Search “Speech”

TechCrunch rounds up AI agents that live in text messages

TechCrunch reviewed more than ten AI agents that require no app download and interact directly via text message, covering general-purpose assistants as well as specialized products for home, travel, content creation, and work. These agents can remember context, connect to existing apps, and complete tasks on users' behalf, such as scheduling appointments, booking restaurants, sending emails, and shopping online. The article also lists each product's launch date, funding situation, and subscription price, such as Instinct raising $1 billion at a $10 billion valuation in September 2026, Fambot securing $3.5 million in pre-seed funding, and Town completing a $55 million Series A round.

AI assistant Instinct, founded by a 23-year-old college dropout, saw its valuation rise from $2.5 billion to $10 billion in two months

AI assistant Instinct saw its valuation rise from $2.5 billion to $10 billion in less than two months, with its Series C led by Sequoia Capital, Benchmark, and Coatue, and only 14 full-time employees. Founder Noah Shinn is 23 years old, dropped out of college, published the Agent paper Reflexion at NeurIPS as first author, and later joined Sierra as a research scientist. Instinct uses authorized permissions for email, calendar, credit cards, and more to book tickets, shop, cancel subscriptions, and negotiate insurance for users via text message or phone instructions; users spend an average of more than $1,300 per month through it, but its excessive permission requests have also sparked privacy controversy.

Photon, which held a funeral for mobile apps, raises $4.5M seed round to replace apps with agents

AI startup Photon has closed a $4.5 million seed round, co-led by Gradient and A*, with participation from Vercel, Sequoia China and others. The company helps developers build agents on messaging platforms such as iMessage and WhatsApp, and has attracted more than 40,000 developers to sign up, with revenue growing 10x in four months and message volume up 5x last month. Its hosted platform offers 99.95% availability and has achieved SOC 2 Type II and HIPAA compliance certification.

Cerebras: Latency Debt Is Dragging Down AI Experience, GPUs Have Hit a Bottleneck

Cerebras proposes the concept of "latency debt": model parameters have grown from 340M in BERT-Large to over 1T in Kimi-k2, the share of inference tokens has risen from under 10% to over 50%, context has expanded from about 1,000 tokens in GPT-2 to 10 million tokens, input length has doubled every year and output has tripled, causing latency to accumulate continuously. Cerebras says its Wafer-Scale Engine (WSE) keeps data on-chip, eliminating the memory bandwidth bottleneck of GPU inference, and OpenAI has already purchased 750 MW of compute from it.

OpenAI Team Responds to Computer Use Progress Doubts, Recaps One-Week Sprint on Decisions API

In a podcast after OpenAI DevDay, Ari Weinstein, who is in charge of the Computer Use agent, responded to Dwarkesh's doubts about Computer Use progressing slowly, saying it is now completely different from a few months ago because the model has become better at debugging and retrying, and can write JavaScript to execute multiple steps at once while combining screenshots, accessibility, DOM, and Playwright. He also mentioned that Dots provides each Dot with a cloud Linux virtual computer, and that Computer Use has entered the Agents API. Nikunj Handa, head of the OpenAI API team, introduced updates to GPT-6 such as asynchronous tool calling, mid-turn steering, WebSockets, UltraFast inference, prompt caching and pre-warming, and compaction, and said the Decisions API is based on Luna weights, with no new model trained, and took only about a week from the emergence of the idea to a prototype.

English translation is being prepared.

OpenAI and Meta Bet on AI Agents First, Hardware Later

The Verge reports that OpenAI and Meta plan to use cute software agents in the coming year to pave the way for physical AI hardware. OpenAI unveiled its agent platform Dots at DevDay and is partnering with former Apple designer Jony Ive; public documents show the hardware could ship as early as February 2027. Meta, meanwhile, has tied its recently released Muse agent directly to a pendant-style device called Muse Charm, which it plans to launch before this year's holiday shopping season. Both companies hope to first get users accustomed to and dependent on agents, then guide them toward accepting physical forms.

Frequent rear-end crashes involving assisted driving: system disengages at the last second, but liability still falls on the driver

On the Anshun section of the Shanghai-Kunming Expressway, a vehicle with assisted driving enabled rear-ended a large truck at night; when the system signaled disengagement, it was less than 10 meters from the vehicle ahead, leaving the driver no time to take over. The article reviews similar cases in Binzhou, Shandong, and on the Jiqing Expressway, noting that all mass-produced vehicles on the market are legally L2-level assisted driving, that the Supreme People's Court has made clear the driver remains the responsible party after activating the system, and that automakers need only prove they issued a warning and disengaged to be exempt from liability. Two mandatory national standards for intelligent connected vehicles taking effect in 2027 require L3 systems to give drivers at least 10 seconds to respond after issuing a takeover request, otherwise triggering a deceleration and pull-over to a stop.

Anthropic's Thariq on the Next Phase of Claude Code: Claude Mods, Projects, and Agent Security

Anthropic's Thariq Shihipar introduced the next phase of Claude Code on the Latent Space podcast, including Claude Mods for customizable harnesses, Projects, Claude Tag, and recent releases such as Opus 5.5 and Sonnet 5.5. He calls prompting the core skill for using Claude Code, recommending that users first spend time filling in context and allocate effort by task—for example, high or maximum for code review and security, and low or medium for UI-type tasks. He also discussed agent security, including sandboxing, prompt injection, Auto Mode, and Anthropic's "Pacing the Frontier" position.

Generating Images and Videos with vLLM-Omni on SageMaker AI (Part 2)

On Amazon SageMaker AI, deploy two endpoints using the same AWS vLLM-Omni DLC image: a real-time endpoint running FLUX.2-klein-4B to generate images, and an asynchronous endpoint running Wan2.1-VACE-1.3B for image-conditioned video generation. A text prompt first generates a PNG, which is then sent along with a motion prompt to the video endpoint; the generated MP4 is retrieved from Amazon S3, and the example also provides a command-line workflow and an optional Streamlit interface.

AI "Takes Over" the Digital Trade Expo: Wang Xingxing Discusses Millimeter-Level Errors in Robots, Anheng Information Demonstrates "Using AI to Counter AI"

The 5th Global Digital Trade Expo was held in Hangzhou. Wang Xingxing, founder of Unitree Robotics, stated that fine manipulation by robots involves errors of a few millimeters that cannot be automatically corrected, and that whoever solves the millimeter-level error problem of alignment in the physical world will essentially solve the core challenge of robotics. Anheng Information exhibited "Hengnao," China's first AI-native code audit agent, and "Lieying," the world's first AI-native honeypot capable of capturing autonomous attacks by AI Agents, turning "using AI to counter AI" into deliverable products.

English translation is being prepared.

How to Transcribe Zoom Meetings: AI and Native Methods

ElevenLabs has released a guide introducing how to transcribe meeting audio using Zoom's built-in tools and third-party solutions. Zoom live captions are free, but cloud storage and downloadable transcripts require a paid Workplace account; Zoom AI Companion supports live transcription in 46 languages, while post-meeting transcription supports only 19 languages. Developers can build low-latency real-time transcription pipelines through ElevenAPI WebSockets.

English translation is being prepared.

An agentic conversational video intelligence solution built on AWS

AWS has released an agentic video intelligence solution based on the Strands Agents SDK, in which a single AI agent orchestrates Amazon Bedrock, Amazon Rekognition, and Amazon Transcribe at runtime, allowing users to query video content by asking questions in natural language. Responses for already-analyzed content are returned within 1 second, while the first analysis of a new video takes 5–10 minutes; one media and entertainment company used it to reduce the manual review time for more than 200 multi-hour recordings by about 80%.

What is Human in the Loop (HITL)? Human-AI collaborative AI workflows

Human in the Loop (HITL) is an AI workflow that allocates tasks to AI and humans according to preset boundaries: AI handles predictable, rule-based tasks, while humans handle complex decisions, compliance requirements, and edge cases. It is used both for review, approval, and handoff to humans in real-time business processes, and for data labeling, output evaluation, and error correction in machine learning development. ElevenLabs' ElevenAgents can be used to build such HITL workflows.

TypeSafe CEO Diogo Almeida on Jev: Building a System One Model for Production, Not for God

InstructGPT paper author and TypeSafe CEO Diogo Almeida detailed his new model Jev on the Latent Space podcast, describing it as a System One large programmatic model oriented toward code consumption, with its core innovation being an unpublished RLCD (reinforcement learning for calibrated decision-making), optimized for "intelligence per dollar" rather than human preference feedback. He criticized RLHF for causing mode collapse, sycophancy, and hallucinations, arguing that safety alignment suits products like ChatGPT but fragments intelligence at the API layer, and said TypeSafe would rather be seen as a data lab, refusing to publish benchmark tests. He revealed that Jev's daily token consumption has exceeded one trillion since launch, and previewed follow-up models such as ReasoningJev.

What are agentic workflows? A practical guide to automation

Agentic workflows let AI agents reason at each step and decide the next action themselves, rather than executing along a fixed script. They consist of an LLM reasoning layer combined with RAG, memory, and guardrails; high-risk steps still go through fixed approvals, and anything beyond their authority is escalated to a human. ElevenLabs' ElevenAgents Workflows editor can explicitly map decision points, routing between specialized sub-agents or human agents.

English translation is being prepared.

Why I still don't buy into true RSI

The Interconnects author states that he still does not buy into true recursive self-improvement (RSI), and leans more toward the "lossy self-improvement" view he has proposed: the research that can be automated is too narrow, the returns from parallel AI agents diminish, and resource bottlenecks and political factors constrain the building of strong LLMs. He points out that current AI safety concerns stem more from the cultural anxiety brought about by the large-scale deployment of agents by frontier labs such as OpenAI and Anthropic, rather than from any publicly disclosed breakthrough.