China Dismisses AI Slowdown Calls, Blasts U.S. “Fearmongering”
Beijing has pushed back against growing calls from U.S. tech leaders to slow the development of AI, accusing them of using fear to contain China’s progress in the technology.
- The remarks follow an essay by Anthropic’s CEO calling for a slowdown in AI development, in which he also said that a “Chinese lead in AI would pose grave danger for the United States and the world.”
- China’s foreign ministry urged all parties to “promote an open, inclusive and benevolent approach to AI,” and said, “Fearmongering, confrontation and malicious competition will only disrupt the process of global AI governance and serve no one’s interests.”
- The state-run Global Times newspaper struck a more blunt tone, blasting the call for an industrywide slowdown as “packed with containment provisions targeting China and is, in essence, a ‘Cold War playbook’ for the AI sector.”
- The U.S. and China are planning to discuss frontier AI safety risks in mid-September, and the issue could be raised during a summit between Presidents Donald Trump and Xi Jinping on September 24.
Meanwhile, President Trump also weighed in, taking to Truth Social to dismiss concerns over extreme AI risks. He described critics as “very negative forces” raising scenarios that will not happen, called efforts to limit AI part of a “SICK conspiracy,” and said he wants to ensure the United States remains the industry leader.
Tensions Between OpenAI and Mathematicians Continue to Rise
OpenAI’s push to solve major mathematical problems is creating growing tensions with researchers who say AI labs are moving too quickly and putting attribution, collaboration, and open research at risk.
- The rift started after NYU professor Tristan Buckmaster accused OpenAI of pressuring him over credit for work involving an Anthropic researcher, and questioned whether OpenAI used related Codex work in its own Navier Stokes proof.
- 25 Fields Medal winners have now signed an open letter warning that AI labs are threatening mathematical research as they compete to solve famous problems.
- On Thursday, OpenAI withdrew its sponsorship of a math event at Caltech following criticism from researchers at the university.
- The mathematicians argue that proofs generated by AI need to be properly verified, explained, documented, and connected to previous work before they can become part of the mathematical canon.
- Mathematicians now fear that using Codex could feed their work straight back into OpenAI’s models, letting frontier labs spend millions racing to beat them to a proof, a dynamic that will incentivize secrecy over open research.
They say the value in math isn’t just the proofs and who gets credit, but the intellectual superstructure that nourishes students, finds new questions and ideas, and integrates them into broader human civilization. The broader issue extends beyond mathematics, as they see AI as a general threat to intellectual work.
How to create your own 1980s AI photo using ChatGPT
Step 1: Open ChatGPT and sign in.
Step 2: Upload your photo.
Step 3: Paste this prompt:
Transform my uploaded photograph into a realistic 1980s portrait. Keep my facial identity, facial structure, natural skin tone and recognisable features intact. Give me an authentic 1980s hairstyle, period-appropriate clothing and accessories. Use warm studio lighting, subtle analogue film grain, slightly faded colours and soft photographic imperfections. Make it look like a real photograph taken on a film camera in the 1980s rather than a modern photograph with a vintage filter.
You can tweak the prompt further if you’re looking for a specific look.
Step 4: Once the image is generated, download, share, or further edit it.
You can change the hairstyle or outfit, add jewellery, replace the background, adjust the lighting, etc.
ChatGPT Sites adds faster and more collaborative website building
ChatGPT Sites now lets multiple people edit, save and publish the same Site through a new Build Together feature, while private sharing allows access to specific collaborators. OpenAI has also roughly halved deployment times and added database inspection and custom domain support.
Sam Altman rules out an OpenAI IPO in 2026
OpenAI CEO Sam Altman has confirmed that the company will not go public this year, saying an IPO would be “ill advised”. He said OpenAI wants to focus on addressing safety challenges and working with governments and other AI companies before pursuing a public listing.
OpenAI brings GPT‑Live‑1 to its API
OpenAI is launching GPT‑Live‑1 in the API, giving developers a powerful, natural voice model for building voice-enabled apps and business workflows.
Jensen Huang says Nvidia will grow 70%. He’s not wrong.
Jensen Huang is a founder, CEO, and (apparently) psychic.
Last week, Huang took the stage at Goldman Sachs Communacopia + Technology Conference to make a bold claim. Nvidia’s revenue would grow 70% next year, bringing the company’s top line to $680 billion.
Now, it’s common practice to gas up your brand in the land of Silicon Valley. But that’s not what’s happening here. Huang gave those same numbers to his investors last month. And he’s got the receipts to back it up.
Why Huang’s betting big:
- 70%: The projected annual revenue growth, which would push Nvidia to ~$680 billion.
- $1 to $100: The return ratio Huang claims on Nvidia’s investments.
- $400 billion: Nvidia’s expected revenue this fiscal year.
The shovel seller’s advantage
All the major AI labs (Anthropic, OpenAI, Google) currently rely on Nvidia, even as some race to build their own chips. So Huang’s claims of industry domination aren’t that far off.
Plus Neoclouds, OEMs, hyperscalers, and AI-native startups all funnel data back to the company. This gives Nvidia a real-time map of where AI money and infrastructure are actually flowing.
So yes, critics call his investment deals circular. But Huang shrugs them off.
Haters will always hate the guy selling shovels in the AI gold rush.
OpenAI shares tips to get better results from GPT-4o agents
Your old Codex prompts might be hurting you now. OpenAI just dropped guidance saying that instructions written for older models can actually slow GPT-6 Astra down by wasting its memory and triggering the wrong behaviors.
Here’s the core idea: Astra is smarter, so it needs less hand-holding. All that bloated guidance you wrote to babysit weaker models? It’s now dead weight.
Three things to fix right now:
- Skills are prompt files that load for specific tasks. Make their trigger descriptions narrow and precise, so Astra grabs the right one instead of misfiring.
- AGENTS.md loads on every single task. Strip out mandatory pre-reading and generic “run your tests” reminders. Astra already does that on its own.
- Task prompts should clearly define what “done” looks like, so Astra knows when to stop instead of pausing to ask you.
The move: ask Astra itself to audit your existing skills and AGENTS.md files. Less scaffolding, cleaner results.
Solo dev open-sources free ElevenLabs rival that clones voices and dubs video into 646 languages
ElevenLabs charges per character, caps your usage, and sends your audio to their servers. VoiceStudio flips all of that. It’s a free, open-source tool that runs 100% on your own machine.
A solo builder shipped this, and it already has 25K GitHub stars. Here’s what it actually does:
- Clone any voice from one clean audio clip
- Dub videos into 646 languages (ElevenLabs does 32)
- Generate audiobooks, run transcription, and dictation
- Pick from 14 different text-to-speech engines instead of being locked to one
No billing. No caps. Your audio never touches a server.
To get started: git clone the repo, run bun install, then bun run desktop. It auto-configures all Python dependencies on first launch.
The honest tradeoff: quality depends on your hardware and which engine you pick. But for language coverage and privacy, nothing else comes close right now.
10 free open-source GitHub tools that replace software you’re paying for monthly
GitHub is quietly full of free tools that replace software you are probably paying for monthly. Here are 10 open-source projects worth knowing about.
The two standouts worth digging into first:
- LibreChat lets you run ChatGPT, Claude, Gemini, Mistral, and local models all inside one self-hosted interface. You own your data, control which models load, and pay only for the API calls you make. No more juggling five browser tabs or five subscriptions.
- TradingAgents simulates a real trading firm using AI. Separate AI agents act as a fundamentals analyst, sentiment expert, technical analyst, and risk manager. They discuss, debate, and land on a strategy together. Built for research, not live trading.
The rest of the list covers an open-source Bloomberg terminal clone, a one-click short video generator, an AI email assistant, a voice cloning tool, an API integration platform, a video generation engine, and a Claude Code skills library.
The takeaway: before you renew that SaaS subscription, check GitHub first.
Amodei says pace the frontier. Every other frontier CEO said yes within a day.
Amodei published “We Must Pace the Frontier” on Saturday. The argument starts with a claim about the last three months: “since roughly this summer, AI has been advancing drastically faster, driven primarily by AI’s growing ability to build the next generation of AI.” His main fear is agent swarms. He writes that “in 6 to 12 months such a swarm could be capable of taking over the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage).”
Dario’s plan has three steps, and only the first is something Anthropic can do alone. Embedded third-party evaluators get, in his words, “Desks in our offices, access badges, and company laptops,” plus permissions “mostly comparable to what internal risk assessment teams have,” and the right to publish findings “without editorial control by Anthropic.” Step two is common standards and capability limits across democratic-world labs, which he says needs Washington to “issue a narrow waiver for certain kinds of safety conversations” because of antitrust. Step three is coordination with authoritarian governments, which he concedes is the hard one. On China he asks for no new limits on US labs at all: keep chip export controls, crack down on distillation, harden weight security.
Sam Altman posted that he agrees “we need to pace the frontier,” called it “a primary topic of discussions we’ve had at OpenAI in recent weeks,” and said OpenAI will match the independent-evaluator commitment. Musk replied “Dario is right.” Nadella backed pacing on Sunday and said Microsoft would open its first-party MAI model behavior rules to public consultation. That is four of the largest labs on the same side of an argument inside two days. Personally, I think it’s a bullshit face-saving maneuver and none of these companies intend on slowing development.
The market read “pace the frontier” as a capex cut
Memory took it worst. SK Hynix closed down 6.4% and Kioxia fell 6.4%, Samsung Electronics lost 4.1%, and the KOSPI dropped roughly 3.2% in early Monday trade. Europe followed at the open: as of 07:10 GMT, ASML was off 4.4%, BE Semiconductor 4.8%, Infineon and ASM International both more than 5%, STMicroelectronics 3.5%. SoftBank fell as much as 13%, its worst session since July. US names barely moved by comparison, with Nvidia and Micron roughly flat and Arm up.
Bernstein’s Stacy Rasgon argues the reaction overshoots the text. Semis were already down about 19% from June peaks, and as his team put it, Amodei “is not calling for a halt to training, but rather a shift from ‘extremely fast’ to ‘only somewhat fast.'” The essay asks for no compute cap on US labs. It asks for evaluators and export controls. Bernstein still likes Nvidia, Broadcom and the equipment makers.
The timing is the part worth sitting with. Amodei publishes a slow-down manifesto Saturday. The Financial Times reports Anthropic has told shareholders it expects a second straight quarter of positive adjusted operating income on gross margins above 80%, against $11.5B in Q2 revenue. Business Insider reports Anthropic has picked Nasdaq for an October listing, with Reuters separately reporting talks for Nvidia to anchor with up to $10B. On Monday, Bloomberg reported SoftBank closed an upsized $11.87B two-year loan from roughly 20 banks to fund its OpenAI investment. Altman, meanwhile, told Fortune on Friday that safety makes right now “an ill-advised moment to go public,” and confirmed: “I would say not 2026, yeah.” Every one of the Anthropic financial figures is single-outlet reporting from unnamed sources with no company confirmation, so hold them loosely. The direction of the calendar is not in dispute.
A benchmark built on private production code puts the best model at 38.8%
While everyone argued about how fast to go, a small shop called Specific published Real-SWE, which swaps public GitHub issues for licensed private codebases: a social events app with 200,000-plus users, a fintech platform processing 100,000-plus bank statements, an enterprise AI sales product. Tasks touch roughly 11 files each, against about 6 on comparable public benchmarks.
Nobody passes. Fable 5.1 resolves 38.8%, GPT-6 Astra 33.8%, Gemini 3.8 Flash 31.2%, GLM 5.3 28.8%, Grok 4.6 and Muse Spark 1.3 both 23.8%, Kimi K3 18.8%, GPT-5.6 Sol 16.2%. That is a long way from the 90-plus Terminal-Bench numbers these same models post.
Caveat it properly: 10 tasks, 8 runs each, 640 scored rollouts total, and the harness and dataset are not open, only sample tasks on request. That is a small sample from an interested party, and it is not a replacement for a public benchmark. It is still the cleanest available argument that the gap between benchmark scores and production engineering is enormous, and it landed on the weekend the industry decided capability growth is the emergency.
AI leaders are actually asking to slow AI down
Anthropic CEO Dario Amodei wrote a 3,800-word essay arguing that AI capabilities are moving too fast for safety work to keep up. Sam Altman, Elon Musk and DeepMind’s Demis Hassabis basically agree.
Amodei wants frontier labs to slow down, let outsiders test their models and eventually get countries to agree on some shared safety rules. His big worry is recursive self-improvement: AI getting good enough at AI research that it starts making better AI itself. He thinks we could be 6-12 months away from swarms of agents doing basically everything humans do on the internet.
The problem is nobody wants to be the first one to slow down. Trump rejected calls for an AI slowdown. His argument is pretty simple: “whoever wins AI wins.”
And China said no too. The state-run Global Times accused the US of trying to limit China’s AI progress through “technological barriers and regulatory monopolies,” calling the proposal a “silent AI Cold War.”
Funny enough, AI 2027 basically predicted this exact situation. In its scenario, experts start freaking out after the leading US lab develops a superhuman AI coder/researcher, but the president refuses to slow down because China might take the lead. We’re obviously not at the superhuman AI researcher bit yet… but the political argument is already here.
There are a couple other explanations for why the labs suddenly want to slow down. Maybe they’re worried about open-weight models catching up. Or maybe they’re all hitting the same walls and seeing something similar internally.
I’m not convinced they’re secretly staring at ASI though. If they were genuinely sitting on intelligence that could run circles around humanity, you’d expect to see some pretty ridiculous breakthroughs outside of chatbots too.
OpenAI isn’t going public this year
Despite confidentially filing for an IPO, Sam Altman says 2026 would be an “ill-advised” time to actually do it. OpenAI had reportedly been aiming for Q3/Q4 this year, but Altman now says “we’ve got a lot of stuff to do.”
Part of it is AI safety. Altman says OpenAI doesn’t want the pressure of being public while everything around the technology is moving this quickly, and that it’ll IPO when the business and “society” are ready.
There’s also the less philosophical bit: tech stocks have been volatile and OpenAI has its own financial challenges. The company is reportedly leaning toward 2027 instead.
So the OpenAI IPO is still coming… just probably not this year.