AI News Flash | OpenAI's New AI Tool "Codex" Built Into ChatGPT

Hello from the world of AI! 🚀In today's newsletter, we cover the following hot topics:

• OpenAI's new AI tool "Codex," now built into ChatGPT, opens a research preview
• DeepSeek's AI writing style found strikingly similar to ChatGPT's
• Manus AI officially opens access to all users

OPEN AI

OpenAI's new AI tool "Codex," now built into ChatGPT, opens a research preview

Source: TechCrunch

On May 16, OpenAI announced a research preview of "Codex," its most advanced AI coding assistant yet, running on "codex-1," built atop the "o3" reasoning model and tuned for software dev tasks.OpenAI says codex-1 writes cleaner, more readable code, follows instructions more faithfully, and re-runs tests until they pass.Codex runs in a secure cloud sandbox and, via GitHub integration, can pre-load a repo. OpenAI says it handles simple coding tasks, bug fixes, codebase questions, and test runs in roughly 1 to 30 minutes.It also runs multiple dev tasks in parallel, and OpenAI says it doesn't restrict device or browser access while running.OpenAI has rolled out Codex to ChatGPT Pro, Enterprise, and Team subscribers, fairly open for now, though limits are coming within weeks — after which users can buy extra tokens. Plus and Edu access is expected soon too.

Source: TechCrunch

"Vibe coding" — AI-assisted programming — is booming. Google and Microsoft's CEOs say roughly 30% of code at their companies is now AI-written. Anthropic launched "Claude Code" in February, and Google beefed up "Gemini Code Assist" in April with more practical agent capabilities.AI coding startups are also booming: "Cursor" hit $300 million in annual revenue by April 2024 and is now raising funds near a $9 billion valuation. OpenAI is pushing into the space too — reportedly buying coding tool maker "Windsurf" for about $3 billion, with Codex seen as part of that push.This cements OpenAI's presence in AI-assisted software dev — Windsurf was a strategic buy — and Codex's launch shows OpenAI is building its own ecosystem of AI coding tools.Users with access can type a prompt into a ChatGPT sidebar and hit "Code" to generate or fix code, or "Ask" for codebase questions, with progress and history shown right below the prompt box.

Source: TechCrunch

At a Codex briefing, OpenAI's head of AI agent research, Josh Tobin, said the goal is an assistant that acts like engineers' "virtual coworker" — and OpenAI is already using Codex internally to speed up repetitive dev work, docs, and new features.

DEEPSEEK

DeepSeek's AI writing style found strikingly similar to ChatGPT's — sparking debate on copyright and AI ethics

New research found DeepSeek's generated text matches ChatGPT's writing style about 74.2% of the time.The study, by AI-comparison firm Copyleaks, was posted to arXiv.org on March 3. Copyleaks suggests DeepSeek may have trained on ChatGPT's outputs — findings Forbes calls meaningful for IP, AI regulation, and the field's future.

Source:Reuters

Copyleaks used a proprietary algorithm to pull a "stylistic fingerprint" from major LLMs — OpenAI, Anthropic's Claude, Google Gemini, Meta's Llama, and DeepSeek — to see which model wrote which text. Most models were clearly distinguishable, but most of DeepSeek's output was classified as OpenAI's.This suggests DeepSeek's output structure may be extremely close to ChatGPT's, likely fueling further debate over data-use transparency and legal questions in AI.Shai Nisan, Copyleaks' head of data science, who led the study, compared the method to "a handwriting expert identifying an author" — spotting the stylistic "fingerprint" that tells different generative models apart."What we found this time is genuinely fascinating and significant — DeepSeek's style is remarkably close to OpenAI's, a level of overlap we didn't see with any other model," Nisan told Forbes.The finding raises fresh questions about DeepSeek's training data and process. "We can't say for certain right now that DeepSeek used OpenAI-derived data, but it's grounds to question the transparency and ethics behind its development," Nisan said, adding: "Given how closely it resembles a leading OpenAI model, DeepSeek's architecture, training data, and design process deserve closer scrutiny."Nisan also said that if OpenAI-generated data were used for training without permission, the IP implications would be "serious," with potentially "far-reaching legal and ethical consequences" for the AI industry as a whole.Overlapping training datasets can naturally make writing styles converge over time, but Copyleaks' technology is built to detect structural stylistic matches beyond that kind of natural convergence. In other words, the overlap observed here may not be explained by shared data alone, and could stem from deeper design or algorithmic factors.Neither DeepSeek nor OpenAI has commented publicly on the matter.Last year, DeepSeek's V3 model was seen answering some questions by claiming "I am ChatGPT." Per TechCrunch testing and social-media reports, the model even claimed to have been "released by OpenAI as GPT-4 in 2023."This raised the possibility DeepSeek is using "knowledge distillation," a common technique that builds smaller, high-performing models from a larger model's outputs — often used for efficient, task-specific AI.On January 29, 2024, OpenAI told the Financial Times it found signs DeepSeek used its outputs for distillation — possibly violating OpenAI's terms and raising ethical and legal questions.DeepSeek hasn't commented on these allegations, and its lack of transparency keeps drawing criticism.

MANUS

Manus AI officially opens access to all users — no more waitlist

China's advanced AI service "Manus AI" has scrapped its waitlist, giving every user instant access — a big step toward a more open service.

Manus AI, an innovative AI service out of China

Per official information, since May 12, 2025, anyone can use Manus AI just by signing up — the pre-registration requirement from launch is gone, making it far more convenient.A daily free allotment of 300 credits lets users try out most features. These credits are valid only for the day they're issued and don't roll over.New accounts also get a welcome bonus of 1,000 credits, giving newcomers close to full access from day one.Manus AI also offers paid professional plans with more credits, extra features, and advanced options to suit needs from individuals to enterprises.

Source: manusAI

What sets Manus AI apart is its autonomy and general-purpose versatility — handling business and everyday tasks automatically, no hand-holding needed. Use cases include:

• Designing personalized travel plans — for a trip to Japan, for example, it can produce a full itinerary, detailed maps, and sightseeing guides in one go.
• Deep-diving into stock markets and financial metrics to deliver precise reports and compelling investment proposals.
• In education, generating high-quality lecture texts, presentation videos, and study materials.
• Sifting through huge datasets to find and recommend the best-fit vendor for a given need, streamlining procurement.

As a general-purpose AI, Manus AI enriches personal life and drives major efficiency gains for businesses.

Source: manusAI

Per its official site, Manus AI is positioned as "AI that connects thought and action," covering everything from daily routines to advanced work like legal/policy analysis or building mind maps from technical documents.With access restrictions lifted, individuals and small businesses alike can adopt Manus AI more freely — though anyone handling sensitive data should stay mindful of privacy and security.Some places, like Tennessee, have already restricted Manus AI over security concerns — worth trying the daily free credits first before considering a paid plan.Getting the most from Manus AI while keeping data safe will be key going forward.

In Closing

We hope today's content helps you make the most of AI technology and stay informed. If you found this newsletter useful, please feel free to share it with colleagues or friends who might be interested. We will keep curating and delivering the most notable topics so that we can continue bringing you valuable information. Look forward to seeing you in our next

news

. Thank you for your continued support.