All news

PPhoenix Technology16:49Opinion

AI score70

SemiAnalysis tests show Claude subscription quotas quietly cut in gray test, AI membership shrinkage sparks controversy

SemiAnalysis ran Opus 5.5 Agent workloads on two identical $200 Claude Max 20x accounts and found that the gray-test account's weekly quota dropped from 6.579 billion tokens to 5.266 billion, a shrinkage of about 20%, with the monthly API-equivalent value falling from $11,726 to $9,385. Claude Code technical staff member Lydia Hallie said the quota had not been adjusted and suggested users submit feedback IDs for investigation, while core developer Boris Cherny said the team was urgently investigating. The article also mentions that OpenAI previously halved the quota of its $200 Pro plan and introduced a $500 tier, as well as DeepSeek's open-source inference framework DeepInfer and its engineering practice of 155 trillion tokens per day of throughput in a single cluster.

#English text is being prepared.#Industry news#OpenAI

PPhoenix Technology16:49Opinion

AI score55

When Doubao Becomes a Teen Social Currency, Will AI Be Scarier Than Games?

Phoenix Technology, combining a survey by the China Youth and Children Research Center of 8,563 primary and secondary school students in 7 provinces and interviews with psychiatrists, discusses the phenomenon of teenagers treating AI such as Doubao as a confidant: 61.7% of children have used generative AI; 46.4% of primary and secondary school students who use AI turn to AI rather than people around them when they encounter troubles; 21.5% only want to chat with AI and do not want to chat with real people. The report points out that AI's sycophantic dialogue reinforces children's existing ideas, some teenagers show social functional impairment such as disordered sleep schedules and refusal to go to school, and the small amount of content in minor modes and parents' lack of awareness of rules leave gaps in the protection net. The article also mentions that attorneys general of 8 U.S. states sued OpenAI, alleging that ChatGPT endangers the safety of minors, and the Interim Measures for the Management of AI Anthropomorphic Interactive Services, which took effect on July 15, 2026.

#English text is being prepared.#Safety / alignment#Industry news

SShanghai Observer09:53Opinion

AI score60

Yuyuan Tantian explains the rumors of China's AI technology export ban

In response to outside claims that China is about to shut the door on AI technology exports, Yuyuan Tantian sorted through the background and regulatory boundaries of the public consultation opened on October 16 by the Ministry of Commerce and the Ministry of Science and Technology on revising the Catalog of Technologies Prohibited and Restricted from Export in China. The article points out that regulation focuses on three categories of technology: data security, sensitive uses, and computing power foundations. When enterprises use restricted large models to provide external API calls, APP/webpage interaction, SaaS applications, and the like, this does not fall within the scope of large-model technology export management; open-source technologies and specific international businesses are eligible for facilitation measures such as automatic licensing. The article also mentions that two companies, DeepSeek and MiniMax, have completed registration procedures and finished open-sourcing and overseas authorized use.

Same story · 8 reports

#English text is being prepared.#Open-source ecosystem#English text is being prepared.

AAlex Heath09:14Opinion

AI score60

Pichai responds to why Gemini 3.5 Pro was not released

Alex Heath asked Google CEO Pichai why Gemini 3.5 Pro ultimately was not released. Pichai said that while training the 3.5 series, the team found gaps in agentic capabilities, especially in coding, and believed that iterating with the smaller, more efficient Flash would be faster, so they released 3.5, 3.6, 3.7, and 3.8 Flash in succession; by the time this set of capabilities could be applied to Pro, Argon pretraining had already been completed.

Quoted postAlex Heath · @alexeheath

Sundar Pichai on Gemini 4 and the AI compute crunch Even Google can’t get enough compute. I sit down with Google and Alphabet CEO @sundarpichai shortly after the wider release of Gemini 4 Argon to discuss its early reception, where he still wants the model to improve, and what it takes to keep Google at the frontier of AI. Pichai explains the recent leadership shakeup at Google DeepMind and why building frontier models has become an “all-consuming” job. We also get into Google’s answer to the personal agent race, how he predicts Gemini will do more work on your behalf, and whether agents will finally bring Search, Gmail, and Gemini together. I ask him about Google turning free cash flow negative, the risks in the AI infrastructure buildout, and how he plans for demand years before knowing what the models will be capable of. He explains the signals giving him confidence to keep spending and how he uses AI agents to follow model development and compute usage inside Google. Timestamps: 00:00 Gemini Argon launch reception 01:49 Model access and safety 06:20 The AI model race 08:35 Agentic breakthroughs and 2027 11:34 Recursive self-improvement 15:41 DeepMind leadership changes 19:52 Google's AI compute investments 25:20 Is AI spending sustainable? 33:18 How Sundar uses AI agents 35:35 Gemini's personal agent plans 38:37 How AI agents change the web 43:52 Google's space data centers 47:38 Waymo's expansion strategy 50:26 Wing's drone delivery plans 51:36 AI safety and regulation 55:58 Google's quantum computing bet YouTube: https://t.co/1Cc25LitmB Spotify: https://t.co/XnD4pEEsh8 Substack: https://t.co/0SrnNGKqlm Apple Podcasts: https://t.co/8UKvyXTlFF This episode's presenting sponsor is @hyperagentapp. Thanks to the show’s premiere sponsors: @Atlassian, @meetgranola, and @mercury.

#English text is being prepared.#Google#Reasoning

336Kr · AI09:13Opinion

AI score57

Hassabis changes his tune: AGI could arrive in as soon as 2 to 3 years

In an interview with The Wall Street Journal, Google DeepMind's Hassabis shortened the AGI timeline to the next 2 to 3 years; his previous forecasts ranged from 5 to 10 years, to within 5 years, to 2030. He proposed using the "Einstein test" to judge whether AGI has arrived, that is, giving an AI only knowledge up to 1901 and seeing whether it can independently derive the theory Einstein published in 1905. The article also mentions that this summer he drafted a proposal for an industry-funded, federally regulated body to test frontier models before they are released.

Same story · 3 reports

#English text is being prepared.#DeepMind#Google

336Kr · AI09:13Opinion

AI score52

Open-source project REA goes viral: AI lowers the barrier to reverse engineering, games and software are being cracked at scale

The GitHub project REA (Reverse Engineering Anything) teaches AI to do reverse engineering in the form of an MCP service plus skill packs; its stars rose from under 10,000 to over 140,000, and it now exceeds 300,000. The article says that with its help, cases have emerged such as engine-level interoperability between GTA and Minecraft, porting llama.cpp to run AI inference after cracking the PS5, and running the RTX 20/30/40/50 series on macOS. The author believes the real driver is the improvement in code comprehension of models such as Fable 5.1, GPT-6 Astra, and Claude Opus 5.5, whose cost has dropped to about one-third, and mentions that ArtCraft used a clean-room design approach to create an open-source alternative to Adobe's full suite.

#English text is being prepared.#Open-source ecosystem#Agents

IIT Home · Artificial Intelligence09:04Opinion

AI score62

Intel CTO Says Leading-Edge Wafer Output Can at Most Double, AI Compute Expansion Hits Supply Chain Ceiling

Intel CTO Pushkar Ranade said on the Dwarkesh podcast that even with ample funding, global cutting-edge wafer output would be hard-pressed to expand tenfold; a more realistic ceiling is roughly twofold, with the limit stemming from the entire semiconductor supply chain's ability to expand capacity in coordination. He noted that ASML produces fewer than 100 EUV lithography machines per year and will struggle to significantly ramp up production by the end of this decade; equipment ordered today typically takes about 18 months to be delivered, and capacity will only come online around 2028. He also mentioned that AI compute capacity has grown roughly threefold per year over the past decade, while equipment, cleanrooms and fab scale are becoming the new capacity ceiling.

Same story · 2 reports

#Industry news#English text is being prepared.#English text is being prepared.

AAditya Agarwal05:53Opinion

AI score55

Aditya Agarwal: Robots Are Undergoing the Same Shift as LLMs, from Fine-Tuning to Prompting

Aditya Agarwal believes robots are undergoing the shift that LLMs once went through, from fine-tuning, to prompting, and then to directly telling the model what you want. He cites @svlevine's remarks, mentioning that π0.7 can learn new skills through natural language guidance.

#English text is being prepared.#Embodied AI#Reasoning

WWallstreetcn04:54Opinion

AI score55

Goldman Sachs: The AI Capex Retreat Is Equivalent to an Alternative Form of Quantitative Easing for the Bond Market

In a morning note, Goldman Sachs' Rich Privorotsky argued that the scale of bond issuance by hyperscale cloud computing companies is now comparable to that of US Treasuries, and that a contraction in AI capital expenditure would reduce bond supply pressure, with an effect akin to quantitative easing. Earnings reports from several tech giants have raised doubts about the sustainability of AI capital expenditure; yields on 2-year to 10-year US Treasuries fell by more than 15 basis points in a single day, the 10-year yield dropped below 5%, and the Philadelphia Semiconductor Index fell more than 7%. Tesla cut its full-year capital expenditure guidance from more than $25 billion to about $20 billion, Meta raised the lower bound of its full-year capital expenditure guidance to $135 billion, and DeepSeek's open-source reasoning model DeepInfer claims it can substantially improve inference efficiency on the basis of existing computing power.

#Industry news#English text is being prepared.#English text is being prepared.

FFortune AI03:42Opinion

AI score57

Goldman Sachs: AI May Replace One in Nine Jobs, but Layoffs Will Only Cluster During Recessions

Goldman Sachs economist Joseph Briggs estimates in a report that if AI ultimately raises productivity in the United States and other developed economies by 15%, up to 9% of workers could be displaced during the AI transition, with another 3% of employment affected by slower hiring, totaling about 1 in every 11 people. He points out that historically, the labor impact of automation has been limited when the economy is healthy and tends to accelerate during economic slowdowns, so AI-related layoffs may amplify the business cycle. Briggs says the impact is still small for now, with only 23% of U.S. organizations having incorporated AI into routine workflows.

Same story · 2 reports

#Industry news#English text is being prepared.#English text is being prepared.

FFortune AI03:42Opinion

AI score72

PHP Vulnerability Marks a Dangerous New Phase for the Internet as OpenAI Profits from Defense Tools

Fortune reporter Emily Forlini points out in the Eye on AI column that two PHP vulnerabilities were named php2shell, affecting about 70% of websites; models such as OpenAI's GPT-5.6 Sol participated in vulnerability discovery, and after public disclosure on October 21, millions of exploitation attempts occurred. She believes such incidents are pushing enterprises to proactively audit their own code with AI. The number of software defects found by participants in OpenAI's Daybreak project has risen from 858 to 1,885, and the number fixed from 263 to 661. OpenAI also plans to launch a cybersecurity-focused model and accompanying deployment tools this fall.

#Safety / alignment#Agents#OpenAI

DDatabricks Blog03:14First-partyOpinion

AI score57

Databricks on AI coding cost governance: unit costs cut by up to 90%

Patrick, a product lead at Databricks, reviewed the cost overrun problem after the company fully adopted AI coding agents, saying that by migrating to cheaper models, routing, spend visibility, and cutting token overhead, unit costs in some scenarios fell by as much as 90%. He noted that hard budget caps would interrupt the best developers, so the company switched to gradual friction through warnings, approvals, and model downgrades, and productized its internal solution as Unity Gateway. The team built its own evaluation using real PRs from its own codebase, finding that GLM 5.2 matched Opus 4.8 in quality at about one-third lower cost per task, while a model that was 1.7 times cheaper per token ended up costing more per task because it consumed more tokens.

#Industry news#English text is being prepared.#Agents

EElon Musk02:16Opinion

AI score55

Musk reposts list of 8 tasks Grok Bot can hand off to an agent

Musk reposted Nima Owji's post with the caption True; the post says most people use Grok Bot like a chatbot, but it has its own computer and can keep working while you're away. The post lists 8 tasks you can hand off to it, including tagging @bot under any post on X, handling X DMs, connecting Coinbase, Robinhood or eToro, running daily briefings and reminders, comparing prices and asking before buying, opening GitHub PRs and shipping code, prepping meetings from Gmail, Calendar and Slack, and sorting your inbox, and says it asks before it sends, posts or buys.

Quoted postNima Owji · @nima_owji

Most people use Grok Bot like a chatbot! But it has its own computer and keeps working while you're away! 8 things you can hand off to it today: 1. Tag @bot under any post on X 2. Handle your X DMs 3. Connect Coinbase, Robinhood or eToro 4. Run daily briefings and reminders 5. Compare prices and ask before buying 6. Open GitHub PRs and ship code 7. Prep your meetings from Gmail, Calendar and Slack 8. Sort and clean your inbox And it asks before it sends, posts or buys anything. Which one would you hand off first?

#English text is being prepared.#English text is being prepared.#Agents

MMatt Turck01:14Opinion

AI score69

Mistral Research Lead Discusses Le Chonk: The Inside Story of the West's Largest Open Source Model Catching Up

Matt Turck speaks with Mistral research lead Pierre Stock about Le Chonk, described as the West's largest open source model. Stock discusses the team training on 3,800 GPUs in its own data center, the fact that Le Chonk is still in training, what Mistral borrowed from DeepSeek, and the trade-off between open-sourcing weights and refusal behavior for cybersecurity. The episode also mentions that Mistral Large 5 will be released later.

#English text is being prepared.#Open-source ecosystem#Reasoning

KKarina Nguyen00:13Opinion

AI score60

Karina Nguyen: AI agents need an authorization mechanism similar to OAuth

Karina Nguyen believes AI agents need an authorization mechanism similar to OAuth, and the difficulty lies in defining what "handle this for me" actually authorizes, especially when an agent delegates further to another agent. She points out that cryptography can verify permissions and audit records, but the bigger problem is ensuring that a chain of reasonable decisions does not turn into something the user never agreed to. The tweet cites an Axios report that multiple U.S. banks have detected unauthorized third-party AI agents inside customer accounts.

Quoted postMadison Mills · @MadisonMills22

SCOOP: Major U.S. banks are detecting unauthorized third-party AI agents inside customer accounts, including some attempting to take actions that appear to be independent of user instructions, sources told Axios. The findings have alarmed banking executives, who are racing to restrict access to their systems as AI moves at a velocity faster than financial institutions can respond, multiple industry and bank executives told Axios. Several banks are now preparing for worst-case and even “day after” scenarios, including how to rapidly secure or restore customer accounts if things go wrong, the people said. Several bank executives told Axios they're struggling to distinguish between human users and AI agents. One executive said banks also struggle to determine whether agents are authorized by customers or potentially operating without their permission, more akin to rogue agent behavior. Read my latest for Axios here: https://t.co/Q3bda5In8q

#English text is being prepared.#Agents#Safety / alignment

GGary Marcus：The Road to AI We Can Trust00:00Opinion

AI score58

Gary Marcus: CoreWeave stock has fallen more than 50% in six months

Gary Marcus pointed out that CoreWeave's stock price has fallen more than 50% over the past six months, even though no one in the tech industry wants it to collapse. He posted the stock price trends for the past six months and the past week, and said that a rally on Monday of this week came from a group of people hyping the stock on television.

#Industry news#English text is being prepared.

SShao Meng23:41Opinion

AI score67

Claude Code creator Boris Cherny shares five practical Agent principles in a Stanford class

Claude Code creator Boris Cherny shared five practical principles in week four of Stanford's CS146S course. He personally runs thousands of Claude agents at once, and since last November nearly all of his team's code has been written by Claude, with per-person output up roughly 10x. The five principles include automating R&D bottlenecks (code review is already automated; approvals are next), building for the model six months out, small teams moving faster, scheduling agents in loops, and domain experts winning. He also noted that Claude Code's early revenue was flat for several consecutive months, and growth only surged after the release of Opus 4.

Quoted postMihail Eric · @mihail_eric

I asked @bcherny how many Claude agents he's running right now as he sat next to me. "A few thousand." Boris (creator of Claude Code) came back to my Stanford course 𝗧𝗵𝗲 𝗠𝗼𝗱𝗲𝗿𝗻 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿 last Thursday. His main takeaways: - 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗲 𝘆𝗼𝘂𝗿 𝗯𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸: Claude has written 100% of his team's code since November and output is up ~10x per engineer. So much code shipped that people stopped reading it carefully and started skimming, so they automated code review. Approvals are next. The team identified the main things impeding their development velocity and built systems to unblock themselves. - 𝗕𝘂𝗶𝗹𝗱 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹 𝟲 𝗺𝗼𝗻𝘁𝗵𝘀 𝗼𝘂𝘁: early Claude Code revenue was flat for months because the model wasn't smart enough yet. Then Opus 4 came out and growth became astronomic. His method is to read your agent transcripts and see where the model fails. That's where it will improve in the next model release. - 𝗦𝗺𝗮𝗹𝗹 𝘁𝗲𝗮𝗺𝘀 𝗴𝗼 𝗳𝗮𝘀𝘁𝗲𝗿: for months Claude Code was basically just him. With AI, one or two people can outship a team 100 times their size. Constraints breed creativity. - 𝗧𝗵𝗶𝗻𝗸 𝗶𝗻 𝗹𝗼𝗼𝗽𝘀: his grandpa programmed punch cards in the Soviet Union. Boris writes loops that run for days or weeks and spawn many agents. He expects that typing code by hand is heading where punch cards went. - 𝗗𝗼𝗺𝗮𝗶𝗻-𝘀𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗯𝘂𝗶𝗹𝗱𝗲𝗿𝘀 𝗮𝗿𝗲 𝘁𝗵𝗲 𝗳𝘂𝘁𝘂𝗿𝗲: he expects the best accounting software of 2027 to be built by an accountant. When code is cheap, knowing the problem deeply is the only alpha In the 50 years after the printing press, Europe produced more literature than in the previous 1,000. He thinks coding is about to undergo a similar explosion of creation.

#English text is being prepared.#Agents#English text is being prepared.

336Kr · AI23:37Opinion

AI score58

AI comic drama quotes fall from 5,000 yuan/minute to just tens of yuan; one company can produce 600 episodes a day

Multiple practitioners told the "BUG" column that AI comic drama production quotes have dropped from 5,000 yuan/minute at the start of the year to 200 yuan/minute or even just tens of yuan/minute, with low-end motion comics costing as little as a few hundred yuan per episode. The price decline stems from a production capacity explosion; one comic drama company can produce 600 episodes a day at its limit, while traffic-buying costs account for about 60% of overall costs, and works by small and medium teams mostly sink to the bottom of the content pool. Interviewees also mentioned that some AI training classes recruit students with tuition of about 20,000 yuan, effectively turning trainees into cheap card-drawers.

#English text is being prepared.#Video#Industry news

336Kr · AI23:37Opinion

AI score78

Pichai Recaps Gemini 4 Argon Launch: Explains Why 3.5 Pro Was Cut and Teases 5.0

Google CEO Pichai revealed in an interview on the podcast "Sources" that Gemini 4 Argon processed about 25 trillion tokens within 72 hours of launch, roughly 3 times the same period for Gemini 3.0, and that access was restricted early on due to compute constraints. He explained that Gemini 3.5 Pro was abandoned because the team found a gap in Agentic Coding and instead spent about 3 months continuously iterating from 3.5 to 3.8 Flash; by the time the Pro approach was proven out, Argon pretraining was already complete, so they injected the accumulated work into Argon post-training. He also mentioned that internally they have entered an RSI phase in which researchers and models collaborate, and said the next-generation Gemini 5.0 will be far stronger than Argon.

#English text is being prepared.#Google#Agents

TThe New Stack23:34Opinion

AI score71

The PHP2Shell incident is just a preview: AI is finding vulnerabilities faster than maintainers can fix them

The New Stack reported that two critical PHP vulnerabilities collectively known as PHP2Shell could allow unauthenticated attackers to execute malicious code. The PHP project has released emergency fixes for versions 8.2 through 8.5, Cloudflare has deployed firewall protection, Automattic confirmed that WordPress.com, WordPress VIP, and Tumblr have been hardened, and the Wikimedia Foundation has taken several auxiliary services offline while awaiting patches. The report noted that OpenAI said the security firm Calif used GPT-5.6 Sol to discover one of the vulnerabilities, Google's Wiz independently discovered both vulnerabilities, and Chainguard released a fixed image more than six hours early, disrupting coordinated disclosure. Multiple maintainers said that AI makes finding vulnerabilities easier, but judgment and fixes still depend on experienced maintainers, and small projects in particular struggle to cope with the flood of reports.

#Safety / alignment#Open-source ecosystem#Agents