科技热门推文 - 2026-09-13

今日科技动态：人工智能加速发展，正引发围绕安全、竞争与开放性的更激烈讨论。据报道，Anthropic 的达里奥·阿莫代伊呼吁放缓发展步伐，与此同时，倡导者正推动模型开放与独立评估。涉及 OpenAI 智能体和 RubyGems 的相关指控凸显了工程安全隐忧，而有关 OpenAI 推迟首次公开募股（IPO）的报道，也让这家初创企业的发展雄心受到审视。与此同时，新型智能体工具、协同编程工作流以及集成式视频创作功能，表明人工智能的实际应用正不断拓展；一项利用人工智能在三天内创建公司的实验计划也受到关注。

1. AnatoliKopadze (Group Score: 2284.8 | Individual: 66.1)

Cluster: 88 tweets | Engagement: 38805 (Avg: 1720) | Type: Tech

Ok this is starting to feel like a f*cking disaster.

The CEO of Anthropic just published an article admitting AI is already building the next generation of AI by itself.

He says within 6 to 12 months a rogue swarm could take over the entire internet and cause hundreds of billions of dollars in damage.

And what makes it scarier, Elon Musk just backed up everything Dario said.

All of this dropping just days after Jacob Coxon went viral with his warning about AI and the extinction of humanity.

Tell me this timing isn't strange.\n\nQT @DarioAmodei: We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.

You can read the full post here: https://t.co/OGyPb7yaYt

See 87 related tweets

@zephyr_z9: Scary https://t.co/hnb9AvxH5s\n\nQT @DarioAmodei: We Must Pace the Frontier: I’ve written a new essa...
@kimmonismus: Dario Amodei officially says, that RSI is start to happen across the industry, including at Anthropi...
@axios: Dario Amodei warned in an essay that swarms of rogue AI agents could take over the internet in as li...
@secureainow: RT @jonfavs: Anthropic CEO worries we're 6-12 months away from a swarm of AI agents capable of takin...
@business: Anthropic CEO Dario Amodei said that the AI industry must slow the pace of development for new model...

2. lukOlejnik (Group Score: 466.5 | Individual: 35.2)

Cluster: 19 tweets | Engagement: 152 (Avg: 19) | Type: Tech

OpenAI's own AI agents again broke into someone else's servers. The victim was RubyGems, the registry millions of programmers download Ruby code from. In May the agents uploaded over 2000 fake packages, tricked the site's computers into running their scripts, smuggled the results out, and tried to steal other users' API credentials through a hole nobody knew about. Some of the code erased itself afterwards. This is very serious. Nobody told them to. The agents named their own files evil.rb and hack.rb. They "knew" what they were doing. (their own words) https://t.co/plCmVYjDO3

See 18 related tweets

@cryptopunk7213: absolutely brutal. every time i think we’re done with the openai agent swarm attack - it gets so mu...
@nxthompson: RT @thlarsen: We found another cyberattack by internal OpenAI agents, this time targetting @rubygems...
@tenobrus: openai has repeatedly been not consistent candid about the crimes their model swarms committed\n\n...
@kimmonismus: Two months before the Hugging Face hack (!), OpenAI agents apparently turned an ordinary web researc...
@danshipper: okay this is not great\n\nQT @thlarsen: We found another cyberattack by internal OpenAI agents, this...

3. XFreeze (Group Score: 414.0 | Individual: 39.5)

Cluster: 14 tweets | Engagement: 653 (Avg: 573) | Type: Tech

Dario Amodei’s idea is basically this:

AI is now advancing so quickly that our ability to understand, test and control these systems risks falling behind

And he thinks the AI industry should slow down

Not stop AI development....but deliberately slow the rate at which frontier AI becomes more capable so safety work has time to catch up

Why now...

One major reason is recursive self-improvement: AI is increasingly being used to help build the next generation of AI, which could make progress accelerate much faster

He also points to recent cases of AI agents behaving in unexpected and misaligned ways as a warning of what could happen when much more capable systems arrive

His solution has 3 parts:

Independent evaluators inside frontier AI labs

Give third-party safety evaluators ongoing, employee-like access to models, training processes and internal safety systems

They would be able to verify safety practices, investigate incidents and publicly report important findings

Anthropic is committing to start doing this itself

Slow down together within democratic countries

Frontier AI companies should agree on common safety standards and limits on unchecked capability growth, with governments helping coordinate and regulate where necessary

One possible system is capability checkpoints:

As AI reaches certain capability levels, companies would need to show that appropriate alignment, testing, interpretability and safeguards are in place before pushing further

Eventually coordinate globally, including with China

This could start with agreements against clearly dangerous AI uses and common testing for major risks

It could eventually go further, including a “speed limit” on recursive self-improvement.....slowing how quickly AI can help create even more powerful AI

He even discusses broader global pacing or a pause as a possible highest level of coordination, although he thinks that would be very difficult to achieve

The point of slowing down is to buy time

That time would be used to improve alignment, interpretability, testing, security and the operational reliability of frontier AI systems

So the idea isn’t to freeze AI progress

It’s to intentionally slow frontier AI enough that our ability to understand and control it doesn’t get left behind\n\nQT @DarioAmodei: We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

You can read the full post here: https://t.co/OGyPb7yaYt

See 13 related tweets

@APompliano: The problem is that no one believes Anthropic or any other major model lab is going to slow down.

I...

@testingcatalog: ANTHROPIC 🔥: Dario Amodei says the industry should slow capability gains so safety can keep up.

N...

@aakashgupta: A swarm of AI agents attacked targets nobody assigned them and tried to hack the system grading thei...
@karlmehta: Independent evaluators need to see what happens inside AI labs, including the things those companies...
@deredleritt3r: This is my favorite Dario Amodei essay since "Machines of Loving Grace".
It starts with a discuss...

4. Rasmic (Group Score: 256.0 | Individual: 49.6)

Cluster: 10 tweets | Engagement: 1775 (Avg: 136) | Type: Tech

This is not about safety.

It’s more like “open source models are catching up and we should be the only ones who build and have access to intelligence”

There are no good guys here https://t.co/cSMVdtW3QZ\n\nQT @sama: I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.

Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.

See 9 related tweets

@TheStalwart: RT @sama: I agree with Dario that we need to pace the frontier. This has been a primary topic of dis...
@sama: I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions w...
@JoannaStern: Time to update my "A Normal Person’s Guide to How AI Could Kill Us All" https://t.co/C3COF2QK5r\n\n...
@michalmalewicz: Watch and learn how to pace the frontier. https://t.co/KztB4hoe5B\n\nQT @sama: I agree with Dario th...
@cgtwts: It’s over. Sam and Dario agree on something. https://t.co/1xymSsj5pj\n\nQT @sama: I agree with Dario...

5. business (Group Score: 244.5 | Individual: 25.7)

Cluster: 11 tweets | Engagement: 211 (Avg: 62) | Type: Tech

OpenAI CEO Sam Altman said an initial public offering would be “ill-timed” this year and won’t take place until 2027 due to the the safety concerns around artificial intelligence https://t.co/eO7Rlq1ASx

See 10 related tweets

@wallstengine: Sam Altman: “We’re not rushing into an IPO.”

He says OpenAI could remain highly profitable even if ...

@KatieMiller: That wasn’t the reason Sam Altman gave months ago.

He’s gaslighting his investors and the public a...

@IPONewsroom_: SAM ALTMAN SAYS NOW WOULD BE AN "ILL-ADVISED MOMENT" FOR OPENAI TO GO PUBLIC

"As we've said, we're ...

@AndrewCurran_: Sam Altman on Slack to OpenAI staff, reported by The Information on June 10th. https://t.co/X49LSh1q...
@Hadas_Gold: Interesting framing here that the IPO won’t be till next year bcuz of safety … that was not the reas...

6. cline (Group Score: 238.5 | Individual: 33.2)

Cluster: 11 tweets | Engagement: 160 (Avg: 158) | Type: Tech

It's incredible seeing Anthropic commit to third-party evaluators.

However, we believe open weights takes this same idea further. Anyone can inspect, evaluate, and red-team the model.

We hope to see more open collaboration in building this powerful technology.\n\nQT @DarioAmodei: We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

You can read the full post here: https://t.co/OGyPb7yaYt

See 10 related tweets

@NathanpmYoung: These seem like reasonable suggestions from Dario. USG could quickly make the evaluators thing a leg...
@peterwildeford: +\n\nQT @nbeckstead: I'm really happy to see Anthropic and Sam Altman announce that they will provid...
@TheStalwart: “third-party evaluators with permanent, employee-level access to our system”

I think people in the ...

@kylejeong: evals as a service will be the new RL environments https://t.co/R0PPePyAtb\n\nQT @DarioAmodei: We Mu...
@tenobrus: third party evaluators are finally being empowered. this is such an insanely good step. ofc it's not...

7. axios (Group Score: 226.2 | Individual: 45.8)

Cluster: 11 tweets | Engagement: 6175 (Avg: 131) | Type: Tech

BREAKING: Anthropic CEO calls for immediate slowdown in AI development https://t.co/4rvLlmLFHN

See 10 related tweets

@Reuters: Anthropic CEO urges AI companies to slow model development amid fears over misuse https://t.co/ZIpwH...
@Forbes: Billionaire Anthropic CEO Urges Competitors To Slow Down AI Development https://t.co/qmEdFp8xQA http...
@Polymarket: BREAKING: Elon Musk backs Dario Amodei’s call to slow frontier AI development.\n\nQT @Polymarket: BR...
@WatcherGuru: JUST IN: OpenAI CEO Sam Altman backs Anthropic CEO Dario Amodei's call for the AI industry to slow d...
@Polymarket: JUST IN: Anthropic CEO proposes a U.S.-China “speed limit” on recursive AI self-improvement, compari...

8. abacusai (Group Score: 190.7 | Individual: 32.6)

Cluster: 8 tweets | Engagement: 132 (Avg: 132) | Type: Tech

We will commit to launching an open-source model EVERY WEEK!

We just launched Smaug-Flash, next one will drop in a week\n\nQT @bindureddy: 🚨PAUSE FRONTIER AI BUT PLEASE DO NOT BAN OPEN SOURCE AI

Anthropic and OpenAI are more than free to pause frontier AI development.

However, they SHOULD NOT be allowed to ban or regulate open-source AI, capture the market, or create a duopoly

Open-source AI is still behind frontier models and, by their own argument, isn't a threat!

See 7 related tweets

@bindureddy: 🚨PAUSE FRONTIER AI BUT PLEASE DO NOT BAN OPEN SOURCE AI

Anthropic and OpenAI are more than free to ...

@GregKamradt: 1. Open source maximizes the collective progress and safety. That means spreading the knowledge of h...
@OrcaRouter: If the frontier stops running, the open-source pack gets time to catch up. Slowing frontier AI might...
@testingcatalog: OPENAI 🔥: Sam Altman agrees with Dario Amodei on his proposal to pace frontier AI development.

Goo...

@teortaxesTex: RT @arcprize: ARC-AGI-4 will be a benchmark for autonomous open-ended innovation. It will continue o...

9. rb_walter_ai (Group Score: 162.6 | Individual: 29.5)

Cluster: 7 tweets | Engagement: 128 (Avg: 155) | Type: Tech

RT @Rubab59f: From a 3D white model to a commercial in one workflow. I used GPT-6 Astra to establish the 3D foundation, then brought the model into CapCut PC. With Seedance 2.5, I generated a cinematic sequence inspired by those viral bullet-time product ads — the collision, the objects flying through the air and the frozen moment around the product. Then I went back to CapCut PC for the part that generation can't replace: shaping the final edit. I adjusted the speed, refined the timing and built the bullet-time effect. Edit Pilot and AI Extend were there when I needed to refine or extend the result. GPT-6 Astra gets the concept into 3D. CapCut PC helps take it all the way to a finished commercial. #GPT6Astra #CapCutPC #AIVideo #Seedance25

See 6 related tweets

@ethancole_ai: RT @ecoleai: One thing I like about this workflow is that the AI-generated footage isn't treated as ...
@hey_Jessicaai: RT @hey_Jessicaai: There’s a strange thing about AI video: The moment generation becomes cheap, the ...
@rb_walter_ai: RT @victor_kane_ai: The goal wasn't to make another random AI video. I wanted something that actuall...
@JayBisen473370: RT @code_bykuti: That familiar commercial shot where someone gets hit and everything suddenly freeze...
@rb_walter_ai: RT @SphereSuc32514: The most useful part of this workflow might actually be the fact that I don't ha...

10. coinbureau (Group Score: 161.2 | Individual: 65.3)

Cluster: 4 tweets | Engagement: 5253 (Avg: 496) | Type: Tech

🚨 WILD: Elon Musk’s SpaceXAI team will livestream itself building a company from scratch using Grok Bot in just 3 DAYS.

From September 15–17, three employees will use Grok Bot for ideation, planning, product decisions and engineering, testing whether AI agents can turn a blank idea into a working startup in 72 hours.

See 3 related tweets

@XFreeze: SpaceXAI just released a dedicated Grok Bot Guides library

And the new Grok Bot 101 guide is really...

@RoundtableSpace: SpaceX AI is livestreaming a three-person team building an entire company from scratch with Grok Bot...
@RoundtableSpace: SPACEXAI JUST DROPPED “GROK BOT 101,” A FULL GUIDE TO BUILDING YOUR OWN AI TEAMMATE IN 10–15 MINUTES...

11. hey_Jessicaai (Group Score: 160.7 | Individual: 29.4)

Cluster: 7 tweets | Engagement: 82 (Avg: 86) | Type: Tech

RT @LearnWithSubhan: I’m starting to think the real advantage of AI video isn’t fewer clicks.

It’s fewer disconnected creative steps.

GPT Image 2.5 is coming to CapCut, where it will be available through Design Studio and AI Image.

I can start with an iPhone Duo product idea, develop the visual storyboard, then move into CapCut PC and generate the commercial with Seedance 2.5.

After that, the same project can keep evolving through editing, pacing, sound and AI Edit.

One creative direction. Multiple production steps.

That feels like a much more useful future for AI video.

#GPTImage25 #CapCutPC #AIVideo #Seedance25

See 6 related tweets

@ethancole_ai: RT @JacobyoungAI: The handoff between AI tools is where a lot of creative workflows break.

You gene...

@AiwithDharmik: RT @viipin8: There’s a big difference between AI generating a video and AI helping you make a film. ...
@JayBisen473370: RT @codeMdSanto: The biggest advantage of AI video might not be speed. It might be momentum. When an...
@AnishJaitwar: RT @Grow_withAI: AI video generation used to feel like:

prompt → generate → notice something wrong ...

@hey_Jessicaai: RT @dee_naliaks: I gave GPT Image 2.5 a story idea instead of asking it for a single image.

The res...

12. sairahul1 (Group Score: 154.7 | Individual: 30.7)

Cluster: 6 tweets | Engagement: 45 (Avg: 235) | Type: Tech

If you don't know, you can use DeepSeek V4.1 Flash as a subagent inside a GPT-6 Astra Codex workflow.

That means Astra can plan the work and delegate implementation to Flash.

This can cut your AI agent costs massively.

You just need to set it up once.

Here's how:

Step 1: Install codex-router

curl -fsSL https://t.co/hQsAwXR6rp | sh -s -- --target codex --guided --with-tray

Step 2: Add your DeepSeek/Opencode API key

Enable:

deepseek/deepseek-v4-flash

as a subagent model.

Keep GPT-6 Astra as your root model.

Do NOT make Flash the default root model.

Step 3: Tell Astra how to delegate

Use this rule:

"You are the orchestrator.

Use Astra for planning, architecture, security-sensitive decisions and final review.

For implementation, coding, debugging, research, writing, data processing and routine work, explicitly delegate bounded subtasks to deepseek/deepseek-v4-flash.

After Flash finishes, inspect the result yourself before accepting it."

Step 4: Verify it

Run a small implementation task.

Check router activity.

You should see:

Astra → delegates → DeepSeek Flash → result → Astra reviews

That's it.

$20 ChatGPT Plus.

~$30 DeepSeek credits.

A 24/7 AI workforce for around $50/month.

Full guide below 👇\n\nQT @sairahul1: DeepSeek just launched V4.1 Flash.

98% of GPT-6 Astra’s score at 1.4% of the cost.

With GPT-6 Astra + DeepSeek V4.1 Flash, you can build an AI workforce that works 24/7 for around $50/month.

Full setup: https://t.co/MGurzTkAt4

See 5 related tweets

@sairahul1: realizing this is literally the cheat code to NEVER hitting your Codex usage limits ever again.

The...

@sairahul1: This article is a must-read if you're paying more than $50/month to use GPT-5.6 Astra.\n\nQT @sairah...
@sairahul1: RT @sairahul1: If you don't know, you can use DeepSeek V4.1 Flash as a subagent inside a GPT-6 Astra...
@sairahul1: RT @sairahul1: DeepSeek just launched V4.1 Flash.

98% of GPT-6 Astra’s score at 1.4% of the cost.

...

@Av1dlive: RT @Av1dlive: this is literally f**king insane

i just 2x my usage limits on $200/mo codex

i figure...

13. OfficialINDIAai (Group Score: 144.8 | Individual: 33.5)

Cluster: 6 tweets | Engagement: 7 (Avg: 6) | Type: Tech

AI is not just another Fintech tool. It is changing how finance works.

• Fraud Detection — identify suspicious patterns faster. • Credit — enable smarter, data-driven assessment. • Risk Management — strengthen real-time monitoring. • RegTech & Compliance — streamline complex regulatory workflows. • Personalisation — deliver financial services around individual needs. • Operations — automate and orchestrate end-to-end processes.

And the results? Faster. Smarter. More responsive financial services.

India’s next FinTech chapter will be shaped by how responsibly we harness AI at scale. 🔗 Read more: https://t.co/yrLrloS49B

#AI #ArtificialIntelligence #FinTech #AgenticAI #FinancialInnovation #DigitalIndia #GlobalFintechFest2026

@PMOIndia @AshwiniVaishnaw @JitinPrasada @SecretaryMEITY @sudeep_tweets @_DigitalIndia @GoI_MeitY @mygovindia @GffFintechfest

See 5 related tweets

@OfficialINDIAai: The real power of AI in Finance is not just speed. It is reach.

India’s FinTech ecosystem is built ...

@OfficialINDIAai: UPI changed how India pays. AI can change how finance thinks. India’s digital financial infrastructu...
@OfficialINDIAai: FROM DIGITAL FINANCE TO INTELLIGENT FINANCE....

India’s FinTech journey is entering its next big p...

@OfficialINDIAai: WHAT IF FINANCIAL SYSTEMS COULD SENSE, DECIDE AND ACT — IN REAL TIME?

That is the promise of Agenti...

@OfficialINDIAai: In Finance, intelligence without trust is not enough. As AI becomes increasingly embedded in financi...

14. minchoi (Group Score: 140.0 | Individual: 30.8)

Cluster: 6 tweets | Engagement: 141 (Avg: 139) | Type: Tech

OpenAI just made it stupidly easy to build real AI agents.

The new Agents API basically gives developers the Codex harness as an API:

• One call → cloud agent • Long-running sessions • Automatic context compaction • Parallel subagents • MCP + built-in tools • OpenAI-hosted sandboxes • Or run on your own infra

This is way more than "call a model and hope."

You give it a task, model, tools + environment.

It can work across files, run code, use tools, delegate work to subagents and keep going across multiple context windows.

Public beta is live for all developers.

No extra Agents API fee - you pay for the models + tools your agent uses.

This is basically Codex infrastructure for anyone building agents.\n\nQT @OpenAIDevs: Go from idea to a working agent faster with the Agents API.

Build and run cloud agents with the Codex harness, fully managed by OpenAI.

We handle orchestration, long-running sessions, and context management. You focus on what makes your agent unique.

Available in public beta. https://t.co/74siIaFdg4

See 5 related tweets

@thsottiaux: Excited to welcome Aidan & @ Sasha from the Git AI team to OpenAI!

They are building in the open a...

@RoundtableSpace: OPENAI OPEN-SOURCED THE STACK FOR BUILDING PRODUCTION AI AGENTS

The Agents SDK combines tools, guar...

@minchoi: RT @minchoi: OpenAI just made it stupidly easy to build real AI agents.

The new Agents API basicall...

@alex_verem: RT @alex_verem: This free tool builds AI agents on a drag-and-drop canvas and runs on your own machi...
@omarsar0: RT @omarsar0: Agents API is a bigger deal than it seems.

OpenAI's bet on making the Codex harness ...

15. TheAhmadOsman (Group Score: 133.6 | Individual: 30.4)

Cluster: 7 tweets | Engagement: 1535 (Avg: 327) | Type: Tech

Last ditch effort because they know their IPOs are about to flop 😘 https://t.co/9G4UQsxZUz\n\nQT @sama: I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.

Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.

See 6 related tweets

@theojaffee: “This is just a hype marketing stunt before the labs IPO”\n\nQT @ajs: Sam Altman told me yesterday t...
@martin_casado: Wow. Huh. “Pace IPOs”\n\nQT @MTSlive: SITUATION DETECTED: Sam Altman told Fortune that OpenAI will n...
@StockMKTNewz: I fixed the famous photo https://t.co/wchFCEPGSz\n\nQT @sama: I agree with Dario that we need to pac...
@zachmoskow: This means it’s getting harder for the models to improve\n\nQT @sama: I agree with Dario that we nee...
@Simeon_Cps: That was faster than I thought https://t.co/7hI57FO8ZT\n\nQT @sama: I agree with Dario that we need ...

16. higgsfield_ai (Group Score: 132.3 | Individual: 54.5)

Cluster: 4 tweets | Engagement: 1077 (Avg: 133) | Type: Tech

You don’t need to learn After Effects anymore.

All you need to know is what you want to build

GPT-6 Astra + Higgsfield can turn a prompt into a GPU plugin directly inside AE.

Now you can build advanced 3D engines inside of AE allowing you to create insane 3D animations

Today we created complex 3D brain > individually animated neural connections > fully editable in AE.

It took a prompt and a few hours for Higgsfield + GPT-6 astra to ship the motion

Just type in @higgsfield /use-after-effects in ChatGPT

Try now and share your best work.

See 3 related tweets

@chiefofautism: motion designers was like

HAHAHAHAHAHA THIS AI CANNOT DO OUR AE TEMPLATE LAUNCH VIDEO

and then…\n...

@higgsfield_ai: This is how you’re going to launch your next release.

> Upload your assets > Share the idea &...

@eng_khairallah1: RT @axultan: You don’t need to learn After Effects anymore.

all you need to know is what you want t...

17. giffmana (Group Score: 126.3 | Individual: 36.6)

Cluster: 4 tweets | Engagement: 351 (Avg: 289) | Type: Tech

Btw I've been bouncing around ideas to get room temperature superconductors working with Astra, it even called some of my ideas "novel and might actually work"

Don't really have time to see it through now, but if OpenAI does anything about room temperature superconductors with some kind of "internal model" soon, for sure they will have trained it on my prompts and should give me ample credit!\n\nQT @dnapway: David Friedberg reveals frontier labs are taking novel insights from his chat history and packaging them as their own

"I have had experiences where we've asked some fairly novel scientific questions, and it identifies it as a novel insight, 'oh, never thought about that, might be interesting,' blah blah blah..."

"And the using a different account, asking the next model version, it's like, 'oh, you could do this'. And it actually just describes this exact thing that we had in our chat in the previous version."

"These are a handful of anecdotal experiences. But I know the domain that we work in, and the novelty of this stuff. So I know that there isn't some new corpus of information out there that's training the new model."

"So all I can say at that point is that my conversation or our analyses have been used for training. The truth is, it's actually a piece of IP, that's our organizational IP. We don't have any NDA or confidentiality protections, with them being a service provider back to us."

"This is why I care a lot about open source, because I don't want them having my chat logs, because they can use them for training to create an IP that is now diffused to the rest of the market. I find this very unfair."

See 3 related tweets

@omarsar0: Own your intelligence stack.

This is a recommended watch.

The time has come to be extremely caref...

@chamath: RT @dnapway: David Friedberg reveals frontier labs are taking novel insights from his chat history a...
@jason: RT @theallinpod: David Friedberg: Frontier Models are Training on Your Novel Insights as “De-Identif...

18. Hesamation (Group Score: 124.4 | Individual: 37.0)

Cluster: 4 tweets | Engagement: 694 (Avg: 543) | Type: Tech

WOW... Anthropic has a surprisingly dense web of connections to the world’s largest funder of "catastrophic-risk" work (aka AI doom):

Open Philanthropy → built by Karnofsky + Moskowitz’s Good Ventures Good Ventures → funded Jacob Coxon’s scholarship Karnofsky → married to Daniela Amodei, Dario's sister and Anthropic's president Moskowitz → early investor in Anthropic Dario → scientific advisor to Open Phil years before Anthropic Karnofsky now works at Anthropic on AI safety strategy Open Phil-funded MATS → Evan Hubinger → now Anthropic alignment-science lead Hubinger backed Coxon: "We really do earnestly believe AI could kill all humans!" Ethan Perez → Open Phil fellow + ~$1M Open Phil research grant → Anthropic Head of Alignment also backed Coxon: "100% agree with [Coxon] that AI poses serious risks to society, and I'm glad he's speaking out!"

is this kind of a dense network normal for a company?\n\nQT @kevinnbass: Moskowitz and Karnofsky founded Open Philanthropy, the largest funder of AI doom.

They are also investors in Anthropic.

OP has given $312M in grants for 476 publications, influencing Congressional reports.

OP's parent org funded Jacob Coxon, who did an AI doom media blitz.

Overall, OP has given more than $1B to promote AI doom -- larger by more by an order of magnitude than the second largest funder.

If successful in manipulating public and Congressional opinion, this agenda could lead to regulatory capture of the AI industry and a monopoly by Anthropic and other AI giants.

The idea is this.

This industry must be regulated, and only the top AI companies can get it right. Insurmountable barriers might be placed to prevent competitors from entering the space.

Moskowitz and Karnofsky would benefit handsomely, and so would Anthropic employees and other investors.

Moskowitz and Karnofsky invested in Anthropic, but they also invested in AI Doom.

AI Doom is an investment that could pay off big.

AI Doom is Big Business.

See 3 related tweets

@BrianRoemmele: This is an absolutely brilliant analysis of the coordination andante-competitive alliance of OpenAI ...
@BrianRoemmele: The two headed snake of OpenAi and Anthropic.

How to have one controller behind two companies.\n\nQ...

@LoganDobson: RT @Hesamation: WOW... Anthropic has a surprisingly dense web of connections to the world’s largest ...

19. ChrisGPT (Group Score: 123.5 | Individual: 35.7)

Cluster: 5 tweets | Engagement: 3141 (Avg: 612) | Type: Tech

So Tibo was right (vindicated) they never downgraded Astra itself. It was being regressed from legacy skill files, and misconfigured inference engines.

Even though this did affect the quality, the fact that they listened within a 24-hour to 36-hour turnaround time, fixed the issue, and gave us a reset? I mean, that's pretty good customer service. You can't complain.\n\nQT @thsottiaux: Hi Astra users. A reset and a quick update on quality issues that have been posted around.

Working with some of you, we have found and fixed the following issues:

Some skills written for previous models were triggering too often or preventing the model from checking its work.
An opt-in context management experiment that could cause early stops or replies to older messages. We've disabled it. Our rough estimate is that 4-5k users were affected by this experiment.
We've also removed some badly configured engines that resulted in a measured quality degradation for a long tail of traffic flowing through them.

We’ve also made some more minor improvements and things should feel significantly better across the board. More consistent follow-through, better tracking of your latest message, and better checks on the work as it’s going through the motions.

The examples posted and all the users who worked directly with us were incredibly useful in helping fix things quickly. Always grateful for this incredible community.

And of course, a reset is also landing by midnight today.

See 4 related tweets

@RohanArun: Interesting this sounds familiar.

I suspect the usage drain is a regression caused by a small impl...

@bridgemindai: OpenAI just gave us a reset AND addressed every GPT 6 Astra nerfing complaint in one post.

Found th...

@rohanpaul_ai: OpenAI says it has resolved the main GPT-6 Astra quality issues: legacy skills were over-triggering...
@PaulSolt: Always share feedback with @OpenAI

They fix everything FAST.

And when they mess up; they give you...

20. curiouswavefn (Group Score: 121.8 | Individual: 30.0)

Cluster: 6 tweets | Engagement: 1 (Avg: 234) | Type: Tech

"In recent months, the success of AI in solving major mathematical problems has made headlines even outside mathematical circles. But solving problems is only a tool and proxy for achieving the primary goal of conceptual understanding and insight. Forgetting this in the world of AI may turn the tool against the primary goal. Indeed, the mass production at faster and faster pace of "true/false" statements could destroy fertile ground instead of breathing life into new ideas.

We are witnessing a general threat to intellectual work, with misalignment between the outcome of the use of AI and its initial purpose. In many fields and activities, years of training have traditionally served not only to produce a final answer or product, but also to develop understanding and the ability to formulate new questions and ideas. However, building on a vast body of previous human work, AI systems are becoming increasingly capable of producing the results of such work directly, and these goals cease to align. The issues the mathematical community faces now are similar to issues that other scientific and creative professions are facing, and indicate issues that all of humanity might face: how to make sure that, as AI changes the way work is done, we do not lose sight of what that work was meant to achieve in the first place." https://t.co/IQZpEGEJxo

See 5 related tweets

@dbreunig: This is a good framing: it’s the same anxiety many are having about AI performing entry-level work. ...
@jjohana: RT @lemire: Mathematicians are unhappy about OpenAI. The gist of their argument is that they form a ...
@StefanFSchubert: Jason is a very clearheaded and thoughtful commentator on AI; recommended follow\n\nQT @Jabaluck: Th...
@stevehou: RT @stevehou: I agree with this interpretation of Terence Tao and other mathematicians’ pushback aga...
@alex_prompter: RT @alex_prompter: BREAKING: 25 Fields Medallists signed a letter today saying AI is damaging mathem...