Sponsored by
Stop Losing Your Money. It's time to upgrade your trading platform.
Your current trading platform is probably letting you down
Limited assets (no international stocks, no commodities, no pre-IPO companies)
Limited ability to short
Limited access to leverage
Limited trading hours
Liquid is one of the fastest growing trading platforms, allowing users to trade stocks, commodities, FX, and more 24/7/365 from their phone and computer.
Trading on Liquid is as simple as:
Pick an asset
Pick long or short
Pick your position size and leverage
Place your trade
The best part is that Liquid markets never close. So no matter what is going on in the world, you are able to keep your portfolio positioned properly.
Today:
OpenAI Transforms Codex Into an Untethered, Autonomous Developer
Anthropic and the Gates Foundation Forge a High-Stakes Partnership for Global Good
Perplexity’s ‘Computer’ Democratizes Data Science for the Entire Enterprise
Anthropic’s Cyber-Model ‘Mythos’ Uncovers a Critical Flaw in Apple’s macOS
Microsoft Quietly Scouts New Startups to Prepare for Life Beyond OpenAI
OpenAI has dramatically expanded the capabilities of its Codex coding assistant, transitioning it from a simple autocomplete tool into a comprehensive, multi-platform autonomous agent. Available across macOS, Windows, command-line interfaces, and cloud environments, Codex can now orchestrate complex multi-agent workflows, handle routine background operations, and securely execute code directly on a developer's machine.
Important Details:
Always-On Automations: Codex now works unprompted in the background to handle tedious tasks like issue triage, alert monitoring, and continuous integration/continuous deployment (CI/CD) maintenance.
Multi-Agent Collaboration: Through built-in worktrees and cloud environments, multiple Codex agents can work in parallel across different parts of a project, condensing weeks of engineering work into days.
Advanced Windows Sandboxing: To allow Codex to safely read files and execute local commands without risking system integrity, OpenAI developed a highly customized, restricted execution sandbox specifically for Windows that prevents unauthorized network or file access.
Seamless Cloud Handoff: Developers can offload heavy, resource-intensive jobs to Codex in the cloud and later review the results directly within their local code editor.
Anthropic and the Gates Foundation have forged a four-year, $200 million partnership to deploy advanced artificial intelligence for the public good. The initiative will provide grant funding, dedicated technical support, and usage credits for the Claude AI model to tackle systemic challenges in healthcare, education, and agriculture across low- and middle-income countries.
Important Details:
Accelerating Medical Breakthroughs: The collaboration focuses heavily on utilizing Claude to analyze large datasets and screen potential drug and vaccine candidates for high-burden and neglected diseases, including polio, HPV, and preeclampsia.
Public Health Intelligence: Anthropic is building specialized connectors to help governments and health ministries seamlessly integrate AI into their decision-making processes for workforce deployment, outbreak detection, and supply chain management.
Educational Infrastructure: The partnership is co-developing tools to boost K-12 educational outcomes in the U.S., sub-Saharan Africa, and India, including AI-powered apps for foundational literacy, math tutoring, and career advising.
Empowering Agriculture: The initiative will fund the development of AI tools that deliver real-time, locally relevant agricultural data to farmers in their native languages to improve crop yields and financial planning.
Perplexity has introduced "Computer for Enterprise," a sophisticated AI system capable of executing complex, multi-step workflows across various software platforms. By acting as a digital proxy that safely connects to a company's internal tools, Perplexity Computer enables non-technical employees to perform advanced data science and analytics tasks without waiting on specialized engineering teams.
Important Details:
Direct Database Integration: Through secure app connectors, the system can directly query major enterprise platforms like Snowflake, Salesforce, and HubSpot.
Instant Analytics: A sales or marketing professional can ask natural language questions, and the AI will autonomously write the database queries, run them, structure the results, and generate comprehensive dashboards or financial models.
Intelligent Orchestration: Instead of relying on a single AI model, Perplexity Computer acts as a command center. It intelligently delegates sub-tasks to the best-suited frontier models (such as using Opus for core reasoning, Gemini for deep research, or ChatGPT for long-context recall).
Automated Data Exploration: The system features specialized "skills" for data science, allowing it to autonomously profile new datasets, assess data quality, identify outliers, and discover distribution patterns before conducting formal analysis.
🧠RESEARCH
This paper shows how to train AI models that read images and text to handle much longer inputs. The team finds question-answer practice on long documents beats copying text. Their model, MMProLong, improves document answers, keeps short-task skill, and still works reliably beyond training, reaching 256K and 512K inputs cleanly.
EVA-Bench tests voice agents, AI systems that talk with users to finish tasks. It creates realistic back-and-forth calls, checks bad simulations, then scores accuracy and user experience. Across 12 systems, none did well on both. Accents, noise, and repeated trials exposed major reliability gaps in today’s real deployed voice agents.
This paper argues AI models may not need a separate search tool to find useful evidence. Its method, INTRA, lets the model use attention, meaning its way of focusing, to pick stored facts and answer questions. It beat strong search-plus-answer systems on finding evidence and giving better final answers overall.
📲SOCIAL MEDIA
🗞️MORE NEWS
Apple’s Security Has Been Tough to Crack. Mythos Helped Find a Way In. Researchers used Anthropic’s Mythos AI to help find a way around Apple’s macOS defenses by linking two bugs together. The attack could raise a hacker’s access level, called “privilege escalation,” meaning they could reach protected parts of a Mac that should be blocked.
Microsoft Eyeing Startup Deals for Life After OpenAI Microsoft is looking at AI startup deals so it can depend less on OpenAI over time. It considered Cursor, backed away over rule-maker concerns, and is now discussing Inception, a startup using a different way to build text-generating AI.
OpenAI-Apple Partnership Frays, Setting Up Possible Legal Fight Bloomberg says OpenAI’s Apple partnership has become tense because OpenAI did not get the benefits it expected. OpenAI lawyers are reportedly exploring legal options, which could turn the two-year alliance into a court fight.
2028: Two Scenarios for Global AI Leadership Anthropic argues the U.S. and its allies must keep their AI lead over China by controlling advanced chips and closing loopholes. It also warns about “distillation,” meaning rivals copying AI model outputs to train their own systems faster and cheaper.
Microsoft Starts Canceling Claude Code Licenses Microsoft is pulling back many internal Claude Code licenses and moving developers toward GitHub Copilot CLI. CLI means a text-based command tool, and Microsoft says it wants one main coding assistant it can shape around its own tools, security needs, and costs.
Electrofluidic Muscle MIT researchers built soft artificial muscle fibers that move liquid with electricity, so robots or prosthetics can move more like real muscles. The fibers are quiet, lightweight, and do not need bulky motors or pumps, which could help future wearable exoskeletons and robotic hands.
