it gets automated.

quasa. io/media/the-hidden-hand-farms-of-india-fueling-the-ai-robot-revolution-with-human-motion

See 1 related tweets

@tunguz: It may come as a surprise to most people, but until recently robotics was NOT the field that relied ...

14. rohanpaul_ai (Group Score: 87.4 | Individual: 31.4)

Cluster: 3 tweets | Engagement: 28 (Avg: 56) | Type: Tech

HeyGen just turned AI video generation into a CLI workflow, so an agent can go from script to avatar, then render the video and ship it without leaving the terminal.

The great point here is, once video is exposed as a command, any research agent, reporting bot, or automation stack can treat video as just another output format.

The useful idea here is that video-generation can now sit inside cron jobs, pipelines, internal tools, and autonomous agents that move from script to avatar to finished asset without a human stitching the steps together.\n\nQT @HeyGen: Your AI agent can now generate and ship videos.

HeyGen CLI is now live.

Run one command and your agent handles it all: script → avatar creation → video → delivery

All from the terminal. Just your agent and the CLI.

RT + Comment “CLI” and we’ll DM API credits (must follow) https://t.co/Q5m9ISTog0

See 2 related tweets

@Origin_AI_01: Every agent I run ends in markdown I barely read.

This turns output into something you can actually...

@Origin_AI_01: RT @HeyGen: Your AI agent can now generate and ship videos.

HeyGen CLI is now live.

Run one comman...

15. jukan05 (Group Score: 85.0 | Individual: 29.0)

Cluster: 3 tweets | Engagement: 121 (Avg: 244) | Type: Tech

Wow, when I first mentioned this, it was under HK$60. It’s gone up a lot since then. https://t.co/WVpieMYD5s\n\nQT @jukan05: Samsung Electronics Evaluates ASMPT for HBM TC Bonders… Supply Chain Diversification Continues

Samsung Electronics continues to pursue diversification of its supply chain for thermocompression (TC) bonding equipment, a critical component in high-bandwidth memory (HBM) production. It has been learned that the company recently completed a demo test with overseas semiconductor equipment firm ASMPT on HBM TC bonders and has decided to move forward to the next phase of collaboration.

According to industry sources on the 13th, Samsung Electronics plans to conduct a JEP (Joint Evaluation Project) with Singapore-headquartered ASMPT for HBM TC bonders.

HBM is a type of memory in which multiple DRAM dies are vertically stacked and connected via through-silicon vias (TSVs). Samsung Electronics employs a process that places non-conductive film (NCF) between each DRAM die and bonds them using thermocompression. The equipment used in this bonding process is the TC bonder.

Samsung's in-house subsidiary SEMES has been the primary supplier of TC bonders for HBM. However, since last year, the company has been working to onboard additional domestic and international back-end equipment makers into its supply chain.

Among them, ASMPT had been conducting initial tests on HBM TC bonders with Samsung through Q1 of this year. These tests were at the R&D stage, involving demo equipment demonstrations in a lab setting.

Furthermore, the two companies have reportedly reached an agreement to pursue a JEP for the TC bonder. A JEP involves installing already-developed equipment on a customer's mass production line to verify the equipment's performance and reliability. It can be viewed as the stage where the feasibility of actual deployment is assessed.

An industry source explained, "Samsung Electronics has been continuously reviewing the diversification of its HBM TC bonder supply chain," adding that "the JEP with ASMPT is part of this strategy, and it signals that testing between the two companies is making steady progress."

That said, it remains uncertain whether Samsung will actually adopt ASMPT's HBM TC bonders in mass production. Numerous variables remain before commercialization, including the possibility that equipment performance may fall short in a production environment or that pricing negotiations may not reach a satisfactory outcome.

The timing of Samsung Electronics' adoption of hybrid bonding equipment is also expected to significantly influence TC bonder supply chain diversification. Hybrid bonding is a technology that directly connects chips without using micro bumps between DRAM dies, as in conventional processes. It is attracting attention as a next-generation bonding process for HBM, as it offers advantages in enhancing chip performance compared to TC bonding.

However, hybrid bonding has a high level of technical difficulty and has yet to be commercially deployed in HBM bonding. Additionally, JEDEC (the international semiconductor standards body) is pursuing measures to relax thickness specifications for next-generation HBM. If greater thickness is permitted, it becomes easier to maintain existing TC bonding technology.

Another industry source noted, "Hybrid bonding technology is not yet mature, and even when it is eventually adopted, limited approaches such as applying it to only certain layers are being discussed — meaning TC bonding technology may have more room to grow than previously expected." The source added, "Under these circumstances, Samsung Electronics will likely feel an even greater need to diversify its TC bonder supply chain."

See 2 related tweets

@zephyr_z9: Very interesting https://t.co/r6egkU1g7I\n\nQT @jukan05: Samsung Electronics Evaluates ASMPT for HBM...
@zephyr_z9: Nice ASMPT got more orders\n\nQT @jukan05: Samsung Electronics Evaluates ASMPT for HBM TC Bonders… S...

16. ns123abc (Group Score: 82.0 | Individual: 37.3)

Cluster: 3 tweets | Engagement: 1988 (Avg: 423) | Type: Tech

🚨OpenAI Chief Revenue Officer memo to employees LEAKED:

"Claude has become a religion, that's the level of that mania” "Anthropic's strategy is fear, restriction, and the idea that a small group of elites should control AI" "They made a strategic misstep to not acquire enough compute" "The market is ours to win" "But Microsoft limited our ability to meet enterprises where they are — AWS Bedrock”

OPENAI ENTERPRISE CODE RED ALERT

See 2 related tweets

@zephyr_z9: Spud will be a good model\n\nQT @ns123abc: 🚨OpenAI Chief Revenue Officer memo to employees LEAKED: ...
@zbuckholz: It’s not a religion; it’s just the best product currently available.

That’s it.

OpenAI, xAI, and t...

17. eastdakota (Group Score: 79.4 | Individual: 28.0)

Cluster: 3 tweets | Engagement: 94 (Avg: 31) | Type: Tech

RT @whoiskatrin: cloudflare sandboxes are now generally available

agents get a real computer: terminal, interpreter, live preview urls, secure credentials. it sleeps when idle and wakes on demand

they can do actual engineering work: clone a repo, run tests, fix failures, repeat

that loop is what makes engineers effective. now agents have it

https://t.co/p7ooV6R9eQ

See 2 related tweets

@Cloudflare: AI agents need more than a prompt—they need a computer. 💻

Cloudflare Sandboxes are now GA. Give you...

@CloudflareDev: Cloudflare Sandboxes is now GA. Agents need more than prompt windows. They need terminals, interpre...

18. VKazulkin (Group Score: 78.4 | Individual: 52.5)

Cluster: 2 tweets | Engagement: 3748 (Avg: 115) | Type: Tech

RT @pvergadia: 🤯BREAKING: Researchers just mathematically proved that AI layoffs will collapse the economy: and every CEO already knows it.

The AI Layoff Trap. A game theory paper from UPenn + Boston University is glaringly important!

100K+ tech layoffs in 2025. 80% of US workers exposed. And no market force can stop it.

→ Every company fires workers to cut costs → Every fired worker stops buying products → Revenue collapses across every sector → The companies that fired everyone go bankrupt

It's a Prisoner's Dilemma with math behind it. Automate and you survive short-term. Don't automate and your competitor kills you. But everyone automating destroys the demand that makes all companies viable.

UBI (universal basic income) won't fix it. Profit taxes won't fix it. The researchers found only one solution: a Pigouvian automation tax "robot tax"

The AI trap on the economy is here!

See 1 related tweets

@StefanFSchubert: “Mathematically proved”? That’s not how knowledge acquisition works.\n\nQT @pvergadia: 🤯BREAKING: Re...

19. Hadley (Group Score: 78.3 | Individual: 43.1)

Cluster: 2 tweets | Engagement: 1104 (Avg: 248) | Type: Tech

RT @levie: Another week on the road meeting with a couple dozen IT and AI leaders from large enterprises across banking, media, retail, healthcare, consulting, tech, and sports, to discuss agents in the enterprise.

Some quick takeaways:

Clear that we’re moving from chat era of AI to agents that use tools, process data, and start to execute real work in the enterprise. Complementing this, enterprises are often evolving from “let a thousand flowers bloom” approach to adoption to targeted automation efforts applied to specific areas of work and workflow.
Change management still will remain one of the biggest topics for enterprises. Most workflows aren’t setup to just drop agents directly in, and enterprises will need a ton of help to drive these efforts (both internally and from partners). One company has a head of AI in every business unit that roles up to a central team, just to keep all the functions coordinated.
Tokenmaxxing! Most companies operate with very strict OpEx budgets get locked in for the year ahead, so they’re going through very real trade-off discussions right now on how to budget for tokens. One company recently had an idea for a “shark tank” style way of pitching for compute budget. Others are trying to figure out how to ration compute to the best use-cases internally through some hierarchy of needs (my words not theirs).
Fixing fragmented and legacy systems remain a huge priority right now. Most enterprises are dealing with decades of either on-prem systems or systems they moved to the cloud but that still haven’t been modernized in any meaningful way. This means agents can’t easily tap into these data sources in a unified way yet, so companies are focused on how they modernize these.
Most companies are not talking about replacing jobs due to agents. The major use-cases for agents are things that the company wasn’t able to do before or couldn’t prioritize. Software upgrades, automating back office processes that were constraining other workflows, processing large amounts of documents to get new business or client insights, and so on. More emphasis on ways to make money vs. cut costs.
Headless software dominated my conversations. Enterprises need to be able to ensure all of their software works across any set of agents they choose. They will kick out vendors that don’t make this technically or economically easy.
Clear sense that it can be hard to standardize on anything right now given how fast things are moving. Blessing and a curse of the innovation curve right now - no one wants to get stuck in a paradigm that locks them into the wrong architecture. One other result of this is that companies realize they’re in a multi-agent world, which means that interoperability becomes paramount across systems.
Unanimous sense that everyone is working more than ever before. AI is not causing anyone to do less work right now, and similar to Silicon Valley people feel their teams are the busiest they’ve ever been.

One final meta observation not called out explicitly. It seems that despite Silicon Valley’s sense that AI has made hard things easy, the most powerful ways to use agents is more “technical” than prior eras of software. Skills, MCP, CLIs, etc. may be simple concepts for tech, but in the real world these are all esoteric concepts that will require technical people to help bring to life in the enterprise.

This both means diffusion will take real work and time, but also everyone’s estimation of engineering jobs is totally off. Engineers may not be “writing” software, but they will certainly be the ones to setup and operate the systems that actually automate most work in the enterprise.

See 1 related tweets

@etnshow: Aaron Levie, Box CEO, says enterprise AI strategies are leaning towards "adoption to targeted automa...

20. chris_j_paxton (Group Score: 75.8 | Individual: 33.1)

Cluster: 4 tweets | Engagement: 227 (Avg: 41) | Type: Tech

The average person in the united states does not believe they will personally benefit from AI rollouts and that will become a huge problem\n\nQT @zerohedge: The US social mood is turning dramatically negative on AI https://t.co/djm1jRHe7l

See 3 related tweets

@zerohedge: The US social mood is turning dramatically negative on AI https://t.co/djm1jRHe7l\n\nQT @zerohedge: ...
@NathanpmYoung: For clarity, I think many people who think AI will almost certainly kill us are way too confident. ...
@chris_j_paxton: RT @chris_j_paxton: The average person in the united states does not believe they will personally be...