←Search results

English text is being prepared.

By relevance · Total 2760 results Clear search

Friday

  1. Story1 reports · 1 facts

    vLLM Blog Publishes Technical Retrospective on Kimi K3 Inference Optimization

    On October 16, 2026, the vLLM blog published a technical retrospective titled "Hyper Optimizing Kimi K3 in vLLM: Moving Fast and Responsibly." According to a TMTPost·AGI report, NVIDIA CentML engineer Benjamin Chislett used a Megakernel written in 12,000 lines of CuTe DSL to optimize inference for Moonshot AI's Kimi K3; the vLLM blog reports over 840 tokens/s at batch 1 and a 1.5x to 2x improvement in decoding throughput across multiple configurations, while SemiAnalysis estimates 423 tok/s/user on GB300 NVL72, a 1.68x improvement. The vLLM blog states that agent-optimized inference code is now outpacing human review and maintenance capacity, and the team therefore proposed a fast-track model optimization RFC so that reviewers judge designs from high-level abstractions rather than reviewing code line by line. The report also mentions that NVIDIA is developing optimizations for other models in vLLM, such as DeepSeek V4.1.

    On October 16, 2026, the vLLM blog published a technical retrospective titled "Hyper Optimizing Kimi K3 in vLLM: Moving Fast and Responsibly." According to a TMTPost·AGI report, NVIDIA CentML engineer Benjamin Chislett used a Megakernel written in 12,000 lines of CuTe DSL to optimize inference for Moonshot AI's Kimi K3; the vLLM blog reports over 840 tokens/s at batch 1 and a 1.5x to 2x improvement in decoding throughput across multiple configurations, while SemiAnalysis estimates 423 tok/s/user

    Matched fieldsSummary

  2. Story1 reports · 1 facts

    Khanna says Democrats will strictly scrutinize AI labs if they win the House

    Greg Casar, chair of the U.S. Congressional Progressive Caucus, said that if Democrats win a majority in the House of Representatives next week, they will push for strict scrutiny of AI labs and summon the CEOs of major AI companies to testify at hearings, not ruling out the use of subpoenas to compel their attendance. Previously, on October 29, Casar and 19 other Democrats sent a letter to OpenAI demanding a response regarding the dismissal of three safety researchers; OpenAI said the personnel decision was unrelated to the safety concerns raised. At present, this statement remains at the stage of a policy declaration before the election, with no subsequent developments.

    Greg Casar, chair of the U.S. Congressional Progressive Caucus, said that if Democrats win a majority in the House of Representatives next week, they will push for strict scrutiny of AI labs and summon the CEOs of major AI companies to testify at hearings, not ruling out the use of subpoenas to compel their attendance. Previously, on October 29, Casar and 19 other Democrats sent a letter to OpenAI demanding a response regarding the dismissal of three safety researchers; OpenAI said the personnel

    Matched fieldsSummary

  3. Story1 reports · 1 facts

    Cursor discloses malicious Claude Code extension incident in its extension marketplace

    Cursor disclosed that a tampered Claude Code VS Code extension appeared in its IDE extension marketplace, affecting up to 5,000 users, and stated that the malicious version was not developed by Anthropic. Cursor took the extension marketplace offline the same day, worked with the relevant service providers to remove the malware, restored the marketplace after confirming there was no further risk, and required all users to upgrade to Claude Code for VS Code version 2.1.314. Subsequent reports said the tampered extension was version 2.1.313, went live on October 29 and remained available for about 11 hours, and may have been downloaded by up to 5,000 developers; the malware attempted to install a persistent SSH backdoor, collected the names of environment variables related to usernames, hostnames, public IP addresses, Git email addresses and sensitive credentials and sent them to an attacker-controlled server, and also added an attacker-controlled SSH key to achieve persistence. Cursor has taken down the extension and temporarily shut down the entire extension marketplace to review other plugins and rotate access keys, while working with its hosting provider to shut down the attacker's server, but it has not yet explained how the attacker published the tampered package under Anthropic's name.

    Cursor disclosed that a tampered Claude Code VS Code extension appeared in its IDE extension marketplace, affecting up to 5,000 users, and stated that the malicious version was not developed by Anthropic. Cursor took the extension marketplace offline the same day, worked with the relevant service providers to remove the malware, restored the marketplace after confirming there was no further risk, and required all users to upgrade to Claude Code for VS Code version 2.1.314. Subsequent reports sai

    Matched fieldsSummary

  4. Story2 reports · 1 facts

    Toyota and Waymo announce next phase of autonomous driving partnership

    On October 30, 2026, Toyota and Waymo announced that their autonomous driving partnership has entered its next phase. According to information published on the Waymo blog that day, the two parties will begin initial testing and validation of a Toyota bZ4X autonomous driving fleet in Japan in 2027, and will then expand to the UK and other Waymo operating markets over the following years. The vehicles will be produced at Toyota's Takaoka plant in Toyota City, Aichi Prefecture, with both complete vehicle assembly and integration of Waymo's autonomous driving hardware completed at the same plant. Waymo co-CEO Dmitri Dolgov said that in Waymo's operating areas, its serious injury crash rate is up to 20 times lower than that of human drivers. Another Waymo release the same day said the partnership will expand the scale of the bZ4X autonomous driving fleet in Japan, with testing and validation beginning in 2027. No further progress has been disclosed at present.

    On October 30, 2026, Toyota and Waymo announced that their autonomous driving partnership has entered its next phase. According to information published on the Waymo blog that day, the two parties will begin initial testing and validation of a Toyota bZ4X autonomous driving fleet in Japan in 2027, and will then expand to the UK and other Waymo operating markets over the following years. The vehicles will be produced at Toyota's Takaoka plant in Toyota City, Aichi Prefecture, with both complete v

    Matched fieldsSummary

  5. Story1 reports · 1 facts

    Digital Life Kazik says most automated tasks have been switched to Qwen 3.8 Flash

    On October 30, 2026, Digital Life Kazik stated that Qwen 3.8 Flash performed unexpectedly strongly in some simple accuracy task evaluations. He said that in many of the automated tasks of the product he works on, more than half have already been switched from Claude Opus 5.5 to Qwen 3.8 Flash. Currently, this claim comes from content posted by Kazik himself, has not been cross-verified by other sources, and there have been no subsequent developments.

    Matched fieldsSummary

  6. Story1 reports · 1 facts

    ARC Prize says models scoring high on ARC-AGI-2 also tend to score high on ARC-AGI-3

    ARC Prize proposes a judgment: models that score higher on ARC-AGI-2 tend to also score higher on ARC-AGI-3. ARC Prize also points out an obvious exception—Gemini 4 Argon at the low reasoning tier scores only 42.1% on ARC-AGI-2, but reaches 97.5% on ARC-AGI-3 under the provider adapter harness. Currently this judgment is released by ARC Prize alone, and there are no other developments or facts yet.

    Matched fieldsSummary

  7. Story1 reports · 1 facts

    Rohan Paul says Anthropic's revenue curve will become legendary

    On October 28, 2026, Rohan Paul offered the judgment that Anthropic's revenue curve will become legendary. Citing a Bloomberg chart, he said that Anthropic's revenue in the first half of 2026 was nearly 400 times its full-year 2023 revenue, and that no other company in the chart comes close to that multiple; Nvidia, the second-fastest grower, remained below 50 times throughout its first three years after going public, while Alphabet, Meta, Microsoft and SpaceX essentially stayed near the 1x baseline. The same material also states that Anthropic's full-year 2025 revenue was about 100 times that of 2023, so its revenue in just six months of 2026 was about 4 times its full-year 2025 revenue. At present, this judgment appears only in this single source, Rohan Paul, and there have been no subsequent developments.

    On October 28, 2026, Rohan Paul offered the judgment that Anthropic's revenue curve will become legendary. Citing a Bloomberg chart, he said that Anthropic's revenue in the first half of 2026 was nearly 400 times its full-year 2023 revenue, and that no other company in the chart comes close to that multiple; Nvidia, the second-fastest grower, remained below 50 times throughout its first three years after going public, while Alphabet, Meta, Microsoft and SpaceX essentially stayed near the 1x base

    Matched fieldsSummary

  8. Story1 reports · 1 facts

    Rohan Paul said that Sunra resells multimodal model calls at the manufacturer's original price

    Rohan Paul stated on October 29, 2026 that Sunra has released the Sunra API, through which users can call all major image, video, audio, and 3D models with a single key, billed at each model vendor's own API prices on a pay-as-you-go basis, with no subscription fee and no markup. Rohan Paul also said that Seedance 2.5 is priced on Sunra at half of what two other gateways charge, and that this unified entry point also makes it easier to run A/B tests on models from Google, OpenAI, ByteDance, and Kling. At present, this claim appears only in Rohan Paul's account, and there is no independent evidence from Sunra officially or from other sources.

    Rohan Paul stated on October 29, 2026 that Sunra has released the Sunra API, through which users can call all major image, video, audio, and 3D models with a single key, billed at each model vendor's own API prices on a pay-as-you-go basis, with no subscription fee and no markup. Rohan Paul also said that Seedance 2.5 is priced on Sunra at half of what two other gateways charge, and that this unified entry point also makes it easier to run A/B tests on models from Google, OpenAI, ByteDance, and

    Matched fieldsSummary

  9. Story1 reports · 1 facts

    meng shao said he used the Liepin robot recruitment consultant as an AI assistant for a day

    On October 28, 2026, meng shao posted that he had used Liepin's robot recruitment consultant as an AI assistant for a day. His experience was: he did not hit any usage limit, but the replies were somewhat slow and somewhat dumb. Currently, this experience is only found in Meng Shao's (meng shao) own sharing, with no official response from Liepin or cross-verification from other sources.

    Matched fieldsSummary

  10. Story4 reports · 1 facts

    ChatGPT launches SMS interaction feature for dot

    ChatGPT announces the launch of an SMS interaction feature for dot: users can text dot via iMessage or RCS, assign tasks using text, photos, or voice messages, and also have it keep an eye on something continuously, with dot proactively sending messages to report progress. The feature is being rolled out in preview to Pro users in the US. According to the official statement, only users themselves can message their own dot, and dot can only reply to the user; moreover, dot and OpenAI will not be given access to users' other SMS conversations. To set it up, select Texting on dot's profile page; dot shares the same context and plugins across different entry points.

    ChatGPT announces the launch of an SMS interaction feature for dot: users can text dot via iMessage or RCS, assign tasks using text, photos, or voice messages, and also have it keep an eye on something continuously, with dot proactively sending messages to report progress. The feature is being rolled out in preview to Pro users in the US. According to the official statement, only users themselves can message their own dot, and dot can only reply to the user; moreover, dot and OpenAI will not be

    Matched fieldsSummary

  11. Story1 reports · 1 facts

    Jensen Huang proposes that doubling model speed could double AI lab revenue

    On October 29, 2026, Jensen Huang put forward a judgment: under the premise that demand exceeds capacity, doubling model speed (for example, through software or kernel optimization) could double AI lab revenue. The logic is that inference revenue equals the number of tokens served multiplied by the price per token; a faster kernel can increase tokens per second under the same GPUs, power, and data center conditions, and if all the additional tokens can be sold at the current price, a 2x speedup could double revenue with zero additional capital expenditure. Currently, this claim is presented in a report by Rohan Paul, with no other developments or official response yet.

    On October 29, 2026, Jensen Huang put forward a judgment: under the premise that demand exceeds capacity, doubling model speed (for example, through software or kernel optimization) could double AI lab revenue. The logic is that inference revenue equals the number of tokens served multiplied by the price per token; a faster kernel can increase tokens per second under the same GPUs, power, and data center conditions, and if all the additional tokens can be sold at the current price, a 2x speedup

    Matched fieldsSummary

  12. Story1 reports · 1 facts

    AYi asserts that now is the last golden window for ordinary people to do video self-media.

    On October 28, 2026, the WeChat public account "A Yi AI Notes" published an article. The author, AYi, handed a how-to article to Claude Opus 5.5 and wrote only a single prompt: "Make me a video based on this article." That was all it took to complete the entire video. On this basis, AYi made the assertion that now is the best era for ordinary people to do video self-media, and also the last golden window the era has left for individuals. He believes that Opus 5.5 has compressed a pipeline that previously required four or five specialized roles and several days of work into one person, a few prompts, and a finished product in just a few minutes. At present, this claim appears only in this one article, with no cross-verification from other sources or subsequent developments.

    On October 28, 2026, the WeChat public account "A Yi AI Notes" published an article. The author, AYi, handed a how-to article to Claude Opus 5.5 and wrote only a single prompt: "Make me a video based on this article." That was all it took to complete the entire video. On this basis, AYi made the assertion that now is the best era for ordinary people to do video self-media, and also the last golden window the era has left for individuals. He believes that Opus 5.5 has compressed a pipeline that p

    Matched fieldsSummary

  13. Story1 reports · 1 facts

    Lydia Hallie says dynamic workflows don't need to use hundreds of subagents

    Lydia Hallie put forward the view that dynamic workflows do not need to contain hundreds of subagents, and that she herself would use about 10 subagents to handle any task she wants to run repeatedly in the same way. She pointed out that a dynamic workflow is an editable JavaScript file, so it gives full control over the subagents and the order in which they run, making it more predictable than skills. Currently, this view was published by Lydia Hallie herself, and there are no other developments or facts yet.

    Lydia Hallie put forward the view that dynamic workflows do not need to contain hundreds of subagents, and that she herself would use about 10 subagents to handle any task she wants to run repeatedly in the same way. She pointed out that a dynamic workflow is an editable JavaScript file, so it gives full control over the subagents and the order in which they run, making it more predictable than skills. Currently, this view was published by Lydia Hallie herself, and there are no other development

    Matched fieldsSummary

  14. Story1 reports · 1 facts

    Shao Meng recounts a review of the 2026 OpenAI SWE interview process

    Shao Meng published a personal retrospective of the 2026 OpenAI software engineer (SWE) interview process. The retrospective shows that the process includes recruiter screening, two technical screens of about 60 minutes each on the same day, a take-home assignment of about 48 hours, a technical deep dive that probes the code line by line, and a 4-round onsite stage. The onsite coding is a progressive multi-part problem involving token-level streaming diff and rollback; the system design question is "Design ChatGPT," involving GPU allocation and autoscaling under non-stationary traffic. There is also an optional fifth round in beta, Agentic Coding, which requires using an AI coding agent to handle problems in large codebases that are difficult to address by hand. At present, this retrospective is Shao Meng's unilateral account, and no other sources have cross-corroborated it.

    Shao Meng published a personal retrospective of the 2026 OpenAI software engineer (SWE) interview process. The retrospective shows that the process includes recruiter screening, two technical screens of about 60 minutes each on the same day, a take-home assignment of about 48 hours, a technical deep dive that probes the code line by line, and a 4-round onsite stage. The onsite coding is a progressive multi-part problem involving token-level streaming diff and rollback; the system design question

    Matched fieldsSummary

  15. Story1 reports · 1 facts

    Vaibhav Srivastav said Codex's weekly active users have grown more than 14-fold since the beginning of 2026

    Vaibhav Srivastav reported on October 28, 2026 that OpenAI held a DevDay Exchange event in France, where it officially disclosed that Codex's weekly active users have grown more than 14-fold since the beginning of 2026. The same disclosure also mentioned that ChatGPT Business seats have grown nearly 70% over the past year, work users have tripled since launch, and France has entered the global top ten for active OpenAI API developers. Currently, this claim appears only in this one report, with no other developments or facts.

    Vaibhav Srivastav reported on October 28, 2026 that OpenAI held a DevDay Exchange event in France, where it officially disclosed that Codex's weekly active users have grown more than 14-fold since the beginning of 2026. The same disclosure also mentioned that ChatGPT Business seats have grown nearly 70% over the past year, work users have tripled since launch, and France has entered the global top ten for active OpenAI API developers. Currently, this claim appears only in this one report, with n

    Matched fieldsSummary

  16. Story1 reports · 1 facts

    Peter Yang says ChatGPT and browser use are extremely helpful for handling medical paperwork

    On October 29, 2026, Peter Yang stated that using ChatGPT together with the browser feature has been a great help in handling medical paperwork, and said this combination is extremely helpful for his handling of medical paperwork. Currently, this statement comes from information posted by Peter Yang himself, and there has been no other factual development or institutional response.

    Matched fieldsSummary

  17. Story1 reports · 1 facts

    Peter Yang predicts that farmers' markets will see a revival

    On October 29, 2026, Peter Yang published his view that the job AI can never replace is selling handmade trinkets at a local farmers' market, and on that basis predicted that farmers' markets are about to see a revival. At present, this judgment appears only in this article published by Yang himself, with no independent reporting or subsequent developments from other sources.

    Matched fieldsSummary

  19. Story1 reports · 1 facts

    Rime releases speech-to-speech model Prism

    Rime releases Prism, a speech-to-speech model for AI voice agents. According to Testing Catalog, Prism writes each reply as text before speaking it, allowing enterprises to edit the transcript and block things the agent should not say. The model also keeps the call uninterrupted and can look up accounts or freeze cards during the conversation. Rime says no other speech-to-speech model takes this approach, making Prism better suited to regulated enterprises such as banks. There is currently no information on further developments.

    Rime releases Prism, a speech-to-speech model for AI voice agents. According to Testing Catalog, Prism writes each reply as text before speaking it, allowing enterprises to edit the transcript and block things the agent should not say. The model also keeps the call uninterrupted and can look up accounts or freeze cards during the conversation. Rime says no other speech-to-speech model takes this approach, making Prism better suited to regulated enterprises such as banks. There is currently no in

    Matched fieldsSummary

Wednesday

  1. Story20 reports · 4 facts

    Google releases world model Genie 4

    On October 27, 2026, Google DeepMind released the world model Genie 4, focused on real-time interactive simulation and introducing native audio to the series for the first time. According to the official introduction, the model is built on capabilities from the Gemini family: Gemini Omni is responsible for making the simulation match the appearance and behavior of the real world, while other Gemini models provide the complex reasoning needed to maintain the simulation's coherence; users can control scenes in real time using natural language, and each world also instantly generates ambient sound, music, and interactive dialogue. The next day, Google DeepMind further explained that Gemini 3.8 Flash breaks down goals in a loop and checks frames to maintain causal coherence across multi-step events, and that Gemini 4 Argon has also been used to jointly simulate tasks with Gemini Robotics; the company said it remains an early experimental model, that object permanence and complex physical interactions still drift during long runs, that all generated content carries a SynthID watermark, and that it is currently being previewed to trusted testers first. On October 28, Ethan Mollick posted that Genie 4 can run DOOM, and tested it generating a game-like environment as it goes, saying the experience is not perfect. On October 29, Peter Yang used Genie 4 to build 5 worlds in real time, arguing that world models are the next big direction for AI after LLMs.

    On October 27, 2026, Google DeepMind released the world model Genie 4, focused on real-time interactive simulation and introducing native audio to the series for the first time. According to the official introduction, the model is built on capabilities from the Gemini family: Gemini Omni is responsible for making the simulation match the appearance and behavior of the real world, while other Gemini models provide the complex reasoning needed to maintain the simulation's coherence; users can cont

    Matched fieldsSummary