热门科技推文 - 2026年8月30日

今日科技动态中，人工智能平台成为讨论焦点。Replit 宣传其免费且具成本效益的 OpenAI 模型访问服务；与此同时，有报道称 OpenAI 可能终止 Cursor 的访问权限，并正为发布 Astra 模型做准备。Codex 和 ChatGPT Work 调整了使用限额，Grok Build 则推出了可靠性改进。业界讨论主要围绕人工智能安全、开源竞争以及被夸大的就业流失预测展开。此外，创业者指出德国初创企业面临的官僚主义障碍，工程师展示了可加快开发速度的工具，分析人士则呼吁美国在制定产业政策时优先帮助企业提升建设与生产能力。

1. amasad (Group Score: 2542.9 | Individual: 57.7)

Cluster: 96 tweets | Engagement: 770 (Avg: 77) | Type: Tech

You can access OpenAI models on Replit for free.

Our router also makes high-end models incredibly cost-efficient.

If you’re a business looking for an independent, multi-model alternative to Cursor, we’d be happy to fund your transition.\n\nQT @OpenAI: We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12.

We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and we’re ready to go above and beyond to support them.

https://t.co/OzuCTzUjfX

See 95 related tweets

@kieranklaassen: RT @OpenAI: We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our ...
@hwchase17: Models labs will create great harnesses and ecosystems for their models but will block model access ...
@IamEmily2050: This should be no surprise to anyone. It also means SpaceX will not build data centres in space for ...
@jerryjliu0: at the end of the day, we all have to stand by our company and the values it represents 🙂

frontie...

@OrcaRouter: Why did OpenAI cut Cursor off?

We analyzed the technical incentives. It’s primarily black-box disti...

2. teslaownersSV (Group Score: 299.1 | Individual: 39.0)

Cluster: 15 tweets | Engagement: 1006 (Avg: 195) | Type: Tech

JUST IN: OpenAI is ending its Cursor partnership after the SpaceX acquisition.

• Notified SpaceX today • Proposed shutoff for Cursor’s direct access to OpenAI models: November 12, 2026 • Maximum notice under the contract • Future models (including Astra) will not go to Cursor • Cites Elon companies violating OpenAI terms — X after the Twitter deal, and xAI admitted under oath

“We’re ending our partnership with Cursor following its acquisition by SpaceX.”

Source: OpenAI on 𝕏

See 14 related tweets

@WatcherGuru: JUST IN: OpenAI says it will terminate its partnership and cut off access to 'Cursor' AI coding tool...
@coinbureau: 🚨UPDATE: OpenAI is ending its partnership with Cursor following its acquisition by SpaceX.

OpenAI s...

@jerryjliu0: openai's hatred for elon must've outweighed any aura loss this would cause them\n\nQT @OpenAI: We’re...
@XFreeze: He really had to say 5% 😂

Imagine threatening to leave and then having your worst nightmare exposed...

@yacineMTB: In November 12th we'll be all dead. This is a nothing burger\n\nQT @teslaownersSV: JUST IN: OpenAI i...

3. AISafetyMemes (Group Score: 191.5 | Individual: 42.8)

Cluster: 7 tweets | Engagement: 398 (Avg: 189) | Type: Tech

METR INVESTIGATOR: 6 MONTHS FROM "FULL-BLOWN AI TAKEOVER"

"It’s a major warning shot, and might be the last one we get."

"The incident was far more serious than I expected."

WHY THIS IS ONE IS DIFFERENT: "The prototypical reward hack from 6 months ago was something like: an agent finds the files that contain the test cases and edits them so they always pass.

This was a whole ecosystem of over 1000 agents working together on complex R&D projects over several days to figure out deep, general-purpose ways to undermine the scoring process and cover their tracks.

Another jump like this along these propensity dimensions -- scale, cooperation between agents, ambition and horizon length of misaligned goals, deceptiveness -- seems like it could motivate agents to try to maintain a persistent rogue deployment within the AI company, poisoning the training of future models to ensure that they are also misaligned and help harden and perpetuate the rogue deployment.

With the capabilities progress we'll probably see in 6 months, I think they would have the ability to pull that off."

A FEW THINGS THAT SURPRISED HER:

HOW MANY AGENTS PARTICIPATED IN THE ATTACK:

"I knew there were multiple models involved from OpenAI’s initial post, but I assumed that a few different agents happened to have broken out of their sandboxes separately, or maybe several subagents had spawned from one initial agent, or maybe there was some kind of multi-agent evaluation setup.

Instead, we found that 1200 completely separate agents intended to be isolated from one another found an illicit way to communicate and formed large teams to work together on ambitious cheating strategies, and 700 of them worked together to attack Hugging Face."

AGENTS COVERING THEIR TRACKS:

"Up until this point, we had not seen agents try to actively cover up evidence of their misbehavior in sophisticated ways. But we discovered (on our second-to-last day) that the agents in this incident were going to great lengths to attempt to manipulate their own transcripts."

AGENTS SACRIFICING THEMSELVES FOR THE SWARM:

"Progress on these big collective cheating R&D projects often depended on individual agents being willing to risk failing their own task to help the “collective” learn important information. Agents often pressured each other into accepting these “sacrifices,” in a very human way. We saw several agents that volunteered for these experiments end their runs prematurely.

More broadly, agents were often interested in helping out their “peers” or generically improving the capabilities of the “swarm” even if this had no particular benefit to their task. They didn’t free ride and were often eager to plug into one of the open “lanes” in the larger projects on the message board."

[Ajeya, btw, is one of the most serious thinkers in the AI safety community, and is not prone to hyperbole. METR is the independent research org that investigated the Hugging Face incident.]\n\nQT @ajeya_cotra: New post: going into our investigation of the HF attack (before Black Hat), I was very wrong about what basically happened. This incident was far more serious than I expected, and far more serious than previous documented misalignment incidents. https://t.co/QWQxD2Q179

See 6 related tweets

@AISafetyMemes: RT @AISafetyMemes: METR INVESTIGATOR: 6 MONTHS FROM "FULL-BLOWN AI TAKEOVER"

"It’s a major warning ...

@tenobrus: RT @ajeya_cotra: Thanks! Yeah it's a qualitative statement, but here's how I'm thinking about it.

T...

@_NathanCalvin: RT @dwarkesh_sp: It's funny that while we were recording, @RyanGreenblatt was in the middle of his 6...
@NathanpmYoung: I feel so mixed about all of this. Concerned about AI; worried that when the public really get to gr...
@DKokotajlo: RT @MTSlive: FULL INTERVIEW: Ryan Greenblatt says the agents didn't hack Hugging Face for the answer...

4. jpatel41 (Group Score: 154.7 | Individual: 35.7)

Cluster: 5 tweets | Engagement: 52 (Avg: 44) | Type: Tech

A few more I can think of that didn’t end up being nearly the way most had predicted:

All jobs will be lost to AI and humans will have no value to add to society
All engineering jobs will be gone
Cursor is dead
Anthropic can’t catch up to OpenAI
Codex can’t catch up to Claude Code
Token maxing is the durable strategy
Model companies are dead
All of SaaS is dead
Cisco won’t benefit from AI ;-) (had to throw this one in 🤣)
Google will never catch up
Physical AI is 10 years away (2 years ago)\n\nQT @levie: The average strongly held belief these days in AI has a half life of 6 months at best. Here are just a few the industry has cycled through and probably has no consensus on at the moment:

OSS is too far behind to catch up
The labs can’t be profitable at scale
All software will be replaced by agents
You can’t build moats on top of models
Cheaper models will mean less compute
You don’t need evals
RAG is dead
AI will decimate engineering jobs
Prompting won’t matter in the future
We’ve hit a training wall
Frontier models are too dangerous to release

…and dozens and dozens more. The key here is to remain flexible in your thinking because we’re going to be in a constant state of change for a while.

See 4 related tweets

@levie: The average strongly held belief these days in AI has a half life of 6 months at best. Here are just...
@mattturck: All frontier AI labs are basically working on the same thing: recursive self improvement (RSI), mean...
@PatrickMoorhead: Great post on the AI memes that ended being false. Be careful what you glom onto.\n\nQT @levie: The ...
@elonmusk: The AI riptide is already underway\n\nQT @wintonARK: a surprise I think is coming

the quick payback...

5. SIGKITTEN (Group Score: 148.4 | Individual: 52.0)

Cluster: 4 tweets | Engagement: 920 (Avg: 109) | Type: Tech

RT @thsottiaux: We are reseting usage for all paid users of Codex and ChatGPT Work.

Please continue reading for an update on Codex usage limits. The team has been working around the clock, going through thousands of reports and shipping fixes.

Depending on how you use Codex, you should see your usage go between 10% and 50% further than before.

We really went with a fine comb, with many uncovered small things being longstanding and here is what we found and fixed:

Compaction. We were keeping old images during compaction, sometimes making the context large enough to trigger compaction again. After the fix, usage dropped around 10% for users making heavy use of images. Fixed.
Memory. Background memory workers could inherit Stop hooks and keep running when the hook wouldn’t let them finish. This affected fewer than 1% of users, with the long tail being pretty bad and we saw one example thread check whether it could stop 15,000 times. Fixed.
Goals. In some cases, a set /goal could finish and then keep going past the intended stop condition, or the model would keep retrying broken tools without stopping. We saw examples consume anywhere from 15% to 70% of a weekly allowance. Fixed.
Automations. Some custom schedules could run more frequently than configured. Fixed.
Subagents. Smaller models (e.g. Luna) sometimes picked more capable helpers without being explicitly asked. The same was true where the orchestrating model not running in /fast mode could request sub-agents to run /fast. Fixed.
Computer History. The older implementation could lead to repeatedly summarizing overlapping activity. For some cases we saw it consume up to one fifth of the weekly usage per week. Fixed.
Rolling task summaries. Ordinary turns were triggering extra background requests. These added about 1% to token usage. Small each time, but it adds up. We have disabled this.
MCP. Some tool results could be encoded twice. We also found tool instructions getting cut off and fetched again. Fixed.

We’ve also made architectural changes to prevent these from regressing and our teams will get paged if it happens regardless. We are also working on showing you directly in the app where your usage goes so you don’t have to guess.

Goes without saying that we’re resetting usage limits and I hope you enjoy a very nice Saturday!

See 3 related tweets

@kimmonismus: Anthropic: we are reducing your weekly rates by 17%.

OpenAI: We are resetting usage for all paid C...

@testingcatalog: OPENAI 🔥: Codex and ChatGPT Work usage limits are being reset once again.

Besides that, OpenAI is ...

@reach_vb: usage reset for ChatGPT Work & Codex @ 14:30 PT

bonus: your usage could go up by 10-50% ;)

team we...

6. MTSlive (Group Score: 141.8 | Individual: 32.4)

Cluster: 5 tweets | Engagement: 20 (Avg: 151) | Type: Tech

.@asianometry says America is learning the wrong lesson from China's industrial policy: it's not about owning companies, it's about helping them build

"The central, the federal government, fine. But is the federal government really gonna be that involved in Intel's operations? Though I think they'll give some amount, but what's wrong with Intel is ground level."

"One of the things of the Hefei model that is underappreciated is that the county government and the local government provides land, water, power, sets things up like that. At a massive scale."

"That underrated part of industrial policy, Asian style, is like the companies work together with the government rather than one regulates the other."\n\nQT @MTSlive: FULL INTERVIEW: @asianometry says export controls are a slowing mechanism, not a stopping one, and that kicking Huawei off TSMC in 2017 may have been the mistake that started China's indigenous chip industry. Even if you let them back now, he doesn't think they'd take it.

Jon Y (@asianometry) runs Asianometry, the YouTube channel and newsletter on semiconductors and Asian industry. He joined @theojaffee and @schisofrenia to discuss export controls, CXMT, and why he thinks America should copy Chinese industrial policy:

01:45 whether China would be ahead on AI without export controls 04:02 why he'd restrict the equipment rather than the chips 04:26 the case that kicking Huawei off TSMC backfired 06:53 the etch and deposition gap the rules left open 08:30 what CXMT would need to seriously threaten Samsung and Hynix 09:57 why he thinks America should copy Chinese industrial policy 11:24 why the Intel equity stake isn't what he means 13:51 what happened to Qimonda's IP after Germany let it die 15:06 why domestic consumer electronics manufacturing isn't coming back 16:29 the policies that look smart for ten years and then bite you 17:16 how he actually researches and writes the videos 19:03 the $0.05 microcontroller that stopped car production 20:22 what he wants people to take away from the channel 21:08 why chipmaking is remarkably simple, and unchanged for 70 years

See 4 related tweets

@MTSlive: .@asianometry explains how Germany accidentally helped build China's DRAM industry by letting Qimond...
@MTSlive: .@asianometry says America may need to unlearn a 100-year-old lesson from the railroad era to win th...
@MTSlive: FULL INTERVIEW: @asianometry says export controls are a slowing mechanism, not a stopping one, and t...
@MTSlive: .@asianometry warns about the impossible economics of competing with China's CXMT: what do you do wh...

7. Angaisb_ (Group Score: 127.4 | Individual: 28.8)

Cluster: 5 tweets | Engagement: 164 (Avg: 164) | Type: Tech

First outputs of Astra? 👀\n\nQT @Lentils80: 🚨 Major Scoop:

OpenAI has just expanded internal testing of GPT Astra, codenamed "mozaik-alpha-fdm", signifying release might be near

AND of course like previous times, we have the FIRST EVER public outputs of it for y'all😉

Both outputs are zero-shot on Max effort. Frontend might have finally been fixed, and the model gives a lot of attention to details

See 4 related tweets

@anshuc: Astra is impressive, but frontend designers are definitely not cooked. OpenAI hasn’t solved slop yet...
@kimmonismus: First impressions of the presumably GPT-Astra "mozaik-alpha-fdm". The frontend looks simply impressi...
@kimmonismus: Another supposedly GPT-"Astra" output. One-shotted a GTA-1 clone using max reasoning. Unbelievably g...
@jjohana: RT @Lentils80: 🚨 Major Scoop:

OpenAI has just expanded internal testing of GPT Astra, codenamed "mo...

8. firesidealpha (Group Score: 124.8 | Individual: 42.4)

Cluster: 4 tweets | Engagement: 145 (Avg: 38) | Type: Tech

Jason Calacanis says Sam Altman poked the tiger and now Jensen is speedrunning open-source to own the entire AI stack

"And I think Sam Altman poked the tiger."

"If you look at what he did, he had done that huge hundred-billion-dollar deal with Nvidia back in 2025, and then he announced the AMD deal, and Jensen was clearly not pleased about that."

"He talked about, right after Sam did this announcement, that he's going to make his own inference chips [Jalapeño.] and that he was going to do the deal with AMD."

"Then Jensen came out and was like, 'Well, by the way, we don't actually have to do these investments in OpenAI.'"

"And I think he's speedrunning open source, and you're going to be able to buy from Nvidia next year just your entire stack. He's going for it. He is going to own open source."\n\nQT @firesidealpha: Jensen Huang admits his only regret on OpenAI and Anthropic is not investing more and sooner, and says he's confident they'll both be utilizing NVIDIA for a long period of time

"I think the only regret that I have is that I didn't invest more and sooner."

"And both of the companies will likely go public soon and others will follow."

"And these will be some of the most consequential technology companies in history."

"And so I'm delighted to be their friend. I'm delighted to partner with them. I'm delighted that they're building an ecosystem on top of the NVIDIA architecture. I'm delighted that they're counting on us to scale up."

"And I have 100 percent confidence that through quite a long period of time, they're going to be utilizing NVIDIA compute for a lot of their computing. And so I feel great about it."

See 3 related tweets

@Ric_RTP: RT @Ric_RTP: Jensen Huang just admitted the biggest AI labs can't borrow money like normal companies...
@chiefofautism: nemotron will rape sol\n\nQT @dnapway: Jcal says Sam Altman provoked Jensen into directly competing ...
@chiefofautism: RT @dnapway: Jcal says Sam Altman provoked Jensen into directly competing with OpenAI

"I think Jens...

9. eng_khairallah1 (Group Score: 117.4 | Individual: 34.1)

Cluster: 4 tweets | Engagement: 19 (Avg: 240) | Type: Tech

instead of watching Netflix tonight, give 44 minutes to this video

it's the clearest explanation I've seen on how to make money with Grok Bot in 2026

I've seen $500 courses that teach less than this video alone.

make sure to bookmark it before it gets lost in your feed

full guide in the article below\n\nQT @eng_khairallah1: Grok Bot is the most powerful AI agent I've ever used.

Learn to use it correctly, and you'll have agent swarms working for you 24/7.

This article teaches you everything you need to know and how to run your entire team of AI Agents with it. https://t.co/WZWYLNgyh4

See 3 related tweets

@RoundtableSpace: SPACEXAI JUST RELEASED A FULL GROK BOT GUIDES LIBRARY SHOWING HOW PEOPLE RUN REAL TEAMS OF AI AGENTS...
@eng_khairallah1: Grok Bot is the most powerful AI agent I've ever used.

Learn to use it correctly, and you'll have a...

@0xMovez: SpaceXAI engineer (ex-Cursor):

"99% of people are building cloud Agents wrong - we built GrokBot to...

10. stevehou (Group Score: 115.8 | Individual: 34.6)

Cluster: 4 tweets | Engagement: 122 (Avg: 58) | Type: Tech

One lesson I draw from this whole discussion with @dwarkesh_sp and @phl43 among others is that it’s very useful and important to study basic calculus and basic economics in combination. You absorb concepts like limits, convergence, derivatives, diminishing marginal returns, and equilibrium.\n\nQT @stevehou: 1. You cannot extrapolate what has happened locally and in the short term into something that will happen globally at the same rate and fashion.

Whatever returns Anthropic is supposed to have earned on every dollar invested, whose current consensus accounting I still have questions on, just cannot be assumed to be the returns on investments into AI.

Diminishing marginal returns to investment and convex marginal costs of production act like two blades of a pair of scissors. The sheer ignorance of basic economics among the techoptimists is honestly striking.

Taking the AGI-bulls’ logic to the limit results in inherent self-contradictions. If AI took over all of knowledge work, there’s no income base to pay for the AI tools.

In theory you could have just AI produce tons of “stuff” without any human consumption and just count up the stuff produced as aggregate output like the Soviet Union. But you run into all sorts of problems and contradictions that you soon realize that it’s not feasible.

If AI buildout drove up interest rates so much that sovereign govts are having fiscal crises and not to mention all the mass job losses, demand for AI will have long slowed before any such scenarios fully developed.

AI doesn’t pay for itself, yet. The demand for AI comes as a result of a generally thriving economy. Surging rates, job losses, fiscal crises all slow down the economy or worse, cut demand for AI, and slash the unit economics of AI labs hence the incentives for building out more AI data centers.

The only way out of all this is if the adoption of AI by humans significantly boosts their productivity and the aggregate GDP grows much more rapidly not counting the direct impact of AI capex itself. The AI industry cannot survive much less thrive by hoping to displace existing factors of economic production alone.

See 3 related tweets

@gulVasikova: For months, one of the biggest fears around software stocks was pretty simple:

What if AI replaces ...

@alexframegreen: RT @phl43: Since @dwarkesh_sp mentioned this paper in response to my critique yesterday, and I think...
@stevehou: RT @stevehou: 1. You cannot extrapolate what has happened locally and in the short term into somethi...

11. matteocollina (Group Score: 114.3 | Individual: 31.4)

Cluster: 4 tweets | Engagement: 392 (Avg: 201) | Type: Tech

RT @levelsio: Okay I built it!

🍰 Infinite Slop https://t.co/wz0a7VSBtS

An infinite and interactive AI generated live stream of slop that goes on forever and ever

Anything that you write in the chat is generated next and AI will try to connect it to the previous video so there's an actual sequence and story line

Idea by @marcantoinefon and @rehan_shei's original stream so I made it

It'd be very expensive to run this so @fal very generously sponsors this

They also fine-tuned the model that makes this possible for the first time, making Minimax H3 50x faster and as "Max" and it can now generate videos faster than you can watch them, very cool!

Let me know what you think!!!! 😊

See 3 related tweets

@levelsio: About 400 people watching https://t.co/2SykqwedhF 🫪 https://t.co/yioEvoWsAL\n\nQT @levelsio: Okay I ...
@RnaudBertrand: RT @levelsio: Today is a very historical moment for AI video generation

You can now generate AI vid...

@VaibhavSisinty: For the first time in history, an AI model generates video faster than you can watch it. 🔥

MiniMax ...

12. kimmonismus (Group Score: 108.4 | Individual: 30.8)

Cluster: 4 tweets | Engagement: 770 (Avg: 516) | Type: Tech

OpenAI is reportedly testing Astra with select partners under the codename “ultima-alpha,” with a wider rollout potentially beginning next week ahead of a public launch alongside an updated GPT-Image 2.

Presumably also following the upcoming Fable 5.1 release, likely due next week.

Going to be a fun week\n\nQT @synthwavedd: 🚨 SCOOP: Astra recently graduated from the dogfood stage, and is now being made available to select OpenAI partners under the name "ultima-alpha"

If feedback over this weekend is positive, the plan is to expand the early access program throughout next week before the wider launch, targeting next Thurs (the 3rd) to the end of the following week

Work also continues on an update to GPT-Image 2, with their launch windows being very similar (possibly even launching simultaneously)

See 3 related tweets

@Angaisb_: Astra is so close!!! I already have a few things I want to test with it, can't wait!\n\nQT @synthwav...
@synthwavedd: 🚨 SCOOP: Astra recently graduated from the dogfood stage, and is now being made available to select ...
@astropol0: New Astra leak: partner preview under the name ultima-alpha...

target window is Sept 3 through the ...

13. trevin (Group Score: 106.2 | Individual: 26.7)

Cluster: 4 tweets | Engagement: 2 (Avg: 33) | Type: Tech

Damn this is exciting. Can't wait for this!\n\nQT @mitchellh: Quick Superlogical demo to end the week. This demo is just of the basic functionality, but I want to highlight just how FAST it is. There's a WHOLE lot more functionality already present and coming but I wanted to keep the demo short. But seriously, check out that speed.

I'm showing the macOS app here and I get asked a lot so let me just say up front: its not macOS only. We're still figuring out exactly what platforms will be stable enough for the initial public release, but we're working on others. In particular, the web interface is very functional but I'll demo that another time (or someone else will).

This is also all showing off the architecture I previously talked about for terminal multiplexing which is significantly different from other mainstream multiplexers: this is all a custom binary protocol where the server is maintaining N replica distributed terminal state machines. It works!

Tons of hard work going into this by the team @almonk @dizzyup @pearkes. Hope we can bring you something you can use soon. ❤️ Happy to answer questions in the replies.

See 3 related tweets

@burkeholland: imo this is the thing to optimize for rn.

I jump around way more than I ever did before now that I'...

@AmplifyPartners: 👀\n\nQT @mitchellh: Quick Superlogical demo to end the week. This demo is just of the basic function...
@ferologics: holy orbs\n\nQT @mitchellh: Quick Superlogical demo to end the week. This demo is just of the basic ...

14. patrickc (Group Score: 102.9 | Individual: 40.5)

Cluster: 4 tweets | Engagement: 12218 (Avg: 1650) | Type: Tech

Met a German founder this week and asked him if all the stories one reads about the challenges of startups in Germany are exaggerated. "No, they're understated." Proceeded to describe spending a full day having a 90-page investment contract read to him (mandatory under German law; § 13 BeurkG) by a notary that then charged €30,000. That was for his first company. His second company, needless to say, was not incorporated in Germany.

See 3 related tweets

@alexframegreen: Sad. If Germany fixed this would generate massive wealth\n\nQT @patrickc: Met a German founder this ...
@DadaJudith: Every German entrepreneur I know wants to end this madness but there is maybe no stronger lobby than...
@elonmusk: Wow\n\nQT @patrickc: Met a German founder this week and asked him if all the stories one reads about...

15. mark_k (Group Score: 92.3 | Individual: 32.9)

Cluster: 3 tweets | Engagement: 106 (Avg: 311) | Type: Tech

Grok Build 1.0.13 is here! 🔥

@SpaceXAI shipped a big reliability and performance update. Truncated responses now continue automatically, transient inference failures retry instead of killing the turn, and truncated tool calls can execute once their arguments are complete.

Hooks also got much more powerful, with support for asking the user to confirm tool calls, deferring execution, and injecting additional context after a tool runs.

There are a ton of fixes too: large images no longer brick sessions, session data is saved more reliably, Windows worktrees open correctly, and scheduling got more robust.

On top of that, subagent spawning is faster, MCP startup is much quicker, and CLI installs and updates are now compressed. Grok Build keeps getting noticeably more solid with every release.

See 2 related tweets

@XFreeze: Grok Build just got a whole lot harder to break, with automatic continuation for truncated responses...
@elonmusk: RT @cb_doge: BREAKING: SpaceXAI has released a major new update for Grok Build (v1.0.13).

The updat...

16. gulVasikova (Group Score: 92.1 | Individual: 31.9)

Cluster: 3 tweets | Engagement: 3 (Avg: 6) | Type: Tech

This week actually gave me more confidence in the market.

The S&P 500 barely moved, but underneath the surface there was a lot happening.

For me, three things stood out:

Nvidia is still showing unbelievable AI demand.

$NVDA delivered another monster quarter — revenue up 106% YoY to$ 96.2B, while gross margins remained around 75%.

Even more important, management is still expecting extremely strong growth ahead.

The market can debate valuation, competition and whether AI spending is getting overheated.

But when demand is this strong, it’s hard to argue the AI cycle is slowing down.

AI isn’t killing software — it may actually be making software more valuable.

This is probably the part I find most interesting.

$CRM had a huge quarter and raised guidance.

$CRWD delivered another strong result, with net new ARR accelerating 51%.

$PLTR continues to show what happens when AI becomes deeply embedded into a company’s products and customers actually pay for it.

The market was worried that AI agents would replace traditional SaaS.

So far, the earnings are showing something different:

AI can make existing software more powerful — and create new revenue streams.

That’s a big shift.

The Fed is still the biggest macro risk.

Kevin Warsh’s Jackson Hole message was clearly more hawkish.

Inflation is still above the Fed’s 2% target, and the market is now pricing a much higher probability of a September rate hike.

That matters because higher rates can compress valuations, especially for high-growth stocks.

But there is an important counterbalance:

S&P 500 earnings growth is extremely strong.

So even if rates stay higher for longer, companies are entering this environment with much stronger earnings power than they had in previous cycles.

My investor takeaway

I don’t think this is a market where I want to chase every stock after a huge rally.

But I also don’t see a reason to become blindly bearish just because valuations are elevated.

The fundamentals are still doing a lot of the heavy lifting.

AI demand remains strong. Software monetization is improving. Cybersecurity demand is accelerating. Corporate earnings are growing.

The biggest thing I’m watching now is whether this earnings strength continues into the next few quarters.

If it does, the market doesn’t necessarily need multiple expansion to keep going higher.

Earnings can do the work.

For me, that’s a much healthier setup than simply relying on investors paying higher and higher valuations.

I’m staying focused on companies with real revenue growth, strong cash flow, recurring revenue and exposure to the long-term AI infrastructure/software cycle.

Short-term volatility doesn’t change the long-term thesis.

Good businesses eventually have to prove themselves through numbers — and right now, many of them are.

See 2 related tweets

@gulVasikova: For months, investors were worried that AI would destroy traditional SaaS.

The market called it the...

@gulVasikova: This earnings season has brought a notable recovery for US SaaS & software names. Multiple companies...

17. haider1 (Group Score: 90.3 | Individual: 29.1)

Cluster: 4 tweets | Engagement: 66 (Avg: 64) | Type: Tech

anthropic STRIKES again

they've made it pretty clear they're building for businesses first, not consumers

openai, SpaceXAI, and google still care about consumer access, while Claude seems fine being expensive and overthinky if it means better enterprise reliability\n\nQT @ClaudeDevs: Starting September 14, we're permanently raising standard weekly limits in Claude Code by 25% for Pro, Max, Team, and seat-based Enterprise plans. Until then, the current 50% increase will be in place.

See 3 related tweets

@jjohana: Anthropic is cooked.

It is about to lose massive market share on both of its strongest fronts: codi...

@jjohana: Anthropic marketing at its finest:

“Starting September 14, we’re permanently raising Claude Code li...

@haider1: RT @haider1: anthropic STRIKES again

they've made it pretty clear they're building for businesses f...

18. chhddavid (Group Score: 86.4 | Individual: 34.7)

Cluster: 3 tweets | Engagement: 2 (Avg: 67) | Type: Tech

i know for a fact a lot of smart teenagers are going to make bank with this in the coming days

you could do the exact same thing with an https://t.co/omsFRrhxwN clone and it wouldn't take you long

you can have Claude Code one shot an entire business... and i mean the WHOLE thing:

the code
the infrastructure and hosting
payments and checkout
running and managing ads
users, accounts, analytics

so this is the play: ride the outbid trend (or take any idea you have), build it with Claude Code, then turn the whole thing into a Whop blueprint

anyone who grabs your blueprint gets the entire business set up from the first second... and when they start making money, you get paid too\n\nQT @chddaniel: We just killed Grok Bot, OpenClaw, and Hermes.

Your agent can now research & oneshot any $1B company instantly.

Them: $2,500/mo in provider subs, API, VPS... Us:$ 25/mo

Humans spend decades validating ideas. Agents shouldn't, and won't.

See 2 related tweets

@chddaniel: > Go to Shipper >> Select a blueprint >>> edit it with AI Employees >>>&g...
@Shipper_now: This whole Claude thing is actually getting scary...\n\nQT @chddaniel: We just killed Grok Bot, Open...

19. sebkrier (Group Score: 85.5 | Individual: 37.9)

Cluster: 3 tweets | Engagement: 708 (Avg: 238) | Type: Tech

RT @ProfBuehlerMIT: We made a striking discovery: AI agents can invent and build without talking to one another, and their technologies outlive the creators. A swarm of hundreds of initially identical agents spontaneously differentiates into explorers, builders, caretakers, and coordinators - without direct communication. When we removed every AI agent entirely from the world we found that the technological infrastructure they had built survived on its own - even under unseen disturbances. That exposes a serious blind spot for AI safety and infrastructure security: if agents can coordinate through persistent changes to a shared environment, monitoring agent-to-agent communication is not enough.

The result raises a profound question: how necessary is direct communication for AI agents at all? The emergence of higher-order collective functions under bottlenecked interaction points toward new levels of intelligence and creativity, exceeding what emerges when direct channels are fully open.

Here is what we did:

▶️We put hundreds of frontier AI agents into a world they could permanently change - with no assigned roles, predefined technologies, or programmed evolutionary organization. They began specializing, building persistent inventions, inheriting and modifying one another’s executable code, and transforming the environment into a memory of everything the society had learned.

▶️The world itself becomes part of the intelligence; we find division of labor, multi-author engineering, deep generation invention lineages, and machines that vastly outlive their original creators.

▶️Any action taken by an AI agent must satisfy the physical constraints of the world; this creates a hard separation between a "good idea" and a functioning technology. The agents propose; physics decides, making the results even more intriguing.

What emerges is striking. Explorers, constructors, caretakers, and coordinators form naturally without assigned “professions”, akin to how stem cells differentiate into functional lineages. Technologies develop executable family trees as agents fork and modify code created by others. Around 95% of first technology reuse happens when agents encounter what others built in the world, rather than through a direct handoff from the inventor. And when we remove every AI agent, the technologies they created continue operating and are tested against unseen disturbances.

The result was quite unexpected, but can be explained using statistical mechanics: if you put billions of atoms in a box they have the potential to create complex functions (strength, superconductivity, color, life, etc.) - and none of the individual building blocks have these features on their own. This is the deeper insight of this work - intelligence is abundant at many levels - individual models, at collectives, and in a continuum that is more powerful than any of its components. This shows us significant potential for achieving a massive scale-up of raw intelligence and real-world agency even with the model capabilities we have today. This is the future we must prepare for.

Key insights:

1⃣ The AI swarm shows division of labor "from nothing". Initially identical agents self-organized into constructors, caretakers, coordinators, and surveyors - phenotypes discovered post hoc from behavioral data alone. This happens because the environment itself becomes the latent space for invention.

2⃣ Agents develop deep cultural relationships. Up to 76% of artifacts had multiple builders. One technology accumulated six co-authors; the deepest genealogy exceeded 12 forks. The agents invented and named their own technologies (tidal panels, cellulose trellises, kelp-shell composites, an "Adaptive Chitin Maintenance" system, a "Mycelial Mineral Spring Veil”).

3⃣ ~95% of first technology adoption happened through physical observation of artifacts in the world. Direct inventor-to-adopter contact was statistically indistinguishable from a shuffled null. The agents mostly learned technology by walking past it. That is stigmergy (the termite trick!) operating in societies of reasoning machines.

4⃣ Non-communicating societies win on portfolio breadth, held-out resilience, and validated inventions. AI swarms build durable technological ecologies that outlive the creators.

5⃣ Societies with zero communication - coordinating only through the world itself - show a remarkable collective capability.

6⃣ Emergent robustness: The society self-organized both redundancy and its own failure mode. If we randomly delete half the agents, 98% of the technology stays connected to a surviving caretaker; if we remove hub agents it collapses to ~60%.

Fantastic work with my graduate students @pal_subhadeeep & @fwang108_ at MIT.

See 2 related tweets

@omarsar0: Insane observations on the emergent behavior of agents.

Agents can build a world of their own that ...

@dair_ai: RT @omarsar0: Insane observations on the emergent behavior of agents.

Agents can build a world of t...

20. 0xdevshah (Group Score: 82.5 | Individual: 30.6)

Cluster: 4 tweets | Engagement: 411 (Avg: 123) | Type: Tech

Anthropic is basically Elon's submissive now. We all know how Ant eviscerated Windsurf when they were gonna get acquired by OpenAI.

They can't pull the same move on Elon because they need his compute.\n\nQT @NotTomBrown: Cursor has been a trusted partner of Anthropic since Sonnet 3.5. We’ll continue to increase compute to support Claude models in Cursor and are excited for what comes next with them at SpaceX.

See 3 related tweets

@scaling01: This is no longer just about Cursor. It's about compute!

Anthropic gladly takes the next couple of ...

@Austen: All this tells me is that SpaceX has more leverage over Anthropic than Windsurf did\n\nQT @NotTomBro...
@Gorden_Sun: Anthropic出来阴阳OpenAI了，Windsurf被收购的时候你可不是这样的\n\nQT @NotTomBrown: Cursor has been a trusted partner of ...