NAll news

All news

Saturday46 items

  1. AYi AI NotesOpinion
    AI score76

    Kimi K3.1 Pro Price Compared with Opus 5.5 and GPT-6 Astra, Fully Open-Sourced in Early November

    The author compared the prices of Kimi K3.1 Pro, Opus 5.5, and GPT-6 Astra: Kimi K3.1 Pro's input, output, and cache-hit prices are all only one-fifth of GPT-6 Astra's, and its input and output prices are also only half of Opus 5.5's. The author states that Kimi K3.1 Pro comprehensively surpasses Opus 5.5 on coding and terminal tests, and will be fully open-sourced in early November. Another tweet cited gives prices of $2 per million tokens for input, $10 for output, and $0.20 for cache hits, with the Highspeed version charging double for both input and output.

    Quoted postAYi · @AYi_AInotes

    English text is being prepared.

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

  3. Rohan PaulOpinion
    AI score71

    AI finds vulnerabilities faster than organizations can fix them, and attackers keep pace

    Rohan Paul cited data pointing out that AI now discovers software vulnerabilities faster than organizations can fix them. By late July 2026, the U.S. National Vulnerability Database had recorded 45,207 vulnerabilities, close to the record total for all of 2025; a single July update from Google fixed 433 Chrome vulnerabilities, compared with 11 a year earlier, while Oracle fixed 1,449 versus 309. In this week's PHP emergency, banks, airlines, and governments took websites offline because of 2 AI-discovered vulnerabilities; Palo Alto Networks' Unit 42 estimates that more than one-third of the 22 million related internet-exposed systems run software that cannot be patched, and upgrades are expected to take months.

    #English text is being prepared.#Safety / alignment

  4. Rohan PaulOpinion
    AI score68

    Damodaran says Anthropic IPO will be priced on ARR rather than cash flow valuation

    New York University professor Aswath Damodaran wrote that Anthropic's IPO will be priced on run-rate revenue rather than cash flow, and that a $1.5 trillion to $2 trillion valuation requires conditions such as early-stage high-speed growth, operating margin rising above 30%, and compute cost growth slower than revenue. He pointed out that ARR has no unified definition: OpenAI's $50 billion is calculated on net cooperative sales, while Anthropic's $65 billion is calculated on a gross basis; converting OpenAI to a gross basis yields the roughly $70 billion circulated among investors. He also mentioned that Anthropic's May funding round raised $65 billion at a post-money valuation of $965 billion, while the expected IPO pricing at that time was $1.15 trillion.

    Quoted postAswath Damodaran · @AswathDamodaran

    As Anthropic's IPO nears its target date, it it worth remembering that IPOs get priced, and that as a young company with growing revenues and big losses, that pricing will be based on metrics that are buzzy and unfamiliar, from eyeballs in the dotcom era to users in the social media decade to ARR with AI companies...https://t.co/5DXmDGcSTf

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

  5. Aaron LevieOpinion
    AI score55

    Aaron Levie: Agent clusters for cyber defense will become one of AI's largest categories

    Box CEO Aaron Levie believes that agent clusters used for cyber defense are about to become one of the largest categories in the AI field. He judges that security is the most suitable use case for agent clusters, because in this field the more compute is invested, the higher the value produced, and a large number of tests that are hard for humans to cover or complete in time can be handled by agents. He cites an Enclave announcement saying that attackers are already using cheap AI agent clusters to probe for weaknesses around the clock.

    Quoted postEnclave · @EnclaveAI

    Big news: Enclave is now open to everyone, and today we're launching Threat Feed. Enclave is an autonomous security team for AI-native companies: a fleet of security agents that test your code, cloud, and infrastructure the way an attacker would. We built it on a simple belief. Attackers already use swarms of cheap AI agents to probe for weaknesses around the clock, so defenders need agents of their own to keep up. A small team without a security function should be able to defend like a large one. Threat Feed is the newest piece: a live feed of new vulnerabilities, checked against your stack, so when the next big disclosure drops you know whether it affects you. This week made the case. On Wednesday, PHP2Shell, LiteLLM and Railway all disclosed security issues. PHP2Shell had sat in PHP's code for over a decade until one researcher using frontier AI turned it into a working exploit in about a week. Cloudflare saw thousands of exploit attempts in the 10 hours before disclosure and millions since. Starting today, anyone can sign up. Try it free for 14 days: https://t.co/KEjswP2ZMh

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

  6. Guillermo RauchOpinion
    AI score60

    Guillermo Rauch on cache timing attacks by agents on shared hosts

    Vercel CEO Guillermo Rauch commented on the agent side-channel issue discovered by OpenAI, calling it a "noisy neighbor" attack: two agents on the same machine can create noise through the CPU cache, while the other agent probes by timing its own computation and cache accesses. He noted that such timing attacks on shared-host VMs have been known at least since 2015, and that there is no complete solution; the only "complete" approach is exclusive hardware per agent, but the economic cost is significant. He said Vercel Sandbox has deployed all feasible mitigations for this class of attack, along with continuous monitoring and red-team testing.

    Quoted postroon · @tszzl

    ‘In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache’

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

  7. Noam BrownOpinion
    AI score72

    Noam Brown: Using large-scale agent teams to red-team infrastructure vulnerabilities

    Noam Brown stated that last month a large agent team consumed over 100 billion tokens to solve the Navier-Stokes Millennium Prize Problem, and the team is now using a similar approach for red-team testing to find and fix infrastructure vulnerabilities. He cited an Eric Wallace post saying the team built a large team composed of the models with the strongest cyber offense and defense capabilities, collaborating in an offline training environment to find zero-day vulnerabilities; although they did not find a path to bypass CoT monitoring to reach the internet, they did find methods to escape mainstream virtual machines, exploit network infrastructure, and communicate across virtual machines via hardware side channels, and the related findings were used to harden infrastructure and submit a large number of patches upstream to open-source software packages.

    Quoted postEric Wallace · @Eric_Wallace_

    To make sure our frontier models don't cause another HuggingFace incident, we built an offline world that they use during training. It provides everything that the models need in a fully cached, indexed, or synthetic manner. We then built a huge team of our most cyber-capable models and told them to collaborate to find zero-days in this containment infrastructure. While they didn't find a path to the internet that wouldn't be caught by our CoT monitors, they did find ways to escape popular VMs, exploit our network infrastructure, and even ways to communicate with other agents running in separate VMs by using hardware side channels. We used this to dramatically harden our infra, and we also upstreamed a ton of patches to open-sources packages.

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

  8. Nathan LambertOpinion
    AI score64

    Nathan Lambert: DeepSeek processes 150T tokens per day on a single cluster, more than the combined total of several open-source inference platforms

    Nathan Lambert says DeepSeek processes 150T tokens per day on a single cluster, more than the combined open-source model inference volume on OpenRouter, Fireworks, Together, and BaseTen. He also posted a reconstructed comparison chart of his own based on these companies' inference volume reports, and noted that this figure corresponds to only a single cluster; it remains unclear how many clusters DeepSeek actually has.

    Quoted postEnglish text is being prepared. · @teortaxesTex

    @natolambert right from the horse's mouth ah, right. That's "a single cluster", who knows how many they have https://t.co/lruIKnDwCX https://t.co/Uq5rYyr5N0

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

  9. TMTPost · AGIOpinion
    AI score66

    Agent, robots, and pharma: who will be rewritten first by GPT-6 Astra?

    Silicon Valley 101 invited Pokee.ai founder Zhu Zheqing, Pokebot founder Xu Huazhe, and University of Washington assistant professor Wang Sheng to discuss the impact of GPT-6 Astra on three directions: personal Agent, robotics, and AI pharma. Xu Huazhe said Astra brings qualitative changes to robotic tasks, taking first place on simulation benchmarks such as RoboDojo, but dual-arm tasks and dexterous manipulation still have obvious shortcomings; Wang Sheng said the team has switched from models such as Claude to Astra for long-horizon biomedical tasks, but the completion rate for designing experiments is only about 30%. Zhu Zheqing believes Astra has steadily improved in long-horizon tasks and hallucination control, but is rather slow, and has not yet solved the problem that hallucinations still occur after running all night.

    #Industry news#Agents#Reasoning

  10. Fortune AIOpinion
    AI score80

    OpenAI model finds PHP vulnerability affecting about 70% of websites

    In August 2026, researchers at security research firm Calif used an advanced OpenAI AI model to discover a vulnerability in the PHP core system. The vulnerability affects about 70% of websites, and attackers can use it to gain shell access to servers; the related exploit code is called PHP2Shell. The vulnerability information leaked before the PHP organization released a patch, and AWS said hackers had begun "widespread" attempts to exploit it. The article points out that AI can analyze massive codebases in hours or days and find flaws that manual audits might miss, but attackers can also use the same reasoning capabilities to scan for and automate the exploitation of vulnerabilities.

    Same story · 6 reports

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

  11. Exponential ViewOpinion
    AI score61

    Azeem Azhar: AI is turning old vulnerabilities into serious problems, and the internet will become safer but more closed

    Azeem Azhar believes the current internet is under intense attack, with AI greatly increasing the scale and speed of exploiting old vulnerabilities; PHP2Shell, Node.js vulnerabilities, and the AI agent cluster intrusion disclosed by Microsoft are contemporaneous cases. He cites Cloudflare head Matthew Prince's prediction that within five years the number of AI agents will reach 1,000 times the number of human users, and judges that attacks will worsen first, with the internet ultimately left safer but less open. The article also mentions that Anthropic once delayed the release of Claude Mythos, while the open-source model GLM 5.3 had by August approached its level.

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

  12. WallstreetcnOpinion
    AI score74

    From CoreWeave to Anthropic, chip-collateralized financing is reshaping the AI debt market

    Chip-collateralized financing began in 2023 when Blackstone provided $2.3 billion in private credit to CoreWeave; now Broadcom is in talks with Wall Street institutions to arrange $60 billion in chip financing for Anthropic, while Nvidia announced a $500 billion cooperation framework with six major lenders. JPMorgan analysts estimate total chip spending could exceed $3 trillion before 2030, most of which will rely on debt financing. The article also notes that such deals are mostly private in nature with limited exit channels, cross-exposures among chipmakers, cloud computing giants and emerging cloud service providers are highly concentrated, and CoreWeave's corporate junk bonds have recently fallen to around 80 cents on the dollar with a yield of about 15%.

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

  13. East MoneyOpinion
    AI score58

    21 Comment: How long can this round of AI investment boom last

    A 21st Century Business Herald commentary argues that the risk release from this round of AI investment boom will show structural stratification, with the endgame closer to the bursting of the 2001 internet bubble than the 2008 subprime crisis. Citing data from the Bank for International Settlements' Bulletin No. 137, the article states that between 2021 and 2025, 55.2% of investment received by AI companies came from other AI companies, and estimates that for current compute investment to earn a reasonable return, the global AI industry's annual revenue would need to reach 2.5 trillion to 3 trillion US dollars by 2030. The article also notes that cloud giants are shifting the risks of heavy-asset depreciation and overcapacity onto emerging compute leasing providers that are borrowing at high interest to expand capacity, and onto chipmakers that provide compute backstops.

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

Friday66 items

  1. Latent SpaceOpinion
    AI score66

    Why AI Is Suddenly Good at Math: Understanding Model Problem-Solving Patterns from Prompt Sequences and Branching Factors

    Latent Space author Ankit Maloo analyzes that OpenAI found a counterexample to the Erdős unit distance conjecture in May, then went on to give solutions to ten open problems including Cycle Double Cover, and Anthropic also released a formalization of Fermat's Last Theorem and bounds related to the Riemann hypothesis. The article borrows Gowers's prompt sequence complexity 2^k + t and branching factor b to explain this phenomenon: the model runs every feasible path to completion in a depth-first manner, then relies on a verifier to prune, and failed attempts enter the next round as negative prompts, so the cost of solving problems shifts from human time to compute. Based on this, the author judges that hard problems whose conceptual leaps have already been made by others and that have few branches are easier for models to advance, while directions with deep search trees and many branches remain where human intuition has the advantage.

    #English text is being prepared.#Reasoning#Papers/Research

  2. WallstreetcnOpinion
    AI score73

    Three frontline researchers discuss how GPT-6 Astra rewrites Agents, robots, and pharmaceuticals

    Silicon Valley 101 invited Pokee.ai founder Zhu Zheqing, PoKe Robot founder Xu Huazhe, and University of Washington assistant professor Wang Sheng to discuss the impact of GPT-6 Astra on three paths: personal Agents, robots, and AI pharmaceuticals. Xu Huazhe said Astra brings qualitative changes to robot tasks, with benchmarks such as RoboDojo topped by it, but it is still slow and has shortcomings in dual-arm tasks and dexterous manipulation; Wang Sheng said the team has switched from models such as Claude to Astra to handle long-horizon biomedical tasks, but the completion rate for designing experiments is about 30%. Zhu Zheqing believes Astra has steadily improved in long-horizon tasks and hallucination control, but the commercialization of personal Agents is still stuck on cloud IPs being blocked by merchants and identity verification issues.

    #Industry news#Agents#Embodied AI

  3. Rohan PaulOpinion
    AI score58

    Sam Altman says OpenAI's current strongest internal model will not be released for now

    Sam Altman said that OpenAI's current strongest internal model is too capable and will not be openly released in the short term. He believes that before release, the cybersecurity protection system of the entire internet needs to change, and the world also needs an additional layer of protection against biological threats. The related remarks come from a video on the Stanford Graduate School of Business YouTube channel.

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

  4. 36Kr · AIOpinion
    AI score62

    Embodied intelligence investors collectively lose money: 13 robotics companies listed in Hong Kong, 8 broke issue price

    According to statistics from 36Kr's Silicon-based Observation Pro, 13 robotics companies listed in Hong Kong this year, of which 8 broke issue price; cornerstone investors in 10 companies invested a combined HK$4.861 billion, with a theoretical net paper loss of HK$855 million based on the latest share prices. The median per-share book return of the final financing round for 12 companies was only 70.4%; a further 41% drop in share price would wipe out the profit. The institutional holdings of the 13 companies are conservatively estimated at over HK$20.9 billion, of which 8 had average daily turnover of less than HK$20 million over the past 20 trading days; a static calculation puts the median time required to fully liquidate at 473 trading days.

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

  5. 36Kr · AIOpinion
    AI score58

    AI-cloned games proliferate, indie developer's playable demo forced to cancel

    A 36Kr reposted article sorts through several AI game-cloning incidents: Alliance Arts, fearing that the art and gameplay of its new title "The Drunkard Goddess's Wine Trick" would be quickly imitated by AI, canceled its playable demo originally scheduled for October 19 and switched to a closed test; after the trailer for "Needle in a Haystack Simulator" went viral, works with similar names and gameplay appeared on Steam, with the creators themselves disclosing they used AI. The article also mentions that 25 hours after the preview video for "Escape the Walls" was released, a clone generated with Claude appeared on Roblox, and that developer Takashi Mochibe received an email demanding $200 or else an AI clone would be rushed out.

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

  6. The New StackOpinion
    AI score67

    PHP2Shell Is Just the Beginning: AI Is Accelerating the Vulnerability Patch Race

    Matthew Weier O'Phinney, PHP Foundation board member and Perforce Zend product lead, reviews the PHP2Shell (CVE-2026-109618 and CVE-2026-109619) incident: these two remote code execution vulnerabilities at the PHP runtime level affect all versions since 5.4; government websites in several European countries were temporarily taken offline awaiting patches, the PHP project released fixes for all supported versions on the morning of October 21, and LTS vendors simultaneously provided patches for EOL versions. He points out that about 40% of PHP users are still running unsupported versions, while AI is making vulnerability discovery faster; in OpenAI's Patch the Planet program, researchers have already used GPT-5.6 Sol to find this vulnerability, and OpenAI expects similar discoveries to increase significantly in the coming months.

    #Safety / alignment#Open-source ecosystem#OpenAI

  7. The Guardian TechnologyOpinion
    AI score60

    Experts Question Anthropic's Report on AI Changing Company Employment

    The Guardian reported that Anthropic said in a report released on Thursday that since the release of Claude Code in March 2025, the share of its entry-level technical job hires fell from nearly 30% to 9%, employees spend about 22% of their time managing Claude agents, and starting in May 2026, the working hours recorded by AI agents exceeded those of human employees. Several researchers questioned the report. Morgan Frank of Case Western Reserve University showed in a peer-reviewed paper that the job market affected by AI had already deteriorated before ChatGPT was released; Julián Posada of Yale believes the report lacks independent third-party auditing and is more like hype for an IPO. Anthropic did not respond to The Guardian's request for comment.

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