DDoS Defense Pioneer Launches New Startup, Secures $38 Million in a16z-Led Funding to Build AI Agent-Powered Version of Tailscale

Beating AI Express News reports that Andrej Karpathy, currently part of Anthropic’s pre-training team, has once again discussed how large models should output results to humans. He argues that as AI takes on more specific tasks, humans will spend more time understanding AI outputs, verifying the processes behind them, and making judgments. Making it faster for people to grasp what AI has accomplished will grow increasingly critical. Karpathy’s suggestions have evolved from text to video: For text, teams can draw on the ASD-STE100 standard to use shorter, less ambiguous language; for complex content, they can create diagrams directly, and further develop animated, interactive web pages. He is most bullish on customized explanatory videos, such as having AI generate a 3Blue1Brown-style animated video with voiceover on the spot for a specific problem. This aligns with Karpathy’s earlier views on AI interaction. Back in May this year, he already recommended using HTML to host complex large model outputs, summarizing the evolution of AI outputs as: plain text → Markdown → HTML, and eventually moving toward interactive videos and simulated environments. Compared to long blocks of text, images, animations, and interactive interfaces can turn complex information into far more accessible forms. As code and model capabilities grow cheaper, software that was not worth developing specifically in the past can now be built on demand. A single web page, an interactive tool, or even a short video can be tailored to serve one person, explain one problem, and be discarded after use.

Sources familiar with the matter revealed that Bank of New York Mellon (BNY) is in talks with Payward, the parent company of Kraken. The potential collaboration would cover digital assets, custody, trading, payments, wealth management, and other financial market infrastructure, with related services planned to be delivered to banks, exchanges, and asset management firms via Payward Services. (CoinDesk)

Beating AI News reports that AI2 has open-sourced AstaBrief 8B, a research report model based on Qwen3-8B that serves as the report-writing component of AI2’s research assistant Asta. Asta first retrieves relevant papers and content snippets, then passes them to AstaBrief to organize into a cited research report. Asta’s original "Thinking" mode relied primarily on Claude and used a multi-step process: it first summarized and organized retrieved results, then generated reports by chapter. AstaBrief eliminates these intermediate steps, allowing it to produce a full report in one step once given a research question and relevant paper snippets. AI2 has now integrated AstaBrief into Asta’s Fast mode. AI2’s tests show Fast mode takes an average of 51.1 seconds to generate a report, compared to 178.5 seconds for the original Claude-powered Thinking mode—an approximate 3.5x speedup. In the ScholarQA-CS2 benchmark, AstaBrief scores 90.5 in citation precision and 78.2 in citation recall, while base Qwen3-8B scores 76.2 and 64.6 respectively. The model weights are released under the Apache 2.0 license, with training data also made public. Developers can integrate AstaBrief into their own paper retrieval systems, or use official examples to process local PDFs directly. AI2 also noted that the model’s main training and evaluation were completed in 2025, with comparisons made to state-of-the-art systems at that time; no full retesting was conducted against 2026’s latest models like Claude.

Former President Trump posted on Truth Social that his administration will immediately begin distributing $100 "checks" to more than 20 million seniors to help cover their Medicare Part B premiums. He also stated that if Republicans win the midterm elections, a "$5,000 Trump bonus" will be distributed to every U.S. citizen.

According to three U.S. officials who spoke to Axios, multiple core members of U.S. President Donald Trump’s cabinet held a closed-door, hours-long meeting at Camp David on Friday to discuss next steps for a potential war with Iran and the ongoing conflict between Saudi Arabia and Yemen’s Houthi movement. The Trump administration did not publicly announce the highly unusual gathering. The last comparable meeting took place in June 2025, just days before Israel launched a war on Iran. Officials said the session was chaired by Vice President JD Vance. Attendees included Secretary of State Marco Rubio, Defense Secretary Pete Hegseth, White House Special Envoy Steve Witkoff, CIA Director John Ratcliffe, and Joint Chiefs of Staff General Dan Caine. One official noted of the Middle East crisis: “Some decisions have been made, or at least thoroughly discussed.” The White House declined to comment. Separately, on Friday afternoon, as advisors met at Camp David, reporters asked Trump about his next moves regarding Iran. Trump responded: “If I told you, you’d have a blockbuster news story. But you’ll have to wait and see. Things are going very well. Iran is not in a good position.”

Beating AI Insight News Brief: Conway Research has launched a personal AI assistant called Underdog, raising funding from investors including a16z and Khosla Ventures. The tool currently runs natively on Mac, connecting to users’ emails and calendars to assist with tasks like searching messages, organizing schedules, and handling daily work. The model operates locally, so personal data does not need to be processed by cloud-based AI, and it functions even without an internet connection. a16z has confirmed leading Conway Research’s seed round. Underdog’s ability to run on standard Macs stems from two key factors: its publicly available Woof model has only 4 billion parameters, with 4-bit weights taking up roughly 2.37GB, and the team developed the Husky inference engine, optimized specifically for Woof and Apple Silicon. The team tested the setup on an M5 Max, finding the same model outperformed MLX across 16 tasks, with a maximum speedup of 4.5 times. Founder Sigil Wen notes that daily tasks like searching emails, organizing files, and scheduling do not require calling the most powerful data center-hosted model every time. Small models, as long as they are sufficiently capable and fast, can handle a large number of tasks directly on users’ own devices. Underdog is next set to expand to iPhone, Windows, Linux, Android, and NVIDIA devices. Wen refers to this roadmap as “Underdog’s Law”: cutting-edge AI that today requires data centers to run will likely be accessible on personal devices in approximately six months. The team is also exploring use cases where agents can directly shop on users’ behalf, aiming to eventually build local AI into personal agents that retain long-term personal context and act directly on users’ behalf.