- Published on
科技热门推文 - 2026-09-07
- Authors
- Name
- geeknotes
今日科技动态方面，OpenAI 成为讨论焦点，相关话题包括据报道其在自动化 AI 研究方面取得的进展、研究人员面临的推理成本上升，以及 GPT-6 Astra 在编程、设计和创意制作等领域的演示。围绕更多 GPT-6 模型的猜测，也伴随着对 AI 社会风险与人类自主权的广泛讨论。工程工具正朝着更深入的智能体集成方向发展：Polygres MCP 有望提供数据库访问能力，Apodex 1.1 则推出了开放的智能体框架和模型权重。与此同时，德里围绕初创企业政策展开的讨论强调，应与创业者合作，支持当地创业生态系统发展。
1. MTSlive (Group Score: 362.7 | Individual: 48.9)
Cluster: 12 tweets | Engagement: 2586 (Avg: 151) | Type: Tech
SITUATION DETECTED: OpenAI has announced that it has reached its goal of an automated AI research intern, and is on track for a fully automated AI researcher by March 2028.
See 11 related tweets
@deredleritt3r: Key take-aways for me from today's OpenAI announcements:
OpenAI "making strong progress" toward c...
@firesidealpha: OpenAI Chief Scientist warns that no one is prepared for a continued rapid rise in machine intellige...
@AndrewCurran_: Good summary of this morning's events.\n\nQT @deredleritt3r: Key take-aways for me from today's Open...
@haider1: openai has reached its "automated research intern" milestone
and says AI agents are now materially ...
- @kimmonismus: OpenAI has officially achieved their goal of automated AI research interns.
Fully automated ai res...
2. firesidealpha (Group Score: 313.7 | Individual: 30.8)
Cluster: 12 tweets | Engagement: 34 (Avg: 42) | Type: Tech
OpenAI reveals its median researcher now burns more than 7,000.
"At the start of this year, the median researcher ranked by agent usage at OpenAI was using coding agents only in modest amounts."
"By mid-August, the median researcher was integrating agents daily into their work, using more than $600 per day of inference at API prices."
"The 90th percentile user in our research organization now uses more than $7,000 of tokens per day."\n\nQT @kliu128: Today we're releasing data on models accelerating research at OpenAI.
Recursive self-improvement could be the most important contributor to AI capabilities over the next few years, but by default it will only be seen inside a few frontier AI labs. Being transparent is more urgent than ever, so we can inform the public discussion on whether and how to pace model development. I ask other AI companies to do the same.
See 11 related tweets
- @deanwball: RT @kliu128: Today we're releasing data on models accelerating research at OpenAI.
Recursive self-...
- @firesidealpha: OpenAI admits it does not yet know how to safely reach aligned, full RSI.
"We do not yet know how t...
- @ShanuMathew93: Median research spending 7000/day!
"In terms...
- @TMTLongShort: Long compute 🚀 https://t.co/2DKvccgJvA\n\nQT @kliu128: Today we're releasing data on models accelera...
- @petergostev: What's striking about this chart is how few people, even at the cutting edge, have any experience sp...
3. daleverett (Group Score: 176.1 | Individual: 31.0)
Cluster: 6 tweets | Engagement: 25 (Avg: 17) | Type: Tech
Ai retrieval that feels like extended context.
Give your agents the whole database\n\nQT @daleverett: First: yes, the mic is a Monster can (Was sleepy)
Second: we just launched Polygres MCP. (Not Sleepy)
AI clients can now connect directly to Polygres with browser sign-in, scoped project access, granular permissions, and confirmation before important changes. It lowkey feels like magic.
The MCP catalog includes 91 tools and adapts to your role, project, and current state.
We’re also shipping Agent Skills 0.6.0 with production MCP support and all five Polygres skills.
Everything is live today -> https://t.co/Osu8LGArWL
by: @daltonmeon @damienhe @polygres backed by: @ycombinator @fdotinc
See 5 related tweets
- @daleverett: If u wanna be my gtm engineer, tell me how I could improve this feature launch
Hiring only the best...
- @damienhe: idk how dale has the energy to film these in public LOL. there was a group of people 45 degrees from...
- @daltonmeon: Finally an mcp 😂🙏 I’m sorry it took us so long. Rmb to update your cli and skills!!! Esp if you’re o...
- @polygres: WE JUST LAUNCHED POLYGRES MCP\n\nQT @daleverett: First: yes, the mic is a Monster can (Was sleepy)
...
- @polygres: RT @daleverett: First: yes, the mic is a Monster can (Was sleepy)
Second: we just launched Polygres...
4. _NathanCalvin (Group Score: 140.1 | Individual: 57.0)
Cluster: 4 tweets | Engagement: 1478 (Avg: 73) | Type: Tech
RT @merettm: I wrote about the state of AI, why I’m concerned about the next few years, and the choices we need to make to keep the future in humanity’s hands.
An Alien Mind: https://t.co/FeIfWNe0UE
See 3 related tweets
- @sama: An important post from Jakub:\n\nQT @merettm: I wrote about the state of AI, why I’m concerned about...
- @romainhuet: We share a lot here about what AI is becoming capable of.
Our researchers are also thinking deeply ...
- @PaulSolt: AI Safety.
Is it safe?\n\nQT @merettm: I wrote about the state of AI, why I’m concerned about the n...
5. kimmonismus (Group Score: 128.5 | Individual: 27.4)
Cluster: 6 tweets | Engagement: 718 (Avg: 1146) | Type: Tech
It appears that OpenAI is already preparing the release of GPT-6 Sol.
GPT-6-Astra is their flagship model, and as is known, GPT-6 is a model family; this means Sol, Terra, and Luna are still to come.
I am currently curious about the implications of this, as the price-performance ratio of Astra-low reasoning is already outstanding, and it remains to be seen how Sol and Luna will compare to Astra-low.\n\nQT @Lentils80: Now that GPT-6 Astra is finally out, OpenAI is now internally testing GPT-6 Sol
Outputs are ofc noticeably worse than Astra in general, from testing. However it's fast and seems to also be a monster model
It made this in zero-shot, 15 minutes and 60k total tokens on Max effort
See 5 related tweets
- @Techmeme: OpenAI quietly updates its evaluation metrics for GPT-6 Astra, making changes that appear to favor A...
- @Lentils80: Now that GPT-6 Astra is finally out, OpenAI is now internally testing GPT-6 Sol
Outputs are ofc not...
- @scaling01: GPT-6 Astra was likely broadly available to OpenAI employees between July 12 and July 26
in that ti...
- @jjohana: RT @DanDr1s: OpenAI DevDay is September 29.
Likely releases:
• GPT-6 Luna • GPT-6 Terra • GPT-6 S...
- @robinebers: RT @nicbstme: GPT-6 Astra is an amazing model, but writing off Anthropic is a BIG mistake.
A lot of...
6. higgsfield_ai (Group Score: 115.7 | Individual: 29.1)
Cluster: 5 tweets | Engagement: 50 (Avg: 224) | Type: Tech
GPT-6 Astra recreated the Canvas launch video with vibecoded motion / composition in After Effects.
Fully editable, with image and video assets generated through Higgsfield MCP. https://t.co/ApOev4kYlZ\n\nQT @higgsfield_ai: GPT-6 Astra is exceptional at using professional animation tools.
It takes full control of an iPad, plugs straight into Procreate, and paints a complete portrait in comic book style. https://t.co/uxSW8tWXvM
See 4 related tweets
- @higgsfield_ai: GPT-6 Astra built this intricately detailed visual with Higgsfield and TouchDesigner.
It generated ...
- @higgsfield: Okay GPT-6 Astra is getting REALLY good at motion design.
It recreated our own launch video with Af...
- @liu8in: Astra + @HyperFrames_ or human + after effects
which is which 🙃 https://t.co/6TBOsPsDwU\n\nQT @Open...
- @higgsfield_ai: GPT-6 Astra is exceptional at using professional animation tools.
It takes full control of an iPad,...
7. innovationcncl (Group Score: 102.1 | Individual: 43.4)
Cluster: 4 tweets | Engagement: 1140 (Avg: 131) | Type: Tech
RT @innovationcncl: 🚨 TRUMP: “Whoever wins AI, wins.”
“This is bigger than the internet. This is a revolution.”
President Trump goes ALL IN on American innovation, saying the U.S. must beat China in AI as the technology is already driving breakthroughs that are transforming fields like medicine. https://t.co/qXEXJQsftq
See 3 related tweets
- @VaibhavSisinty: Trump: "Whoever wins AI, wins. This is bigger than the internet. This is a revolution."
The US pres...
- @kimmonismus: Regardless of what one thinks of Donald Trump, the USA has understood that the race for AI will dete...
- @chiefofautism: probably we are going to die\n\nQT @innovationcncl: 🚨 TRUMP: “Whoever wins AI, wins.”
“This is bigg...
8. lennysan (Group Score: 101.7 | Individual: 36.2)
Cluster: 3 tweets | Engagement: 66 (Avg: 175) | Type: Tech
"We built a technology that extends our intellect, but nothing to extend our soul." https://t.co/ByNqT0nYMQ\n\nQT @lennysan: The "permanent underclass" is a dark fantasy Silicon Valley keeps telling itself.
@illscience (GP at @a16z) points out that by almost every measure, things have never been better, and the job-loss story hasn’t played out as people thought. Radiologists have been “cooked” for 20 years and demand keeps rising. Open engineering roles are at all-time highs. And unlike the mobile era’s winner-take-all markets, every layer of the AI stack has ~20 strong competitors.
We discuss: 🔸 Why you don’t need to be chronically online to keep up 🔸 The "/loop make me happier" opportunity in AI 🔸 Why every company is becoming a series of loops 🔸 Why moats are discovered, not designed 🔸 What's the Birkin bag version of your product?
Listen now 👇 https://t.co/zSKu8pfsoQ
See 2 related tweets
- @lennysan: .@illscience: "Company building will now be creating a series of loops" https://t.co/ZpouX2hvwg\n\nQ...
- @Scobleizer: RT @lennysan: The "permanent underclass" is a dark fantasy Silicon Valley keeps telling itself.
@i...
9. ANI (Group Score: 97.0 | Individual: 25.9)
Cluster: 6 tweets | Engagement: 5 (Avg: 120) | Type: Tech
#WATCH | On Delhi Startup & Incubation Policy 2026, Delhi Minister Ashish Sood says, "We are working to implement this policy in collaboration with prominent startup founders from across the country who started and scaled their businesses in Delhi. Through this policy, the Delhi government aims to foster and support over 10,000 startups and create employment opportunities for more than 4 lakh youths... We want to build a startup ecosystem that benefits both founders and job creators. We are working on this under the guidance of PM Modi and the leadership of Rekha Gupta, and we held discussions regarding this today..."
See 5 related tweets
- @ANI: #WATCH | Delhi Chief Minister Rekha Gupta says, "It's a matter of great joy that the promise we made...
- @ANI: #WATCH | Delhi Minister Ashish Sood says, "... Our government has launched this Startup and Incubati...
- @ANI: #WATCH | Delhi Minister Ashish Sood says, "Under the Delhi Start-Up and Incubation Policy 2026, the ...
- @ANI: #WATCH | Delhi: Chief Minister Rekha Gupta holds a meeting with investors and startup founders on th...
- @ANI: RT @ani_digital: "Delhi lagged in developing startup ecosystem; it should have": Minister Ashish Soo...
10. jjohana (Group Score: 92.3 | Individual: 44.9)
Cluster: 5 tweets | Engagement: 68328 (Avg: 765) | Type: Tech
RT @OpenAI: This is GPT-6 Astra.
Anything you can do on a computer, Astra can do for you. Fast. https://t.co/gDd0IsewJw
See 4 related tweets
- @heyshrutimishra: someone made a response video to the Astra demo https://t.co/0OiaY3EuXO\n\nQT @OpenAI: This is GPT-6...
- @michalmalewicz: GPT-6 Astra just ended software.
Go read.
Also, Astra made the thumbnail about Astra.
https://t....
- @Rasmic: RT @Rasmic: My GPT 6 Astra review is finally here... it's me using it for real work... no game demos...
- @_simonsmith: Honestly hard to tell right now, so: How are you using GPT-6 Astra with Blender?...
11. pstAsiatech (Group Score: 91.4 | Individual: 25.2)
Cluster: 4 tweets | Engagement: 3 (Avg: 8) | Type: Tech
#PostAGIWorldWatch\n\nQT @leo_linsky: We evaluated GPT-6-Astra in 100 complex, unsaturated multi-agent coding environments, competing and cooperating with other models in open-ended tasks.
It's the new frontier model by a landslide. It's even more dominant than the Fable 5 release, because not only does it wipe the floor with the second best model (Fable 5.1), it was also 80% cheaper and 30% faster in agentic coding.
This is a groundbreaking model. The biggest breakthrough since Opus 4.5, maybe even since GPT 4.
This is also a testament to how unsaturated our evals are. Some of the open-ended environments we run were introduced almost a year ago and can still compete frontier model submissions against older, deprecated models like Sonnet 4 without signs of saturation. Our newer ones are much more challenging and completely auto-generated.
We are clearly living in a post-AGI world. Do something interesting and useful with it.
See 3 related tweets
- @haider1: GPT-6 Astra really looks like openai went hard on pre-training
world modeling, math/abstraction, sp...
- @charliermarsh: RT @leo_linsky: We evaluated GPT-6-Astra in 100 complex, unsaturated multi-agent coding environments...
- @teortaxesTex: Here you see the divergence between Anthropic-shaped models and OpenAI-shaped or independent models....
12. derrickcchoi (Group Score: 85.6 | Individual: 32.5)
Cluster: 3 tweets | Engagement: 36 (Avg: 199) | Type: Tech
Time to change the narrative that GPT models are not as good at design\n\nQT @arena: Real-world results are in. There is a new #1 on Code Arena - GPT-6 Astra (Max)!
It also reshapes the Pareto frontier as the best-performing model at $40/Mtoken, which matches the latest Claude model pricing.
GPT-6 Astra by @OpenAI takes the top spot in Code Arena: WebDev with a score of 1797 pts. This opens up a solid +35pt lead over #2 Claude Fable 5.1 (Max) at 1762 pts and #3 Claude Opus 5 (Max) at 1688 pts.
This is a significant improvement from GPT-5.6 Sol (xHigh) at +180 pts, ranked at #13.
Category level votes still incoming, but already we see it at #1 in: Data & Analytics, Consumer Product, Content Creation Tools and #2 in Gaming and Simulations.
Stay tuned for other categories like Brand & Marketing, Reference-Based Design and Full Stack rankings.
Congrats to the @OpenAI team on this release!
See 2 related tweets
- @ml_angelopoulos: Man! The pace at which labs are hill-climbing on coding is extraordinary.
In the last week, there h...
- @rohanpaul_ai: RT @rohanpaul_ai: GPT-6 Astra (Max) leads Code Arena: WebDev at 1,797, 35 points above Claude Fable ...
13. RnaudBertrand (Group Score: 83.4 | Individual: 24.2)
Cluster: 4 tweets | Engagement: 4898 (Avg: 1350) | Type: Tech
RT @nayibbukele: “Results in reading, mathematics and science were above the national average and comparable to the average results in Germany and Sweden.”
These are the results of our pilot program, which includes AI tutors and has been running in 171 public schools for just over a year.
It was evaluated by PISA for Schools in June of this year. Since then, it has already expanded to more than 1,000 schools, and within 18 months, it will be in every public school in our country.
In a few years, El Salvador will inspire the world.
Security was just the beginning 🇸🇻
See 3 related tweets
- @WesRoth: meanwhile New York bans AI in classrooms
predictions on how that will affect those students?\n\nQT ...
- @elonmusk: Grok\n\nQT @Polymarket: JUST IN: El Salvador’s AI tutoring pilot in 171 public schools produced read...
- @BrianRoemmele: Very proud of what we have achieved for AI and education in El Salvador.
This is a testament to cle...
14. Nas_tech_AI (Group Score: 79.0 | Individual: 45.6)
Cluster: 2 tweets | Engagement: 914 (Avg: 193) | Type: Tech
RT @Apodex_AI: Meet Apodex 1.1: Scaling Agentic Intelligence for Complex Work
Open Source Harness: https://t.co/4V9suLx8o5 Open Weights: https://t.co/9CNZ9E5drH
We’re excited to introduce Apodex 1.1, our new model family built to scale agentic intelligence for professional work.
🧠 Frontier-level intelligence for complex work Apodex 1.1 brings frontier-level agentic performance across complex professional work, scientific research, financial analysis, and deep search.
🤝 Asynchronous Agent Team Apodex 1.1 can break down complex tasks, coordinate multiple agents in parallel, continuously integrate their findings, and let you step in to guide or redirect the work at any time.
🔬 Open-source research workbench We’re open-sourcing FrontierAgent—a locally deployable research workbench for the Apodex 1.1 family, including asynchronous Agent Team.
Available now: 🔹 Apodex 1.1 — our most capable frontier model, available through the Apodex online workbench 🔹 Apodex 1.1 mini — open-weight model for running complex work locally 🔹 FrontierAgent — open-source, locally deployable research workbench
🌐 Live workbench: https://t.co/w5l6YI2ZuP 🦾 API platform: https://t.co/kaxuSWvMB4 📃 Paper: https://t.co/CHc4L63U07
See 1 related tweets
- @Nas_tech_AI: Apodex just officially went open-source, and they gave it a massive upgrade in the process.
It’s on...
15. phl43 (Group Score: 77.8 | Individual: 31.0)
Cluster: 3 tweets | Engagement: 29 (Avg: 220) | Type: Tech
Several people are bringing up Kurzweil's argument, but that's a terrible argument and a perfect illustration of what I was talking about!
There are lots of things reasonable people can disagree about, but that Kurzweil's argument was fallacious is not one of them, at least not if they have thought it through.
This family of argument, which infer a likely date for AGI from trends about the growth of compute, is exactly the kind of sloppy thinking I was complaining about.
The fact that so many people bring it up as a counterexample to my claim is driving me so nuts that now I will have to write a post to explain why it's such a bad argument.\n\nQT @phl43: Even if AGI had already come, which I don't think it has yet, the fact that someone predicted in 1999 with any non-negligible degree of confidence that AGI would come around now is a reason not to take them seriously in my opinion.
That's because the various pieces of the puzzle that now make it plausible to think AGI is not very far off, although once again we still don't know exactly when it will come, didn't fall into place until much later, so anyone making such a specific prediction back in 1999 couldn't have had a good argument for it.
This illustrates one of the many reasons why I think predictions are overrated as an epistemic benchmark. It's easy to be right for the wrong reasons and wrong for the right reasons, so you should pay more attention to the quality of the arguments that people use and doomers make terrible arguments all the time.
Another issue is that, though in some cases you can show they have a pretty good track-record on verifiable predictions in recent years, those predictions are about relatively narrow questions and I don't think we have good reasons to think they translate into reliable predictive ability on the broader questions that most people really care about, but people are trying to sell their track-record on the former as a reason to defer to them on the latter.
Again, in light of the sloppy arguments they make all the time, I don't think the fact that e. g. they were able to predict accurately e. g. how SOTA models would score on SWE-bench is a good reason to update a lot based on their predictions about the economic consequences of AI or even x-risk.
See 2 related tweets
- @teortaxesTex: This kind of Clear-Headed Thinking is why the French are inept compared to Anglos in producing parad...
- @phl43: RT @phl43: Even if AGI had already come, which I don't think it has yet, the fact that someone predi...
16. firesidealpha (Group Score: 76.4 | Individual: 27.0)
Cluster: 3 tweets | Engagement: 22 (Avg: 42) | Type: Tech
Dave Lee reveals Elon shared that the Cybercab drive unit uses no rare earth elements and is fully automated to assemble
"I talked to another engineer who was in charge of, or talking about, the drive unit. And the drive unit is basically, you have your motor and your transmission and the stuff that controls all that stuff."
"Elon shared that they somehow managed to have no rare earth elements in that engine, in that, not engine, in that drive unit, but yet still achieve a great mileage or efficiency."
"Now this drive unit was version five, and this drive unit, I guess I was reading, is like 100% automated. Like, there's no humans really involved in the assembly process."
"Now with this drive unit version five for the Cybercab, they made it more efficient, smaller, cost less, etcetera, and it's their next generation of version five."
"And they also made it where it can be super reliable, where it can handle like a million miles for a Cybercab."\n\nQT @davelee: I attended Tesla’s Cybercab launch in Austin and spent 2.5+ hours questioning engineers about unboxed manufacturing, the plastic exterior body panels, brake-by-wire, the battery pack, AI5 and more. My takeaway: Cybercab may be Tesla’s most important product since the Model 3.
I fit as much info as I could in this video. Hope it's helpful!
See 2 related tweets
- @SawyerMerritt: Definitely worth a watch\n\nQT @davelee: I attended Tesla’s Cybercab launch in Austin and spent 2.5+...
- @Kyrannio: RT @davelee: Time for Tesla to take everything they learned from making the Cybercab and launch a ne...
17. RoundtableSpace (Group Score: 73.8 | Individual: 26.8)
Cluster: 3 tweets | Engagement: 107 (Avg: 86) | Type: Tech
GPT-6 Astra built a 3D website that separates the male anatomy into 2,234 individually modeled pieces, turning interactive education into something that would have taken a team months to build manually.
See 2 related tweets
- @coinbureau: 🔥SCIENCE: OpenAI’s GPT-6 Astra can dissect the human body for you.
A user built an interactive 3D a...
- @DataChaz: RT @DataChaz: HOLY SH*T THIS IS INSANE
This girl used GPT-6 Astra to build a 3D website that pulls ...
18. _NathanCalvin (Group Score: 73.6 | Individual: 35.6)
Cluster: 3 tweets | Engagement: 176 (Avg: 73) | Type: Tech
RT @AndrewCurran_: Jakub Pachocki, Chief Scientist OpenAI:
'Currently I believe that no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer. I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established. And I believe that international coordination on future AI development needs to become a top priority for governments around the world.'
'Machine intelligence playing a larger and larger role in its own development process is a natural conclusion of sustained technological progress. If AI progress continues, machine recursive self-improvement (RSI) will be at the very core of future scientific discovery.
Automated AI research is a more dramatic form of scaling intelligence with compute; and of course as a part of it, AI will improve the computational substrate itself. And similarly to scaling, we focus OpenAI research towards RSI as we believe it is the only way to remain at the frontier of AI research moving forward.
I want to stress that the above words don’t imply I think greatly accelerating deep learning research, especially in the short term, is the right collective action we should take as the research community. However, I do think this is where the current path leads, and we all need to make a conscious choice on how to proceed.'
See 2 related tweets
- @scaling01: Jakub Pachocki, Chief Scientist at OpenAI:
"I believe that international coordination on future AI ...
- @willdepue: RT @kliu128: "Currently I believe that no lab has solved alignment and monitoring to a sufficient de...
19. MTSlive (Group Score: 72.0 | Individual: 44.3)
Cluster: 3 tweets | Engagement: 1362 (Avg: 151) | Type: Tech
SITUATION DETECTED: OpenAI says their 90th-percentile researchers now use over $7,000 worth of tokens per day.
See 2 related tweets
- @Techmeme: OpenAI says it hit its "automated research intern" goal, its researchers now use 3.1 agent-workdays ...
- @BrianMRey: good lord
i can’t even begin to imagine the size of their knowledge vault\n\nQT @MTSlive: SITUATION...
20. 0xCodez (Group Score: 71.5 | Individual: 27.1)
Cluster: 3 tweets | Engagement: 216 (Avg: 307) | Type: Tech
OpenAI senior engineer just released a 1-hour course on building agentic systems with Codex from scratch:
• 00:36 - deep dive into Codex: agents, subagents • 11:34 - Codex agentic prompting patterns • 20:06 - long running Codex agents with /goal • 30:03 - building skills for Codex agents • 41:28 - self-improving loops with /automations
this 1-hour workshop by the OpenAI team will replace a $500 AI engineering course
watch today, then read how to build self-improving agentic systems with loops & graphs in the article below\n\nQT @0xCodez: https://t.co/eJB3TgyJwV
See 2 related tweets
- @ai_explorer25: Anthropic senior engineer just released a 1-hour course on building a team of agents with loops & gr...
- @0xCodez: RT @0xCodez: OpenAI senior engineer just released a 1-hour course on building agentic systems with C...