SunBrief#71: GPT-5.4 Pushes AI Into Work Mode

Welcome to the SunBrief

Today in SunBrief 🌞

• A team of 8 just beat OpenAI, Gemini, Perplexity, & Anthropic

• OpenAI Launches GPT-5.4 for Professional Work

• Meta sued over AI smart glasses privacy concerns

• Stock Updates

• Netflix Acquires Ben Affleck’s AI Filmmaking Startup InterPositive

• AI Highlights of the Week

• Too Important to Miss

A team of 8 just beat OpenAI, Gemini, Perplexity, & Anthropic

Spine Swarms just hit #1 on GAIA Level 3 and DeepMind's DeepSearchQA — the world's hardest benchmarks for real-world AI performance.

Not with a bigger model. With better architecture.

Spine Swarms is the AI workspace where expert agents work in parallel on a live canvas — each one researching, synthesizing, and producing deliverables you can actually use.

See exactly where every insight comes from. Steer agents in real time. 300+ models, with the right one selected for every task automatically.

OpenAI Launches GPT-5.4 for Professional Work

New frontier model boosts reasoning, coding, tool use, and computer-control agents across ChatGPT, API, and Codex

OpenAI has released GPT-5.4 (as GPT-5.4 Thinking in ChatGPT) alongside GPT-5.4 Pro, positioning it as its most capable and token-efficient model for real professional workloads like coding, research, spreadsheets, presentations, and multi-tool workflows.

Key Points:

• Two tiers: GPT-5.4 Thinking (default pro-work model) and GPT-5.4 Pro (max performance for complex tasks).

• Best-of-both upgrade: Combines GPT-5.3-Codex-level coding with stronger tool use, documents, spreadsheets, and presentations.

• Upfront planning in ChatGPT: Thinking can show a plan before finishing, so you can steer mid-response with fewer follow-ups.

• Native computer use: First general-purpose OpenAI model with state-of-the-art computer-use capabilities for agents operating across apps and websites.

• 1M context support: Up to 1M tokens in API/Codex (experimental in Codex), designed for long-horizon agents and big workflows.

• Stronger benchmarks: 83.0% on GDPval (wins/ties), 57.7% on SWE-Bench Pro, 75.0% on OSWorld-Verified, plus higher tool and web-browsing performance.

• More factual + more efficient: Lower error rates vs GPT-5.2 and uses fewer tokens to solve many tasks, improving speed/cost.

• Rollout + pricing: Available now in ChatGPT, API, and Codex; GPT-5.2 Thinking stays in Legacy Models until June 5, 2026.

Why It Matters:
GPT-5.4 is a push toward “work-ready” AI: faster agents, stronger tool orchestration, and better outputs for the stuff teams actually ship code, research, spreadsheets, decks, and end-to-end workflows.

Meta sued over AI smart glasses privacy concerns

Lawsuit alleges overseas contractors reviewed sensitive footage despite “built for your privacy” marketing

Meta is facing a new U.S. class action lawsuit after reports said workers at a Kenya-based subcontractor reviewed user footage from Meta’s AI smart glasses, including highly sensitive content. Plaintiffs argue Meta’s privacy-forward ads misled customers into believing their captured moments wouldn’t be subject to human review.

Key Points:

• Human review of private footage: Reports said contractor reviewers saw content such as nudity, sex, and bathroom use from user recordings.

• Privacy promises challenged: The suit claims Meta marketed the glasses as “designed for privacy” and “controlled by you,” without clear warnings that human reviewers could see shared content.

• Opt-out concern: The complaint argues users can’t opt out of the review pipeline once content is shared with Meta AI.

• Regulatory attention: The U.K. privacy regulator (ICO) has reportedly opened an investigation following the earlier reporting.

• Who’s named: The lawsuit targets Meta and glasses partner Luxottica of America, alleging privacy-law and consumer-protection violations.

• Meta’s position: Meta says media stays on-device unless users choose to share it, and that when users share content with Meta AI, contractors may review filtered data to improve the product.

Why It Matters:
This case puts “AI wearables” under a harsher privacy spotlight: if human review is involved, companies may need much clearer disclosures, real opt-outs, and stronger safeguards, especially as smart glasses move into the mainstream.

Stock Updates

Netflix Acquires Ben Affleck’s AI Filmmaking Startup InterPositive

Streamer buys the 16-person team and positions the tech as creator-first post-production tools

Netflix has acquired InterPositive, a stealth AI startup founded by Ben Affleck, bringing its full team into Netflix and naming Affleck a senior adviser. Netflix says the tools are designed to support filmmakers in post-production—without turning into a “text-to-movie” generator or replacing human creative roles.

Key Points:

• Acquisition + team move: Netflix bought InterPositive and is hiring its entire 16-person team of engineers, researchers, and creatives.

• Affleck’s role: Affleck will advise Netflix as the company integrates the tech and works with creators.

• Not “generate a movie” AI: InterPositive isn’t a Sora-style video generator; it builds models from a production’s existing dailies and supports post workflows.

• Postproduction upgrades: The tools aim to help with tasks like color and relighting, mixing, and adding visual effects while keeping filmmakers in control.

• No commercial product plan: Netflix says it will offer access to creative partners, but won’t sell the tech broadly in the market.

Why It Matters:
Netflix is signaling a “creator tools” strategy for AI: using machine learning to enhance real productions (dailies → post) rather than automating filmmaking. If it works, it could raise the quality and flexibility of postproduction while also setting a template for how big studios adopt AI without sparking a full “replace artists” backlash.

AI Highlights of the Week

• OpenAI Launches ChatGPT for Excel with Financial Data Integrations

  OpenAI has introduced ChatGPT for Excel in beta, letting users build, update, and analyze spreadsheets faster directly inside Excel.

  It also added financial data integrations like Factiva, Moody’s, MSCI, and FactSet to help teams work with trusted data more easily.

• Luma Launches Creative AI Agents with Unified Intelligence

  Luma has launched Luma Agents, a new system that can handle text, image, video, and audio work across end-to-end creative projects.

  Powered by its new Unified Intelligence models, the platform aims to help brands, agencies, and studios create faster with persistent context and fewer manual prompt loops.

• Anthropic Challenges Department of War Supply Risk Designation

  Anthropic said it will challenge in court a new Department of War decision that labels the company a supply chain risk to U.S. national security.

  The company says the move has a narrow scope, most customers are unaffected, and it will keep supporting national security users during the transition.

• U.S. Weighs Sweeping New AI Chip Export Controls

  The U.S. is reportedly considering new rules that would require government approval to export AI chips to any country outside the U.S.

  If adopted, the policy could give Washington far more control over Nvidia, AMD, and global chip sales while raising concerns about U.S. competitiveness.

Too Important to Miss

NotebookLM Launches Cinematic Video Overviews for Ultra Users

Microsoft, Google, and Amazon Keep Claude Available for Non-Defense Users

X Tests Product Recommendations Linked Directly to Posts

Anthropic’s Chip Strategy Could Give It a Big AI Advantage

Last Week’s Poll Result

• Did the administration overreact by blacklisting Anthropic?

  Yes, it sends a chilling message → 66.67%

  No, national security requires compliance → 18.52%

  Maybe, but leverage is part of negotiation → 14.81%

• Does Nano Banana 2’s “Flash speed + Pro quality” combo change how you’ll use AI images?

  Yes, this is workflow-ready → 37.50%

  Maybe, depends on reliability → 31.25%

  No, still prefer manual design → 31.25%

• Would a major inference speed jump change how you use AI daily?

  Yes, instant responses change everything → 58.82%

  Somewhat, but quality still matters → 23.53%

  No, speed is already fine → 17.65%

Feedback

Know someone who may be interested?