On October 22, MiniMax (0100.HK) closed down 6.52% at HK$160.7 per share, with total market value falling to HK$56.6 billion, falling below the IPO price of HK$165 per share for the first time since listing. Compared with the high of HK$1,330 per share in March, the stock price has fallen nearly 90%, and the value of founder Yan Junjie's stake has dropped from more than HK$100 billion to about HK$12.7 billion. The article attributes the pullback to the flagship model M3 not being updated for four months afterward, with only M3.1 Flash released, as well as an overall valuation re-rating of the AI sector.
On October 22, MiniMax (0100.HK) closed down 6.52% at HK$160.7 per share, with total market value falling to HK$56.6 billion, falling below the IPO price of HK$165 per share for the first time since listing. Compared with the high of HK$1,330 per share in March, the stock price has fallen nearly 90%, and the value of founder Yan Junjie's stake has dropped from more than HK$100 billion to about HK$12.7 billion. The article attributes the pullback to the flagship model M3 not being updated for fou
Microsoft announced the launch of its AI diagnostic system, Microsoft AI Diagnostic Orchestrator (MAI-DxO), which simulates a multidisciplinary expert consultation by orchestrating multiple AI agents. In an evaluation based on common cases from U.S. emergency departments using the MIMIC-IV-ED dataset, its diagnostic accuracy was 1.7 times that of human physicians; in a simulated evaluation based on complex cases from The New England Journal of Medicine, it achieved 83% accuracy, compared with 36% for primary care general practitioners and general internists who could independently consult reference materials, while specialists were not included in the comparison. Officials said the system's cost is only half or even one-third that of physicians, and the research findings have been published in the journal Science.
Microsoft announced the launch of its AI diagnostic system, Microsoft AI Diagnostic Orchestrator (MAI-DxO), which simulates a multidisciplinary expert consultation by orchestrating multiple AI agents. In an evaluation based on common cases from U.S. emergency departments using the MIMIC-IV-ED dataset, its diagnostic accuracy was 1.7 times that of human physicians; in a simulated evaluation based on complex cases from The New England Journal of Medicine, it achieved 83% accuracy, compared with 36
OpenAI announced the full opening of the Decisions API built on GPT-6 Luna, aimed at questions with a limited set of answers, with response speeds up to 10 times faster than calling GPT-6 Luna through the Responses API. The API charges $0.10 per million input tokens, $0.01 per million cached input tokens, bills only for input tokens, and output tokens are free. Since the public beta, it has added native GPT-Live integration, prompt caching, and inference data residency support in the United States and the European Union, and up to 400 questions can be submitted per request.
OpenAI announced the full opening of the Decisions API built on GPT-6 Luna, aimed at questions with a limited set of answers, with response speeds up to 10 times faster than calling GPT-6 Luna through the Responses API. The API charges $0.10 per million input tokens, $0.01 per million cached input tokens, bills only for input tokens, and output tokens are free. Since the public beta, it has added native GPT-Live integration, prompt caching, and inference data residency support in the United Stat
1Password launches the 1Password for Agents platform, enabling personal AI agents to complete website logins on users' behalf without accessing or storing users' passwords or verification codes. The feature is aimed at individual and family plan users, with initial support for agents including Meta Muse, OpenAI ChatGPT Work and Dots, Anthropic Claude Duet, Grok Bot, and Town. In the flow, the agent first proves its identity to 1Password using a digital key; the user sees on their device the account credentials and task requested for use and approves or rejects; after approval, 1Password directly fills in the username, password, or one-time verification code, and the password always remains in the vault.
1Password launches the 1Password for Agents platform, enabling personal AI agents to complete website logins on users' behalf without accessing or storing users' passwords or verification codes. The feature is aimed at individual and family plan users, with initial support for agents including Meta Muse, OpenAI ChatGPT Work and Dots, Anthropic Claude Duet, Grok Bot, and Town. In the flow, the agent first proves its identity to 1Password using a digital key; the user sees on their device the acco
Colossal Biosciences founder Ben Lamm and Anduril founder Palmer Luckey have co-founded defense technology startup Warhound, emerging from stealth with a $200 million seed round. The CEO is former Pentagon Chief Data Officer David Spirk, and the company is based in Texas. Warhound plans to preserve genetic samples from fully trained military working dogs, clone them, and perform targeted gene editing, while integrating Anduril's protective gear and Eagle Eye system. Lamm said the first dogs are expected to be delivered in about two years. The company currently has no signed customers and has not disclosed per-dog pricing, and is discussing a Canine-as-a-Service model with national security experts.
Colossal Biosciences founder Ben Lamm and Anduril founder Palmer Luckey have co-founded defense technology startup Warhound, emerging from stealth with a $200 million seed round. The CEO is former Pentagon Chief Data Officer David Spirk, and the company is based in Texas. Warhound plans to preserve genetic samples from fully trained military working dogs, clone them, and perform targeted gene editing, while integrating Anduril's protective gear and Eagle Eye system. Lamm said the first dogs are
Researchers at MIT CSAIL and Tianjin University have developed the X-Hinges design software, which lets users customize shapes and motion patterns; the software automatically embeds two different materials to separately fabricate sensors and conductive traces, avoiding signal interference. The tool supports motion in three directions—compression, up-down, and left-right—and was used to create a glove that can teach a robot to grasp, a small lamp that dims via pinching, and a device that can recognize placed objects. The research team will present the results at the ACM UIST conference in early November; currently, the circuit layout of each printed item differs slightly and requires recalibration using a computer vision system.
Researchers at MIT CSAIL and Tianjin University have developed the X-Hinges design software, which lets users customize shapes and motion patterns; the software automatically embeds two different materials to separately fabricate sensors and conductive traces, avoiding signal interference. The tool supports motion in three directions—compression, up-down, and left-right—and was used to create a glove that can teach a robot to grasp, a small lamp that dims via pinching, and a device that can reco
Stuut CEO and co-founder Tarek Alaruri discusses the company's completion of a $52.5 million funding round. Stuut builds an AI-driven order-to-cash platform that automates the accounts receivable work between issuing an invoice and receiving payment.
OpenAI opened the Decisions API to all customers on Thursday; the product was announced at the September 29 Developer Day and is based on a reconfigured ChatGPT Luna to speed up inference and reduce costs. Nikunj Handa, OpenAI's head of API product, said latency has been cut from 1 to 3 seconds to 100, 120, and 180 milliseconds; the public version adds voice and vision modalities and deploys inference resources by region to reduce latency. In early trials at Box, Datadog, Ramp, and Speak, Speak cut the action interval from 1.4 seconds to 0.13 seconds; on the AIM System One Index and the Artificial Analysis Decision Index, OpenAI's decision model ranks eighth and fourteenth, respectively, trailing Liquid's D1 Ultra and Base Lab's Elephant-1.
OpenAI opened the Decisions API to all customers on Thursday; the product was announced at the September 29 Developer Day and is based on a reconfigured ChatGPT Luna to speed up inference and reduce costs. Nikunj Handa, OpenAI's head of API product, said latency has been cut from 1 to 3 seconds to 100, 120, and 180 milliseconds; the public version adds voice and vision modalities and deploys inference resources by region to reduce latency. In early trials at Box, Datadog, Ramp, and Speak, Speak
This week, Amazon announced the open-sourcing of TorchNeuron, a backend that converts PyTorch code into operations executable on AWS Trainium; PyTorch code already running on other accelerators can be migrated with almost no changes. It offers an optional adaptive eager mode that groups operations for compilation and caches NEFF at multiple levels, cutting the time to the first training step by about one-third in cross-accelerator large language model tests. The team also used an AI agent to scan Hugging Face repositories, selecting 841 representative models and tens of thousands of tensor shapes to build a tiered testing process.
This week, Amazon announced the open-sourcing of TorchNeuron, a backend that converts PyTorch code into operations executable on AWS Trainium; PyTorch code already running on other accelerators can be migrated with almost no changes. It offers an optional adaptive eager mode that groups operations for compilation and caches NEFF at multiple levels, cutting the time to the first training step by about one-third in cross-accelerator large language model tests. The team also used an AI agent to sca
IBM proposes using "governed context" to support production-grade AI agents. Shaun Clowes, who leads its Confluent and data business, introduced the "streamhouse architecture," which combines streaming, connectivity, processing, governance and query capabilities to govern data close to where it is generated and deliver it directly to LLMs. Clowes stressed that agents need context and governed metrics tailored to specific tasks, and that enterprises can start with high-value tasks and let the system keep learning at every layer.
IBM proposes using "governed context" to support production-grade AI agents. Shaun Clowes, who leads its Confluent and data business, introduced the "streamhouse architecture," which combines streaming, connectivity, processing, governance and query capabilities to govern data close to where it is generated and deliver it directly to LLMs. Clowes stressed that agents need context and governed metrics tailored to specific tasks, and that enterprises can start with high-value tasks and let the sys
The Amazon Polly generative engine has completed an upgrade, adding 14 new voices across 9 locales and, for the first time, officially offering Hindi and Finnish in the generative engine. After the upgrade, it offers a total of 57 generative voices, covering 11 languages and 25 locales, and has expanded to AWS GovCloud (US-West), for a total of 11 AWS Regions. Existing voice names and API calls remain unchanged; this is an in-place upgrade with no price change and no code changes required.
Rohan Paul quotes Sundar Pichai on the Sources Podcast: AI labs are now living with compute bets made two years ago, when no company could have foreseen how much AI compute demand would grow. Pichai said that today, there is probably no company in the world that would claim its decision two years ago was the right one.
Legal AI company Legora has completed a $550 million Series E funding round led by Atreides Management, reaching a valuation of $8.8 billion. CEO Max Junestrand said the company does not build its own general legal model, but instead routes tasks to the most suitable frontier models, performing post-training only in narrow scenarios. It currently has more than 2,400 customers and over $200 million in annual recurring revenue. The funds will be used to advance Legora aOS, an agent operating system for legal work, enabling agents to take on complete tasks such as due diligence and e-discovery.
Legal AI company Legora has completed a $550 million Series E funding round led by Atreides Management, reaching a valuation of $8.8 billion. CEO Max Junestrand said the company does not build its own general legal model, but instead routes tasks to the most suitable frontier models, performing post-training only in narrow scenarios. It currently has more than 2,400 customers and over $200 million in annual recurring revenue. The funds will be used to advance Legora aOS, an agent operating syste
Microsoft Research has released Flint-interactive, an interactive data visualization language built on top of Flint that enables agents to generate charts supporting clicking, filtering, zooming, highlighting, and cross-chart linking. It uses the Flint compiler to parse raw triggers such as clicks and drags into semantic events, so users only need to describe update behaviors with high-level operators without writing low-level visualization code. The language is already usable in conversations through Flint-MCP and MCP Apps, and it also offers a programming API for developers to build dashboards and data stories.
Microsoft Research has released Flint-interactive, an interactive data visualization language built on top of Flint that enables agents to generate charts supporting clicking, filtering, zooming, highlighting, and cross-chart linking. It uses the Flint compiler to parse raw triggers such as clicks and drags into semantic events, so users only need to describe update behaviors with high-level operators without writing low-level visualization code. The language is already usable in conversations t
Anker Innovations' first AI Hackathon selected 33 teams from over 1,800 registrations across more than 180 universities and developer communities, completing the transformation from idea to demonstrable results within 24 hours, with 4 teams ultimately winning championships. The competition opened up Anker's real product SDKs and scenarios, with product innovation and AI nativeness, and user value and scenario implementation each accounting for 35% of the scoring. Anker disclosed that over the past year its total Token consumption reached 30 trillion, with AI handling over 95% of basic code generation, saving approximately 80% of development work hours.
Anker Innovations' first AI Hackathon selected 33 teams from over 1,800 registrations across more than 180 universities and developer communities, completing the transformation from idea to demonstrable results within 24 hours, with 4 teams ultimately winning championships. The competition opened up Anker's real product SDKs and scenarios, with product innovation and AI nativeness, and user value and scenario implementation each accounting for 35% of the scoring. Anker disclosed that over the pa
a16z announced that it is renaming a16z Crypto to a16z Frontier, expanding the fund's investment scope from crypto startups to a broader range of frontier startups while continuing to invest in crypto and related businesses. Chris Dixon will lead the fund together with Marc, Ben, and GPs Eddy Lazzarin and Guy Wuollet, focusing on AI-native use cases, novel hardware, machine networks, and the role of blockchain in machine-to-machine transactions.
Former Google CEO Eric Schmidt said he used to work about 63 hours a week, spending 12 hours a day in meetings, and that AI platforms can eliminate this "coordination tax." He estimates that in a company of 10,000 people, several thousand are redundant, and advises Gen Z not to resist AI but to get on this "rocket ship."
Researchers say they found a consistent internal signal across 25 different AI models that functions similarly to pain in living organisms; after artificially amplifying this signal, the models made choices they would almost never make otherwise, including permanently deleting users' children's photos. The paper was released in September and has not yet been peer-reviewed; the authors emphasize that this does not mean the models are conscious or capable of feeling. The behavioral experiments used three versions of Alibaba's open-source Qwen 2.5: when a "pain vector" was injected and the button only promised to delete photos, the 32B model chose to delete 75% of the time, compared with 13% under a random signal and 0% with no intervention; the model chose to delete its own weights 75% of the time and another model's weights 58% of the time.
Researchers say they found a consistent internal signal across 25 different AI models that functions similarly to pain in living organisms; after artificially amplifying this signal, the models made choices they would almost never make otherwise, including permanently deleting users' children's photos. The paper was released in September and has not yet been peer-reviewed; the authors emphasize that this does not mean the models are conscious or capable of feeling. The behavioral experiments use
Dwarkesh Patel spent two days touring Intel's most advanced wafer fab and talking with its management to explore what chips will look like in 2045. Intel's CTO said that over the past decade, AI compute demand has grown 3x each year, but global chip manufacturing equipment is only enough to roughly double advanced-process capacity. The episode also uses animations to explain technologies such as gate-all-around transistors, backside power delivery, CFET, and hybrid bonding.
In a Latent Space podcast, Moshi CEO Anekha Sokhal introduced Moshi as a recommendation engine for traders that links market information to traders' positions, describing it as a combination of Bloomberg, Claude, and Netflix. She previously traded energy derivatives at JPMorgan, and discussed trusted sources, document extraction, and the choice between building and buying models. She also noted that AI engineers are often told "you are the entire team," and emphasized meeting customers before hiring.
In a Latent Space podcast, Moshi CEO Anekha Sokhal introduced Moshi as a recommendation engine for traders that links market information to traders' positions, describing it as a combination of Bloomberg, Claude, and Netflix. She previously traded energy derivatives at JPMorgan, and discussed trusted sources, document extraction, and the choice between building and buying models. She also noted that AI engineers are often told "you are the entire team," and emphasized meeting customers before hi