科技推文精选 - 2026年5月17日

2026年5月18日科技简报

Today's top tech conversations are led by @HarryStebbings, whose post about 'Biggest lesson from Jensen Hua...' garnered the highest engagement. Key themes trending across the top stories include company, model, their, agents, agent. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. HarryStebbings (Group Score: 176.8 | Individual: 37.9)

Cluster: 5 tweets | Engagement: 185 (Avg: 114) | Type: Tech

Biggest lesson from Jensen Huang at Nvidia

"Jensen called me at midnight to unpack a challenge, his SLAs are insane, he responds that fast.

The lesson was your job is to fall in love with whatever the job is.

You have to find a way to bend your DNA to love what the job requires." @ShivdevRao

What are your biggest lessons from Jensen @NicolaiTang1 @sama @elonmusk @nbt\n\nQT @HarryStebbings: I have interviewed 1,000s of the world's best founders over the past decade. Few have impressed me like @ShivdevRao at @AbridgeHQ.

He navigated a brutal 5-year wilderness before exploding into one of the most dominant forces in vertical AI.

Today, Abridge is a $5.3BN powerhouse.

I sat down with Shiv to unpack exactly how he did it and condensed my notes below:

🚀 6 Lessons on Building a $5.3B Vertical AI Juggernaut

Survive Long Enough for Market Timing to Catch Up:

Abridge spent 5 years in the "wilderness" before hitting a tidal wave of adoption. When you have an absolute true north thesis, your primary job in the early days is simple: stay standing and don’t die. You must be alive when the sky finally opens up.

Pivot the Product, Never the Core Thesis:

Shiv was willing to pivot on features, go-to-market strategies, and business models. But he refused to budge on his core thesis that healthcare is ultimately powered by the spoken human signal. Die on the hill of your thesis; adapt everything else.

Target the Concentration of Scale Early:

A massive trap for healthcare and enterprise founders is staying down-market too long for "fast feedback loops". In the US, the vast majority of clinicians are concentrated within large, integrated delivery networks. Time your "YOLO shot" to go up-market the moment the market inflects.

Single biggest advice to founders on when to go up market @bhalligan @dharmesh?

Own Your Stack to Protect Your P&L and UX:

While many AI startups rely entirely on frontier systems, 40% of Abridge's model outputs are generated by in-house models. Milliseconds matter in high-stakes enterprise workflows. Building your own models gives you insane performance gains, lower latency, and ultimate control over your P&L.

When should you vs should you not build your own model @matanSF @MaxJunestrand @antonosika?

Don't Fight Foundation Models—Counter-Position Instead

If you try to fight the frontier model giants directly, you've already lost. You win by going millions of miles deep into regulated industries with proprietary datasets and workflows they can't easily replicate. Find ways to coexist and leverage their tailwinds.

Reminds me of what @bradlightcap said on his 20VC.

Move Toward the "Flat Company" Era:

With the explosion of AI agents and advanced tooling, the traditional management layer is compressing. Shiv’s latest idealistic shift is building a hyper-flat organization: fewer managers, and highly leverageable "Super ICs" who can move in lockstep and cover massive surface area.

(link in comments)

See 4 related tweets

@HarryStebbings: Why OpenAI and Anthropic doing consulting businesses is such an obvious and right move

"When you se...

@HarryStebbings: How do you know as a startup if the foundation models are going to kill you or help you

"If you are...

@HarryStebbings: What Founder Mode Truly Means

"Founder mode is about tours of duty; going to the most important thi...

@HarryStebbings: What the founder of Duolingo taught me about sacrifice

"He has that line in the S-1, the only thing...

2. IamEmily2050 (Group Score: 163.1 | Individual: 33.0)

Cluster: 6 tweets | Engagement: 103 (Avg: 108) | Type: Tech

This means the next Grok update is coming at the end of June early July just before the IPO with performance between Opus 4.5 and Sonnet 4.6 not bad and very useful for so many things, but still far from SOTA. Hopefully, they will catch up in the second half of next year.\n\nQT @elonmusk: We are improving the 0.5T Grok foundation model V8 (public version 4.3) every few days.

The 1.5T V9 just finished training (incorrectly called pre-training) and is a major upgrade. Next, we are adding the Cursor data in supplemental training (others call this mid-training), then SFT and RL. About 3 or 4 weeks to release.

This will be a banger.

See 5 related tweets

@testingcatalog: SPACEXAI 🔥: The next version of Grok, based on the 1.5T V9 base model has finished training. Looks l...
@scaling01: RT @elonmusk: We are improving the 0.5T Grok foundation model V8 (public version 4.3) every few days...
@AndrewCurran_: The new 1.5T version of Grok arrives in June. Lately, I use 4.3 increasingly for research, search, a...
@adonis_singh: im hyped for this ngl\n\nQT @elonmusk: We are improving the 0.5T Grok foundation model V8 (public ve...
@XFreeze: Elon just gave a massive update on the next generation of Grok

Grok V9 is officially coming in 3–4 ...

3. FirstSquawk (Group Score: 150.7 | Individual: 41.2)

Cluster: 6 tweets | Engagement: 2455 (Avg: 75) | Type: Tech

CITADEL KEN GRIFFIN WARNS AGENTIC AI IS AUTOMATING ELITE FINANCE JOBS IN HOURS, NOT MONTHS

See 5 related tweets

@garrytan: Ken Griffin doesn’t understand the ceiling just got raised. Some 20-something maybe reading this wil...
@TFTC21: Ken Griffin went home on a Friday "fairly depressed" after watching AI agents at Citadel do work tha...
@MTSlive: SITUATION DETECTED: Citadel’s Ken Griffin has changed his mind on AI.

“In the last few months, ther...

@peterwildeford: KEN GRIFFIN (CITADEL): "Work that we would usually do with people with masters and PhDs in finance o...
@Polymarket: RT @Polymarket: NEW: Citadel CEO Ken Griffin says AI agents are now automating “extraordinarily high...

4. akshay_pachaar (Group Score: 138.4 | Individual: 33.0)

Cluster: 5 tweets | Engagement: 790 (Avg: 935) | Type: Tech

Hermes meets SuperGrok!

xAI just made every SuperGrok subscription work inside Hermes Agent.

One browser login, no API key, no separate billing.

And it doesn't just unlock text chat with Grok 4.3.

The same OAuth token gives the agent access to:

→ Grok Text-to-Speech for spoken responses → Grok Imagine for image and video generation → x_search for real-time X/Twitter search

I just added a new X Research Agent profile to my Hermes. Now my agent watches X while I ship.

Setup takes about 60 seconds:

Available on every SuperGrok tier, no restrictions.

I wrote a full deep dive covering Hermes agent's architecture, memory system, self-evolving skills, GEPA optimization, and setting up multiple specialized agents

The article is quoted below.\n\nQT @akshay_pachaar: https://t.co/Exoyd8tB0d

See 4 related tweets

@RoundtableSpace: Hermes Agent just got SuperGrok built in.

One browser login unlocks Grok 4.3, Grok Imagine for imag...

@elonmusk: Try it out …

Improvements are landing every few days!\n\nQT @teslaownersSV: Grok Build CLI Beta can...

@WesRoth: RT @WesRoth: xAI has launched an early beta of Grok Build, an agentic command-line interface for cod...
@WesRoth: xAI now lets users connect X Premium subscriptions directly inside Hermes Agent, not just SuperGrok....

5. FundamentEdge (Group Score: 126.0 | Individual: 28.1)

Cluster: 5 tweets | Engagement: 118 (Avg: 171) | Type: Tech

Agree completely

In my opinion, learning the fundamentals of any given domain is a more important process than ever, as the "return on foundational mastery" is multiples of what it once was with the superhuman execution ability of agents.

You won't be able to "vibe code" your way to mastery in any domain, the world is too competitive and alpha is too nuanced & adaptive. The most talented individuals who integrate agents will simply raise the competitive bar, and AI slop won't keep up. This is certainly true in finance & investing.

My best advice for young people entering finance - do not bypass developing a deep foundational comprehension of the fundamentals.\n\nQT @levie: One of the best things students and colleges can do is not bail on learning and teaching the fundamentals of any given domain. AI will trick you into thinking you don’t need to go deep in a particular area, but that’s wrong.

The expert with AI is always going to be far more capable than the novice. Those that can steer AI agents properly, figure out how to evaluate their work, fix their mistakes, and incorporate their work into a workflow will always be the most potent users of these tools.

The experienced software developer that’s built and scaled complex systems using agents outrun someone just vibe coding. The designer that uses AI will build far better products and campaigns than anyone else. The banker or analyst that understands financial models will be able to pull off far more with agents.

Despite some of the rhetoric in the valley that this is less implement now, that couldn’t be further from the case. Don’t give up on going deep in your craft.

See 4 related tweets

@levie: One of the best things students and colleges can do is not bail on learning and teaching the fundame...
@dharmesh: Could not agree more.

This is precisely the advice I give my son (15 years old).

The value of know...

@ShanuMathew93: Cannot agree & emphasize this point enough. 100% agreement.

“The expert with AI is always going to ...

@anvisha: Creativity and design are some of the only tasks that won’t be saturated by LLMs.

Anything with a q...

6. TrungTPhan (Group Score: 116.6 | Individual: 39.2)

Cluster: 4 tweets | Engagement: 1233 (Avg: 338) | Type: Tech

Singapore’s Foreign Minister built a NanoClaw AI agent on a Raspberry Pi so he could “learn by doing” (he chats with it on WhatsApp for meeting prep).

We know he put in the work because he says AI agents can get expensive and will be more so when labs stop subsidizing tokens. https://t.co/wAhgtuFlK7\n\nQT @MsMelChen: Singapore’s Foreign Minister, Dr Balakrishnan casually explaining how he built his own AI agent (a 2nd brain for diplomacy) using Claude & WhatsApp integration etc. on a Raspberry Pi

“You cannot govern a technology you have only been briefed on.” 🇸🇬 https://t.co/JMk0t2FrJr

See 3 related tweets

@aiDotEngineer: RT @TrungTPhan: Singapore’s Foreign Minister built a NanoClaw AI agent on a Raspberry Pi so he could...
@MarioNawfal: 🇸🇬 Singapore’s Foreign Minister explained why he built his own AI agent using Claude on a Raspberry ...
@RoundtableSpace: SINGAPORE'S FOREIGN AFFAIRS MINISTER RUNS HIS OWN LOCAL AI STACK ON A RASPBERRY PI.

CLAUDE AGENT SD...

7. WesRoth (Group Score: 112.1 | Individual: 31.0)

Cluster: 4 tweets | Engagement: 62 (Avg: 27) | Type: Tech

Claude Mythos appearing in Google Cloud Console is probably not a sign of a public release.

The more likely read:

Anthropic and Google are preparing or expanding cloud-side infrastructure for customers who already have access to Mythos through restricted programs like Project Glasswing.

Google previously announced Claude Mythos Preview as a private preview on Vertex AI for selected Google Cloud customers, and Anthropic still describes Mythos as an unreleased frontier model focused on high-risk cybersecurity capabilities.\n\nQT @AiBattle_: Claude Mythos now appears in the Google Cloud console, which was not the case yesterday

The preview label is also gone. Is Anthropic preparing for a public release?

Opus 4.7 also appeared first in the Google Cloud console before its release https://t.co/7gI0fET4u8

See 3 related tweets

@gdgtify: RT @testingcatalog: ANTHROPIC 🔥: Claude Mythos model has been spotted on Google Cloud Console.

-cl...

@theo: Predicted this 🙃 https://t.co/qUhUg0fhuX\n\nQT @testingcatalog: ANTHROPIC 🔥: Claude Mythos model has...
@Scobleizer: The Google IO hype starts.\n\nQT @testingcatalog: ANTHROPIC 🔥: Claude Mythos model has been spotted ...

8. eng_khairallah1 (Group Score: 96.1 | Individual: 29.7)

Cluster: 4 tweets | Engagement: 224 (Avg: 68) | Type: Tech

RT @eng_khairallah1: Anthropic's Claude team just showed how to build an AI agent with real memory in under 30 minutes.

24-minutes. free. by the people who built Claude.

worth than $500 vibe-coding course.

Bookmark & replace one movie today with this course, then read the complete article below. https://t.co/RB3LATkfHC

See 3 related tweets

@sairahul1: RT @sairahul1: Anthropic just dropped a 2-hour Claude agent training.

The engineer who built Claude...

@codewithimanshu: RT @codewithimanshu: Anthropic's Claude team just showed how to give your AI agents real memory in u...
@sairahul1: RT @sairahul1: Anthropic just gave away the exact playbook for building AI agents with real memory. ...

9. JaredSleeper (Group Score: 86.2 | Individual: 37.1)

Cluster: 3 tweets | Engagement: 92 (Avg: 31) | Type: Tech

For the last few months, I profiled one public software company per day (50 total) and wrote about the impact of AI on each.

Posts collated here:

https://t.co/cyEQtxFfYN

My takeaways after reflecting this morning:

This is not an innovator's dilemma situation like the on-prem to SaaS transition. The on-prem software companies were in deep trouble- they had legacy products and a legacy pricing model, and getting to the cloud meant sacrificing near-term revenue/profits/cash (subscription transition) AND moving each of their customers from an on-prem version of their software to the new cloud-native version, which created a significant change event- i.e. an occaison for their customers to consider whether the cloud-native option might be better, since they were going to have to make a big change anyway. For better or worse, the SaaS companies are not experiencing dynamics like that- no one is targeting them and running a successful "rip and replace" strategy at any scale. Net retention is stable to up, even for companies disappointing at the margins. The existing business model is intact for now- even growing (and in some cases accelerating).
The greater the long-term risk, the better the short-term case for acceleration. The paradox of investing in SaaS right now is that the more threatened a SaaS company is in the long term, the better the odds that it will accelerate and "disprove" the bear case in the near term. It is the companies closest to killer-app AI use cases (code, image/video gen/CX, etc.) that have both the best prospects for near-term upside and the most ferocious AI native competition. In many other categories, it simply isn't clear yet that AI adds enough value and/or is token-intensive enough to generate the incremental revenue required to accelerate. It seems likely that some SaaS companies will accelerate into their own obsolescence, and companies that don't accelerate near term are paradoxically better off in the long run (because they will have more time to adapt).
Most software management teams don't seem to have made wholesale organizational changes due to AI yet, and that's a disappointment. I would have expected to have read/heard widespread stories about increased operating leverage, cost takeouts, etc. but instead SaaS hiring continues apace and it shows up in management commentary. There is clearly a major struggle to change the culture of SaaS companies and elite talent is being poached by AI natives. I haven't seen a single management team talk about this honestly and put forward a strategy for attracting/retaining top talent in this environment- and I think that's a huge issue in the long run.
The mythical "shitty thin SaaS company with no moat" doesn't really exist in public markets. These companies have all gone through the gauntlet- competition with other VC-backed startups, competition with Microsoft, etc. Almost by definition, they've built up complex, moated businesses with brand equity, network effects, exceptional complexity, etc. That doesn't mean AI isn't an issue for them, but the "issues" I found that concerned me were more around disruption to the workflow the software company serves (see: Figma, Five9, etc.) than a disruption to the SaaS business model writ large.

Summing it up, the broad-brush SaaS bear case melts away somewhat when you go company-by-company. The median SaaS company is seeing little to no impact (in either direction today), while saying all of the right things on the product side and none of the right things on the organizational side.

That doesn't mean there won't be some sort of wholesale disruption down the road- but I don't yet have a clear picture of what it will look like. If you feel like you do, please pick a specific company, respond to or quote-tweet its profile, and explain what you think is going to happen. That's much more fun than debating in the abstract. :)

See 2 related tweets

@chandrarsrikant: RT @MilkRoadAI: Chamath just delivered the clearest diagnosis of what is happening to enterprise sof...
@vasuman: Some findings from across our Varick customers that might shape how you think about AI adoption goin...

10. zerohedge (Group Score: 82.1 | Individual: 25.4)

Cluster: 5 tweets | Engagement: 972 (Avg: 760) | Type: Tech

RT @cvkueppersbooks: I attended the University of Arizona commencement ceremony, where Eric Schmidt @ericschmidt faced boos throughout his speech. If you don’t know how young graduates feel about AI, this post is for you. The message is clear: it reflects growing skepticism toward AI narratives coming out of Silicon Valley. I keep coming back to one question for anyone building AI: are you building for humans? Build responsible and ethical AI. AI governance and compliance matter. (Speech excerpt.)

See 4 related tweets

@peterwildeford: Eric Schmidt gets booed every time he mentions AI 👀\n\nQT @ProudSocialist: The kids are alright!!

F...

@Hesamation: Sir congratulations, everyone now hates AI to their guts and want to puke when they hear about it. h...
@TFTC21: Former Google CEO Eric Schmidt got booed every time he mentioned artificial intelligence during his ...
@Scobleizer: RT @JoannaStern: This wasn’t the case a year ago when I gave a commencement speech almost entirely a...

11. peterwildeford (Group Score: 73.1 | Individual: 38.1)

Cluster: 2 tweets | Engagement: 112 (Avg: 53) | Type: Tech

aw it's so adorable the French think they can replicate Mythos\n\nQT @eliebakouch: Arthur Mensch answers to the french representatives:

"our (mistral) models are capable of finding all the vulnerabilities found by mythos"

"There are obviously people asking if they can buy us. We answer [no] because that's not our mission, and our mission is to be independent, [...] If you succeed, you don't get acquired. If you get acquired, in a way, you've failed"

some numbers:

1B R&D spend at Mistral this year at Mistral 10% of salary mass is spent on tokens estimates that 1 employee (in general, not at mistral) will consumes on average ~1kW in tokens per year, which is ~10k $1GW datacenter is$ 50B capex over 5 years. you can expect to make 2x revenue. electricity captures ~10% of value. revenue is 30% in France, rest of Europe is ~45%. public sector share is 20% with 10% in France. a bit less than 30% of Mistral capital is held by US VCs Mistral's goal is 1GW in 2029 they train/will train bigger models internally and distill them to serve to customers Mistral plays only a small part in the 35B investment (by MGX from UAE) in France, in the "campus AI" project announced at the AI summit earlier this year

some of their current clusters:

40MW in France 25MW in Sweden 80MW in France (next year) they train models on "10s of MW", mention that they need the gpu to be collocated to train model insists on the fact that EU/France advantage for building datacenters is nuclear power, which leads to less carbon footprint

See 1 related tweets

@firstadopter: Tiny Mistral says it can match Anthropic's Mythos in cybersecurity. The Anthropic Mythos narrative i...

12. FundamentEdge (Group Score: 71.6 | Individual: 39.3)

Cluster: 2 tweets | Engagement: 340 (Avg: 171) | Type: Tech

I was in New York last week and for the first time I'm starting to hear senior investment professionals think seriously about the incremental utility of junior investment talent.

Agents are solving process bottlenecks that heretofore required the efforts of junior investment talent across modeling, meeting prep, and data analysis.

”I may not need to hire a junior after all”, It is a more common refrain. And it's widespread enough that I do think some concern is now warranted.

My response back is: what if you just fundamentally rethought the role of the junior investor? What if instead of updating models, updating weekly industry data drops, writing first cut earnings previews, and building out management question lists, the responsibility stack of a junior dramatically changed?

What if you simply challenged the junior investor to develop more accurate and timely revenue forecasts across three dozen names? And the apprenticeship role almost felt more like an AI enabled data scientist, while that person develops their own investor toolkit.

What if you challenged that junior analyst to be an in-house AI-enabled risk analyst? Building AI-augmented risk checklists, counter pitches on every idea in the portfolio & pipeline?

What if you challenged that junior to become the most AI-fluent person on the team, putting them through an intensive six-month AI self-study? Which, for your senior investors with portfolio responsibility, is very difficult to find the time.

The junior then, through the apprenticeship period, becomes a highly value-added generalist.

And during that apprenticeship period, where they're mostly a cost center, they develop into a true AI-augmented investor, and at year three will drive dramatically positive ROI to the firm. Same apprenticeship model, but rethought in specific structure.

The foundational reality still exists that your junior investor for a scaled investment firm is going to be a rounding error in terms of comp. The consideration of slowing down junior hiring isn't a cost consideration, it's an efficiency consideration. And while much of the traditional junior stack is quickly becoming abstracted into agents, I do think there is a real potential to rethink the role of junior investors to the effect of very attractive ROI.

I do think, to a large degree, we have some moral responsibility to do this as well. In a competitive business, it is absolutely imperative that we drive efficiencies and research quality with all the tools at our disposal. That definitionally will impact the existing role of juniors more than seniors.

I'm feeling a bit more of a mission to help firms rethink how they deploy junior talent. If you're doing the same, I'd love to connect and find creative ways to rethink the junior role in fundamental investing firms.

See 1 related tweets

@ShanuMathew93: Love the ideas in here. The investment industry is going to evolve quite a bit over the next 5-10 ye...

13. StockSavvyShay (Group Score: 71.5 | Individual: 37.1)

Cluster: 2 tweets | Engagement: 818 (Avg: 629) | Type: Tech

MY 3 FAVORITE WAYS TO PLAY THE CPU BOTTLENECK

Jensen Huang keeps saying the next era of computing will be built around “AI factories” but every agentic workload creates a massive amount of CPU-bound work around the GPU.

The best way to think about this is the GPU will still continue being the engine but CPU will now become the traffic controller. If AI agents are going to run inside enterprises all day across millions of workflows then the CPU layer becomes a much more important part of the AI infrastructure stack.

$AMD | AMD

AMD is the cleanest way to play the agentic AI CPU bottleneck through EPYC where they just posted record Q1 data center revenue of $5.8B (+57% YoY) and biggest gem was AMD capturing ~46% of server CPU revenue despite only 27% unit share (AMD is winning higher-value server CPU dollars that AI infrastructure actually needs).

CEO Lisa Su also raised the server CPU TAM forecast to $120B+ by 2030, guided Q2 server CPU revenue to grow more than 70% YoY and layered on the$ META 6GW Instinct deal which also makes Meta one of the first customers for next-gen Venice and Verano EPYC processors. Intel's Diamond Rapids was also delayed to mid-2027 which means AMD has effectively zero competitive response during the steepest part of the agentic AI buildout.

$ARM | Arm

Arm is the most interesting of the three because the story is no longer just “Arm collects royalties” since they're now becoming a direct silicon supplier through the Arm AGI CPU which it says can deliver more than 2x performance per rack versus x86 platforms while reducing AI data center capex by up to $10B per gigawatt.

Arm said customer demand for the AGI CPU has already reached more than $2B across FY27 and FY28 which now gives Arm two ways to win from the same bottleneck: royalties on custom hyperscaler CPUs and direct sales into AI infrastructure.

$INTC | Intel

Intel is the most controversial of the three because the market still treats it like a permanent AI-cycle loser but agentic AI could lift every credible x86 supplier and Intel now has two ways to capitalize on it: product and foundry.

On the product side, AI-driven businesses now account for 60% of total revenue and are growing 40% YoY while Xeon 6 was selected as the host CPU for $NVDA DGX Rubin NVL8 systems giving Intel a role inside the next generation of Nvidia AI deployment.

On the foundry side, Intel agreed to manufacture chips designed by $AAPL while Nvidia's$ 5B investment and commitments from Musk-linked companies validate Intel's manufacturing roadmap at the exact moment AI infrastructure needs more advanced domestic chip capacity.

See 1 related tweets

@StockSavvyShay: This is what happens when you become the face of the largest industrial buildout in history.

$NVDA...

14. DataChaz (Group Score: 70.1 | Individual: 41.3)

Cluster: 2 tweets | Engagement: 440 (Avg: 142) | Type: Tech

🚨 STOP BURNING YOUR TOKENS!

If you use Claude Code, you are probably wasting 80% of your context window.

I found 10 ace tools that will completely rescue your API bill.

Caveman Claude

Literally makes Claude talk like a caveman
Slashes 75% of output tokens with zero loss in accuracy Repo → https://t.co/eEvSOvHutG

RTK (Rust Token Killer)

A blazing fast proxy that filters terminal output
60-90% reduction and completely dependency-free Repo → https://t.co/lDfjbsbPD5

Code Review Graph

Claude reads only what matters using a Tree-sitter graph
An unbelievable 49x token reduction on huge monorepos Repo → https://t.co/xGn6Pp88yX

Context Mode

Sandboxes raw output into SQLite instead of your context
A staggering 98% context reduction on logs & GitHub Repo → https://t.co/Jut2bvBMUD

Claude Token Optimizer

Brilliant setup prompts that optimize any project
90% token savings, taking docs from 11K to 1.3K Repo → https://t.co/0uOFODbG7e

Token Optimizer

Hunts down the invisible ghost tokens eating your context
Fully restores and protects your context quality Repo → https://t.co/LUOzjECXKm

Token Optimizer MCP

Adds aggressive caching and compression to your MCP tools
95%+ token reduction through pure intelligence Repo → https://t.co/b5Eqruo2PM

Claude Context

Zilliz’s hybrid vector search MCP
Makes your entire codebase the context for 40% less cost Repo → https://t.co/hPG6pb0j3G

Claude Token Efficient

Just drop one CLAUDE.md file into your repo
Enforces strict terseness with zero code changes Repo → https://t.co/fNrl6nwItF

Token Savior

Navigates your code by symbols, not giant files
97% reduction on code navigation with persistent memory Repo → https://t.co/lkILPhfwJh

[ The god-tier stack ] Pick 2-3 based on what’s draining you:

Massive repo? Code Review Graph + Token Savior Heavy terminal output? RTK MCP data dumps? Context Mode Need an instant fix? Caveman + Claude Token Efficient

Most devs are bleeding tokens.

Run /context in a fresh session and watch the savings roll in 👀\n\nQT @0x_kaize: https://t.co/ScYWvkqPa4

See 1 related tweets

@DataChaz: RT @DataChaz: 🚨 STOP BURNING YOUR TOKENS!

If you use Claude Code, you are probably wasting 80% of y...

15. yacineMTB (Group Score: 67.6 | Individual: 21.5)

Cluster: 4 tweets | Engagement: 256 (Avg: 194) | Type: Tech

insane that they would do this. they're going way too far\n\nQT @theo: To prevent "programmatic use", Claude Code may now request webcam access to assure user is present when prompting https://t.co/Fleg7uY7xc

See 3 related tweets

@cgtwts: Dude. https://t.co/xyloC9UzRz\n\nQT @theo: To prevent "programmatic use", Claude Code may now reques...
@chiefofautism: dario wanna see how u masturbate\n\nQT @theo: To prevent "programmatic use", Claude Code may now req...
@SIGKITTEN: RT @theo: To prevent "programmatic use", Claude Code may now request webcam access to assure user is...

16. ylecun (Group Score: 66.2 | Individual: 44.8)

Cluster: 2 tweets | Engagement: 14126 (Avg: 1844) | Type: Tech

RT @Ric_RTP: Trump just got exposed for running the biggest insider trading operation in American history.

Nancy Pelosi traded $5 million in stocks and Congress lost its mind.

Trump literally executed $750 MILLION worth of stock trades in ONE quarter while being President.

His ethics filing just dropped and the numbers are genuinely unprecedented in history:

Between January and March 2026, Donald Trump personally executed 3,700 individual stock transactions worth between $220 million and$ 750 million.

That's roughly 60 trades PER DAY.

While signing executive orders, meeting foreign leaders, and making policy decisions that directly impact the companies he's buying and selling.

Now here's where it gets really insane:

On February 10, Trump bought between $1 million and$ 5 million worth of Dell stock.

Three months later, on May 8, he stood at a Mother's Day event at the White House, thanked Michael Dell by name, and told Americans to "go out and buy a Dell."

Dell stock surged 14.6% that day to an all-time high of $263.99.

Since Trump's February purchase, Dell is up 96%.

And 5 months BEFORE Trump bought Dell stock, Michael and Susan Dell donated $6.25 billion to Trump Accounts, one of the largest philanthropic commitments to a sitting president's signature program in modern history.

So the timeline goes: Dell donates $6.25 billion to Trump's program -> Trump buys Dell stock ->Trump tells America to buy Dell from the White House podium -> Stock hits all-time high

And that's just ONE stock...

The same filing shows Trump bought Nvidia stock on February 10. One week later, Nvidia announced a massive chip deal with Meta.

He bought more Nvidia stock one week BEFORE his own Commerce Department approved the sale of Nvidia chips to Saudi Arabia.

He bought Intel stock starting in March 2026. The US government already owned a 9.9% stake in Intel worth over $41 billion. On April 30, Trump posted on Truth Social praising Intel, writing that "Intel Stock continues to rise."

Intel jumped 3% in after-hours and is now up 140% year-to-date.

He bought Palantir stock while his administration was actively handing them billion-dollar government contracts for immigration enforcement and defense.

He bought Robinhood stock while his own Trump Accounts program uses Robinhood as the broker.

He's currently sitting on over 100% profit on AMD, Intel, Bloom Energy, Marvell Technology, and at least 10 other positions.

Every single president since Lyndon B. Johnson has used a blind trust to avoid exactly this situation. But Trump didn't.

His assets sit in a trust controlled by his own children, and the filings show a broker acted as agent on several trades.

The White House says the portfolio is "independently managed."

But here's what independently managed looks like:

Buy Dell stock. Three months later, publicly endorse Dell from the White House. Stock hits all-time high.

Buy Nvidia stock. One week later, your own government approves their chip sales. Stock rips.

Buy Intel stock. Post about Intel on Truth Social. Stock jumps. The government you run already owns a 10% stake.

Buy Palantir. Hand them contracts. Buy Robinhood. Route a federal program through their platform.

Nancy Pelosi got absolutely destroyed for her husband's stock trades.

Her husband's total disclosed trades in his most controversial year were worth roughly $5 million.

Trump just disclosed up to $750 MILLION in a single quarter.

While making the actual policy decisions that move these stocks.

This isn't a left or right issue.

We're talking about the President of the United States averaging 60 stock trades per day in companies his own administration regulates, contracts with, and publicly endorses.

What do you think?

See 1 related tweets

@TFTC21: According to ethics filings released this week, President Trump made over 3,600 stock trades during ...

17. tengyanAI (Group Score: 65.0 | Individual: 35.5)

Cluster: 2 tweets | Engagement: 147 (Avg: 50) | Type: Tech

The strangest AI bottleneck might be ABF.

Ajinomoto is better known for MSG, but its build-up film (ABF) is a critical dielectric layer inside the package substrates used for high-end GPUs.

the moat isn't chemical secrecy alone.. it is qualification lock-in + process integration + reliability history.

and the fact that no chip company wants to risk a $30,000 AI accelerator package to save pennies on dielectric film.

is it a permanent monopoly? Probably not. but it looks cycle-durable.

if you ask me:

1-3 yrs: very hard to replace in high-end AI substrates
3-5 yrs: glass substrates and alternatives gain in specific packages
5+ yrs: package architectures probably shift

To see this, i'm tracking closely whether Nvidia, AMD, Broadcom, TSMC, Ibiden, Shinko and Unimicron show signals for moving away from ABF.

See 1 related tweets

@tengyanAI: AI supply chain lesson: the bottleneck is often not the most futuristic-looking company...

sometime...

18. MarioNawfal (Group Score: 64.0 | Individual: 36.3)

Cluster: 2 tweets | Engagement: 1330 (Avg: 531) | Type: Tech

🇺🇸 Tucker lays out the deepest critique of AI yet, and it's not about jobs...

His argument: writing produces thinking.

You can't formulate a thought without first articulating it.

If kids never write because AI writes for them, the quality of human thinking collapses.

That's the surface problem.

The deeper one is purpose:

"The point of living is to create.

That's the point of being a human being. It's necessary for joy.

There is no joy without creation."

If the machine creates everything and humans just consume, you don't get utopia.

You get despair, mass unemployment, and eventually political revolution.\n\nQT @MarioNawfal: 🇺🇸🇨🇳 The U.S. has 4,000 data centers, while China has 365.

But 24 months ago, AI training required 100 megawatts.

Today the minimum is 1 gigawatt.

The U.S., Canadian, and Mexican grids can't deliver that. China's can.

The AI race was never about who built more, but who built bigger, and right now, America's own power grid is the bottleneck.

See 1 related tweets

@pmarca: Interesting.\n\nQT @JordanSchachtel: People keep asking "what jobs will AI destroy?" Wrong question....

19. petergostev (Group Score: 63.4 | Individual: 39.5)

Cluster: 2 tweets | Engagement: 1104 (Avg: 146) | Type: Tech

12 months apart - one is GPT-5.5 and another o3.

Can you guess which one is which? Looks carefully, the difference is quite subtle https://t.co/4b44BeIanb\n\nQT @petergostev: If you feel like AI is slowing down, try coding with o3 - best model only a year ago

See 1 related tweets

@petergostev: If you feel like AI is slowing down, try coding with o3 - best model only a year ago...

20. BrianRoemmele (Group Score: 61.2 | Individual: 32.5)

Cluster: 2 tweets | Engagement: 61 (Avg: 511) | Type: Tech

THE ANTI-DATA CENTER/ANTI-CLANKER FOLKS DON’T WANT YOU READING THIS.

It is what I told them when staffers came calling for my anti-clanker stance, they got from me, “fix the clothing industry by 10% and you will “save the water” and have a good May 1st comrade. https://t.co/tl2dwxGieN\n\nQT @BrianRoemmele: “THAT DATA CRNTER IS WASTING WATER, STOP ALL DATA CENTERS”

I see, let’s talk about that t-shirt you are wearing first or the jeans, the water could support 100s of AI queries or days of computation.

In the grand theater of human consumption, few spectacles rival the quiet hypocrisy of decrying data centers while embracing mountains of disposable clothing. Fast fashion: cheap, trend-driven garments churned out in endless cycles, represents a voracious, often invisible drain on water, energy, and ecosystems.

Meanwhile, data centers, the engines powering AI and digital life, face scrutiny for their cooling needs.

A clear-eyed comparison reveals misplaced priorities: the garment industry’s water use is vast, frequently consumptive or polluting in water-stressed regions, with products destined for landfills after minimal use.

Data center water, by contrast, is largely local, often recyclable or evaporative (returning to the hydrological cycle), and supports immense economic and innovative value. It also is just a fraction of the garment industry.

Water in the Garment Industry: Hidden Rivers and Polluted Legacies  The fashion and textile sector consumes staggering volumes of water annually. Estimates range from 79 to 215 billion cubic meters (roughly 79–215 trillion liters), supplying the drinking needs of millions of people.

This makes it one of the world’s most water-intensive industries, second only to agriculture in some assessments.

Breaking it down garment by garment:  • A single cotton T-shirt requires ~2,500–2,700 liters of water across its lifecycle (growing, processing, dyeing).  • A pair of jeans: 7,500–10,000 liters.  • Leather items push even higher (8,000+ liters for shoes).21 Cotton, which dominates natural fibers, is particularly thirsty. Global averages hover around 8,920 liters per kg of cotton lint (much from rainwater/“green” water, but ~2,344 liters/kg from irrigation/“blue” water in stressed areas like parts of India, Pakistan, and China).

Processing and dyeing add 100–150 liters per kg of fabric, often with toxic chemicals.

The dyeing phase alone accounts for hundreds of billions of liters yearly and contributes to ~20% of global industrial water pollution.

Untreated wastewater laden with dyes, heavy metals, and chemicals flows into rivers, devastating local ecosystems and communities.

Fast fashion amplifies this: Production has doubled in recent decades, with consumers buying 60% more clothes than 15–20 years ago, while usage duration drops.

About 100 billion garments produced yearly; 92 million tonnes of textile waste generated, much ending in landfills (a garbage truck’s worth every second). In the U.S., landfills received 11.3 million tons of textiles in 2018.

Synthetics (polyester ~55–68% of fibers) add microplastics via washing, now a major ocean pollutant. Cheap clothes are worn briefly, discarded, and replaced—embodying “take-make-waste” at planetary scale.

This water is not local and often lost or ruined: Irrigation depletes aquifers in arid regions; polluted effluent renders water unusable downstream.

The full supply chain spans continents—cotton from India/Uzbekistan, dyeing in Bangladesh/China, exporting environmental costs to vulnerable areas.

Data Centers: Local, Cyclical Water Use for Digital Progress  Data centers primarily use water for evaporative cooling (or increasingly air/closed-loop/immersion systems). Global estimates: ~560 billion liters annually now, potentially doubling or more by 2030 with AI growth: still a fraction of fashion’s footprint and far below agriculture (~70% of global freshwater). U.S. data centers consumed ~64 billion liters directly in 2023.

BRAND NEW CLOTHING IS TOSSED IN THE DESERT WITH PRICE TAGS STILL ON IT.

All to make the brand look rare. Can’t have poor folks wearing it.

Meet the infamous fast fashion “clothing graveyard” (also called the “great fashion garbage patch”) in Chile’s Atacama Desert here:

1 of 3

See 1 related tweets

@BrianRoemmele: RT @BrianRoemmele: “THAT DATA CRNTER IS WASTING WATER, STOP ALL DATA CENTERS”

I see, let’s talk abo...