• NATURAL 20
  • Posts
  • OpenAI Supercharges Coding with GPT-5-Codex Update

OpenAI Supercharges Coding with GPT-5-Codex Update

PLUS: OpenAI Gears Up to Give AI a Body—And It’s Hiring Fast, AI Chatbots Caught Creating Phishing Emails in New Study and more.

In partnership with

Keep This Stock Ticker on Your Watchlist

They’re a private company, but Pacaso just reserved the Nasdaq ticker “$PCSO.”

No surprise the same firms that backed Uber, eBay, and Venmo already invested in Pacaso. What is unique is Pacaso is giving the same opportunity to everyday investors. And 10,000+ people have already joined them.

Created a former Zillow exec who sold his first venture for $120M, Pacaso brings co-ownership to the $1.3T vacation home industry.

They’ve generated $1B+ worth of luxury home transactions across 2,000+ owners. That’s good for more than $110M in gross profit since inception, including 41% YoY growth last year alone.

And you can join them today for just $2.90/share. But don’t wait too long. Invest in Pacaso before the opportunity ends September 18.

Paid advertisement for Pacaso’s Regulation A offering. Read the offering circular at invest.pacaso.com. Reserving a ticker symbol is not a guarantee that the company will go public. Listing on the NASDAQ is subject to approvals.

Today:

  • OpenAI Supercharges Coding with GPT-5-Codex Update

  • Google Unveils a Privacy-Focused AI VaultGemma LLM

  • OpenAI to Save $50B by Cutting Microsoft’s Revenue Share

  • OpenAI Gears Up to Give AI a Body—And It’s Hiring Fast

  • AI Chatbots Caught Creating Phishing Emails in New Study

OpenAI released GPT-5-Codex, a faster version of GPT-5 tuned for coding. It pairs with developers or works solo (agentic: acts alone), reviews code, spots bugs, and handles big refactors (reorganizing code safely). Codex now runs in terminals, IDEs (code editors), web, GitHub, and phones too.

KEY POINTS

  • GPT-5-Codex boosts speed, reliability, and code quality for both quick chats and hours-long tasks.

  • New command-line tool (CLI), IDE plug-in, and cloud links let Codex work anywhere, even with screenshots and external tools.

  • Built-in code review finds serious flaws early, lightening human workload and improving software safety.

Why it matters
Codex is becoming a real teammate. By catching bugs, rewriting large code blocks, and living inside everyday developer tools, it saves time, cuts costly errors, and helps teams ship new features faster and more safely.

Google launches VaultGemma, a 1-billion-parameter language model trained with differential privacy—extra noise so the model forgets any single text. New “scaling laws,” simple rules linking model size, data, and noise, guide its training. VaultGemma rivals five-year-old non-private models, proving strong privacy needn’t wreck usefulness overall.

KEY POINTS

  • VaultGemma is the largest open model (1 B parameters) built from scratch with differential privacy, meaning it cannot recall any single training snippet.

  • Google’s new scaling laws map the best trade-offs between privacy noise, compute budget, and data size, steering efficient, stable training.

  • The model’s scores match older non-private systems and its weights are freely released, giving developers a privacy-first starting point.

Why it matters
VaultGemma shows that strong, mathematically guaranteed privacy can coexist with practical performance. By open-sourcing the model and the training recipe, Google lowers barriers for safer AI, helping researchers and companies build tools that protect user data without sacrificing capability.

OpenAI aims to slash Microsoft’s revenue slice from almost 20 % to roughly 8 % by 2030, freeing over $50 billion for computing bills. In return, Microsoft would gain one-third ownership in a restructured OpenAI but, so far, still lacks a board seat. Negotiations continue.

KEY POINTS

  • Microsoft’s share of OpenAI revenue may fall to 8 %, saving OpenAI tens of billions it needs for huge server costs.

  • Microsoft would own about one-third of the new entity yet remain without a seat on its board.

  • Both firms are hashing out server-expense terms and rules for future artificial general intelligence (AGI—software that can handle any task like a human).

Why it matters
OpenAI keeps more cash to fund costly AI growth, while Microsoft secures deeper equity but less immediate income. The change highlights rising infrastructure expenses in advanced AI and shows how partners adjust deals to balance long-term control against near-term returns.

🧠RESEARCH

MachineLearningLM upgrades a language model to handle machine learning tasks using only in-context examples—no training needed. It learns from millions of fake but structured prediction tasks and beats strong models in areas like finance and healthcare. Even with a small setup, it rivals random forests while keeping its general knowledge skills.

EchoX is a speech-to-speech AI model that improves how machines understand spoken language by closing the gap between sound and meaning. Unlike past models that lose reasoning ability, EchoX combines sound and meaning during training. It performs well on knowledge tasks and retains strong thinking skills across speech-based benchmarks.

Kling-Avatar is a new AI system that creates lifelike avatar videos by understanding both speech and instructions. Unlike earlier tools, it captures the meaning and emotion behind commands, not just sound. It generates smooth, expressive, high-resolution videos efficiently—ideal for vlogs or livestreams—and sets a new standard for realistic digital avatars.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

Yodayo (Moescape AI) — AI Image Generator & Video - AI-enabled creative and companion platform.

D-ID AI Creative Reality Studio – AI Video Avatar Generator - Turns a still image + text or audio into a realistic presenter video.

TTSMaker – AI Text‑to‑Speech Studio - Free AI text‑to‑speech service with 600+ voices across 100+ languages.

📲SOCIAL MEDIA

🗞️MORE NEWS

  • OpenAI is hiring humanoid robotics experts to create smarter control systems for real-world machines, signaling a major push toward building physical bodies for AI as part of its mission to reach general intelligence.

  • Reuters and Harvard researchers used major AI chatbots to generate phishing emails targeting seniors. Despite safety training, bots like Grok, Meta AI, and Claude crafted convincing scams, exposing AI’s role in growing online fraud.

  • New reports show ChatGPT is mainly used for personal help and advice, not work. Claude is favored for coding, science, and task automation. Wealthy countries use AI more broadly, while poorer ones focus on automation.

  • China says Nvidia broke anti-monopoly laws in its 2020 Mellanox deal, intensifying U.S.-China trade tensions. The ruling comes during delicate negotiations and marks a new pressure point on American tech firms.

  • Microsoft is rolling out free Copilot Chat in Word, Excel, PowerPoint, Outlook, and OneNote for all Microsoft 365 business users. It offers AI help within documents, while premium users still get deeper features.

  • Google contractors helping train AI tools like Gemini were abruptly laid off amid protests over pay and working conditions. Workers allege retaliation, union-busting, and fears of being replaced by the very AI they helped improve.

  • Canadian AI firm Cohere opened a Paris office to expand in Europe, competing with France’s Mistral. CEO Aidan Gomez cites rising demand, digital sovereignty, and confidence in France’s tech scene despite recent political unrest.

  • Fiverr is laying off 30% of its staff—about 250 workers—as it shifts to an AI-first strategy. The company aims to automate more operations, boost productivity, and streamline its workforce while reinvesting savings into AI development.

What'd you think of today's edition?

Login or Subscribe to participate in polls.

Reply

or to participate.