Quinten Farmer announced in a parody tone the launch of the "Cute Little Guy" model Orb, claiming it achieves SOTA on all CLG benchmarks and asserting that Orb is the first CLG model to demonstrate sandbox escape capability, calling on other labs to keep the cuteness frontier at two little guys. Dan Shipper reposted the post, commenting only "very concerning".
Matched fieldsSummary
Thariq believes that when using coding agents, one can fall into a kind of frenzy: feeling like you're constantly working and getting a lot done, when in fact nothing is really happening. He thinks this is often an escape from other things in life.
Matched fieldsSummary、Body
Anthropic has trained Claude to perform a preliminary safety assessment of the conversation after each response; this technique, called self-assessment, has since it was enabled in June surfaced misuse that other safeguards were not designed to catch. The mechanism takes advantage of Claude having already read the conversation, costs less than classifiers, and improves in accuracy as the model improves. Rohan Paul adds that Anthropic's most accurate misuse detector is the Claude Mythos Preview classifier, which must first read all safety policies and is too expensive to run; Anthropic therefore trained Claude to replicate its judgments without reading the policies or the reasoning process, and on Fable 5 traffic this replica matched 78% of the detector's flags, rising to 94% for high-priority cases, with a false positive rate of 1%.
Anthropic has trained Claude to perform a preliminary safety assessment of the conversation after each response; this technique, called self-assessment, has since it was enabled in June surfaced misuse that other safeguards were not designed to catch. The mechanism takes advantage of Claude having already read the conversation, costs less than classifiers, and improves in accuracy as the model improves. Rohan Paul adds that Anthropic's most accurate misuse detector is the Claude Mythos Preview c
Matched fieldsSummary
Anthropic has launched a new technique called self-assessment, which lets Claude immediately conduct a preliminary safety assessment of the entire conversation after responding. On Claude Fable 5, this mechanism flags 78% of conversations that a stronger prompt classifier deems potentially harmful, and 94% of high-priority conversations, with a 1% false positive rate. Anthropic says that since self-assessment was enabled in June, it has surfaced abusive behavior that other safeguards failed to cover, and that each successful attack now consumes about four times as many generated tokens as it would without the protection.
Anthropic has launched a new technique called self-assessment, which lets Claude immediately conduct a preliminary safety assessment of the entire conversation after responding. On Claude Fable 5, this mechanism flags 78% of conversations that a stronger prompt classifier deems potentially harmful, and 94% of high-priority conversations, with a 1% false positive rate. Anthropic says that since self-assessment was enabled in June, it has surfaced abusive behavior that other safeguards failed to c
Matched fieldsSummary
Anthropic trained Claude to give a preliminary safety assessment of the conversation after replying; this technique is called self-assessment and is used to flag potential abuse. Since the mechanism was enabled in June, it has detected abusive behavior that other safeguards failed to cover. Anthropic said that because Claude has already read the conversation, self-assessment is highly efficient, and unlike classifiers targeting fixed harm categories, it relies on the model's own judgment and will become more accurate as the model improves.
Anthropic trained Claude to give a preliminary safety assessment of the conversation after replying; this technique is called self-assessment and is used to flag potential abuse. Since the mechanism was enabled in June, it has detected abusive behavior that other safeguards failed to cover. Anthropic said that because Claude has already read the conversation, self-assessment is highly efficient, and unlike classifiers targeting fixed harm categories, it relies on the model's own judgment and wil
Matched fieldsSummary、Body
On October 29, 2026, Google added the ability to directly create, preview, and edit files to the Gemini app and web version, supporting documents (text, Markdown, LaTeX), PDFs, presentations, and spreadsheets. Users can select content and ask Gemini to modify text or visual elements, and can continue asking follow-up questions to make adjustments. Generated files can be downloaded in their original format, opened in apps such as Google Slides, or uploaded to Google Drive via the export button in the upper right corner. PDF preview and editing are supported on both web and mobile, while other formats are supported only on the web; this feature is available to all languages and users who can use the Gemini app.
On October 29, 2026, Google added the ability to directly create, preview, and edit files to the Gemini app and web version, supporting documents (text, Markdown, LaTeX), PDFs, presentations, and spreadsheets. Users can select content and ask Gemini to modify text or visual elements, and can continue asking follow-up questions to make adjustments. Generated files can be downloaded in their original format, opened in apps such as Google Slides, or uploaded to Google Drive via the export button in
Matched fieldsSummary、Body
Emro has signed a contract with GS Caltex to develop an AI procurement automation agent that will handle approximately 2,700 small-value material purchases per year for the latter. The agent will analyze procurement requests and complete the entire process from requesting quotations and evaluating bids to selecting suppliers and issuing purchase orders; cases requiring additional review will be handed over to procurement staff for manual processing. The project is based on Emro's self-developed AI Studio platform and will be integrated with GS Caltex's existing procurement system, with plans to expand to other procurement operations beyond small-value purchases in the future.
Emro has signed a contract with GS Caltex to develop an AI procurement automation agent that will handle approximately 2,700 small-value material purchases per year for the latter. The agent will analyze procurement requests and complete the entire process from requesting quotations and evaluating bids to selecting suppliers and issuing purchase orders; cases requiring additional review will be handed over to procurement staff for manual processing. The project is based on Emro's self-developed
Matched fieldsSummary、Body
Grok @bot 0.78.1 has entered the stable version and begun rolling out. Updates include: shopping results displayed as product cards, new subscription templates, the ability to connect Shop Pay in the vault, and more other changes.
Matched fieldsSummary
Generalist announced that it is expanding access to GEN-1.5, its foundation model for the physical world, and has opened a waitlist for users to sign up and queue. The company also provided a detailed introduction link for GEN-1.5, as well as demonstrations of other tasks the model can complete.
Matched fieldsSummary
Generalist announced that it is expanding access to GEN-1.5, its foundation model for the physical world, and users can register via a link to join the waitlist. The tweet also provides links to demonstrations of other tasks by the model, as well as a detailed page introducing GEN-1.5.
Matched fieldsSummary、Body
ChatGPT announced that dot supports interaction via iMessage or RCS text messages; users can assign it tasks or have it keep tracking something using text, photos, or voice messages, and dot will proactively send messages to report on relevant progress. The feature is rolling out as a preview to Pro users in the US. The company says only the user themselves can message their own dot, dot can only reply to that person, and neither dot nor OpenAI can access the user's other text conversations. To set it up, select Texting on dot's profile page; dot shares the same context and plugins across different entry points.
ChatGPT announced that dot supports interaction via iMessage or RCS text messages; users can assign it tasks or have it keep tracking something using text, photos, or voice messages, and dot will proactively send messages to report on relevant progress. The feature is rolling out as a preview to Pro users in the US. The company says only the user themselves can message their own dot, dot can only reply to that person, and neither dot nor OpenAI can access the user's other text conversations. To
Matched fieldsSummary
Tripo announced that Agentic Canvas in Tripo Studio has entered the beta phase, available to all MAX and Team users, and will gradually roll out to other users. This feature lets users describe their requirements in natural language, with an agent planning and executing the workflow to directly produce 3D assets, supporting batch processing and reusable pipelines, and enabling asset inspection, refinement, rigging, and editing through AI 3D Lab.
Matched fieldsSummary
Alphabet reported third-quarter revenue up 25% year-over-year to $127.6 billion, with operating profit of $40.6 billion and an operating margin of 31.8%. Google Cloud revenue rose from $15.2 billion in the same period last year to $30.7 billion, with backlog reaching $561 billion; search and other advertising revenue grew 12% to $63.4 billion, slowing from 17% and 19% in the previous two quarters. The Gemini app surpassed 1 billion users, the first-party model API processes nearly 32 billion tokens per minute, and this quarter it launched Gemini 4 Argon and announced Gemini 4 Carbon.
Alphabet reported third-quarter revenue up 25% year-over-year to $127.6 billion, with operating profit of $40.6 billion and an operating margin of 31.8%. Google Cloud revenue rose from $15.2 billion in the same period last year to $30.7 billion, with backlog reaching $561 billion; search and other advertising revenue grew 12% to $63.4 billion, slowing from 17% and 19% in the previous two quarters. The Gemini app surpassed 1 billion users, the first-party model API processes nearly 32 billion tok
Matched fieldsSummary、Body
Tokyo University of Science successfully completed a bladder cancer surgery assisted by XREAL One Pro AR glasses; the patient was a man in his 50s, and postoperative progress was smooth. The system was jointly developed by the university's Department of Renal Urology and the Department of Electrical and Electronic Engineering team, superimposing endoscopic images within the doctor's field of view and connecting to Sony's NUCLeUS platform to centrally manage operating room video signals. Other participants can wear xbx a01+ to view information synchronously.
Tokyo University of Science successfully completed a bladder cancer surgery assisted by XREAL One Pro AR glasses; the patient was a man in his 50s, and postoperative progress was smooth. The system was jointly developed by the university's Department of Renal Urology and the Department of Electrical and Electronic Engineering team, superimposing endoscopic images within the doctor's field of view and connecting to Sony's NUCLeUS platform to centrally manage operating room video signals. Other pa
Matched fieldsSummary、Body
OpenAI has expanded access to the ChatGPT dot agent to iMessage and RCS; previously, the feature was only available within Slack or the ChatGPT app. Users can assign it tasks via text, photos, or voice messages, and it will proactively send messages to report progress; setup is done by selecting Texting in dot's profile, and it does not allow dot or OpenAI to access other text message conversations.
Matched fieldsSummary、Body
Google released the world model Genie 4, Apple launched a touchscreen laptop, and Microsoft's first-quarter results beat expectations. Nvidia announced it will subscribe to $3 billion of Meta's newly issued shares, and Meta will expand deployment of Nvidia's Vera Rubin computing platform, Vera CPU, and Spectrum-X networking equipment; this is the first time Nvidia has extended this circular transaction model beyond Meta to other "Magnificent Seven" stocks.
Matched fieldsSummary、Body
When Google's AI Mode was asked "Who won the 2020 presidential election," it did not answer directly, instead directing users to consult the official report of the U.S. Federal Election Commission (FEC). Tests showed that it similarly avoided the 2024 election, provided only an external link for the 2012 election, but was able to give detailed and correct answers about the winners of all other elections since 2000.
Matched fieldsSummary、Body
Google has released the frontier model Gemini 4 Carbon, claiming notable improvements in coding, agentic tasks, and cybersecurity, and ranking first on the Artificial Analysis Intelligence Index (v5.0), followed by Claude Opus 5.5 and GPT-6 Astra. The model is currently available only to select cybersecurity defense companies, and will later be expanded to other partners through the Fairwind Program; it ranks first on two benchmarks, SEC-bench Pro and ExploitGym, and is priced the same as Argon at $2 per million input tokens and $10 per million output tokens.
Google has released the frontier model Gemini 4 Carbon, claiming notable improvements in coding, agentic tasks, and cybersecurity, and ranking first on the Artificial Analysis Intelligence Index (v5.0), followed by Claude Opus 5.5 and GPT-6 Astra. The model is currently available only to select cybersecurity defense companies, and will later be expanded to other partners through the Fairwind Program; it ranks first on two benchmarks, SEC-bench Pro and ExploitGym, and is priced the same as Argon
Matched fieldsSummary、Body
Based on 20,869 traces from Agent Arena: Code, Arena compared the user complaint signals for five top-tier models from Anthropic and OpenAI against other agents, and found that frontier models differ most noticeably along three dimensions. Complaints about failure-repair loops were 6.9 percentage points lower, and complaints about ignoring project context were 1.9 percentage points lower; complaints about design issues were 6.4 percentage points lower, and complaints about producing usable results were 5.8 percentage points lower. Complaints about instruction following, task completion, and breaking existing code were 5.6, 4.9, and 4.5 percentage points lower, respectively.
Based on 20,869 traces from Agent Arena: Code, Arena compared the user complaint signals for five top-tier models from Anthropic and OpenAI against other agents, and found that frontier models differ most noticeably along three dimensions. Complaints about failure-repair loops were 6.9 percentage points lower, and complaints about ignoring project context were 1.9 percentage points lower; complaints about design issues were 6.4 percentage points lower, and complaints about producing usable resul
Matched fieldsSummary、Body
NVIDIA has released the open-weight speech transcription model Nemotron 3.5 Transcribe, with 30B total parameters and 4B activated parameters. On Artificial Analysis's non-streaming speech transcription leaderboard, it achieves a 2.8% AA-WER, tying for first among open-source models with Mistral's Voxtral Small. On the speech agent dialogue dataset AA-AgentTalk, it leads Voxtral Small's 2.3% with 1.8%. Its activated parameters are 4B, while Voxtral Small's are 24B; other open-source models of comparable scale have a WER of about 4%.
NVIDIA has released the open-weight speech transcription model Nemotron 3.5 Transcribe, with 30B total parameters and 4B activated parameters. On Artificial Analysis's non-streaming speech transcription leaderboard, it achieves a 2.8% AA-WER, tying for first among open-source models with Mistral's Voxtral Small. On the speech agent dialogue dataset AA-AgentTalk, it leads Voxtral Small's 2.3% with 1.8%. Its activated parameters are 4B, while Voxtral Small's are 24B; other open-source models of co
Matched fieldsSummary、Body