AI News: OpenAI Absolutely Cooked This Week!
This AI news recap covers a busy week dominated by OpenAI announcements, including GPT 5.5 Instant, new real-time voice models, and Codex updates. Anthropic made headlines with a surprising SpaceX compute deal and new 'dreaming' memory features for Claude agents. The ongoing Sam Altman vs. Elon Musk trial also produced dramatic revelations, including leaked text messages from the day Altman was fired.
Summary
The video opens with OpenAI's release of GPT 5.5 Instant as the new default ChatGPT model, described as an incremental refinement rather than a groundbreaking new model. Side-by-side comparisons show it produces more concise, accurate answers — correctly solving a math problem the older model failed — and delivers more personalized responses by leveraging memory. The model is available to all ChatGPT users, including free tier, and is also integrated into Microsoft 365 Copilot.
OpenAI also unveiled three new real-time voice models: GPT Realtime 2 (reasoning-capable voice chat), GPT Realtime Translate (live translation across 70+ languages into 13 output languages), and GPT Realtime Whisper (live speech-to-text transcription). Demos showed impressive capabilities, including the ability to tell the model to stay silent during a side conversation and resume on command. These models are currently API-only but are expected to come to ChatGPT soon. A new safety feature called 'trusted contact' was also added to ChatGPT, allowing adults to nominate someone to be notified if AI detects serious self-harm discussions.
In Codex news, a new Chrome plugin was released allowing Codex to directly interact with the browser, though the host noted it was still buggy on launch day. OpenAI also added virtual 'pets' to Codex — customizable ASCII-style characters that live inside the interface — which the host found amusing given OpenAI's stated goal of eliminating side quests.
On the Anthropic front, two major stories emerged. First, Claude's managed agents now support a 'dreaming' feature — a scheduled background process that reviews past sessions, identifies patterns, restructures memories, and helps agents self-improve over time, going beyond standard memory to become more proactive. Second, Anthropic signed a compute deal with SpaceX, which the host noted was surprising given Elon Musk's past public criticism of Anthropic. The host speculated this is an 'enemy of my enemy' situation, with Musk benefiting from helping Anthropic compete against OpenAI while monetizing XAI's excess compute capacity. Anthropic also committed $200 billion to Google Cloud. These deals are enabling Anthropic to raise usage limits for Claude Code and the API.
XAI launched Grok 4.3, which showed benchmark improvement over the previous Grok model but still trails OpenAI, Anthropic, and Google's top models. Its main advantage is significantly lower cost.
The Sam Altman vs. Elon Musk trial continued to produce dramatic disclosures. Former OpenAI CTO Mira Murati testified under oath that Altman lied to her about whether a new AI model had cleared the company's deployment safety board. Additionally, explosive text messages from the day Altman was fired were revealed, showing a frantic back-and-forth between Murati and Altman, with Microsoft CEO Satya Nadella being pulled in as a mediating force.
The rapid-fire segment covered: Claude now working natively across Microsoft 365 apps (Word, Excel, Outlook, PowerPoint) with cross-platform memory; Adobe Acrobat gaining AI features including PDF chat, presentation generation, and podcast creation similar to NotebookLM; Spotify enabling users to save AI-generated personal podcasts to their library; Apple reportedly developing AirPods with built-in cameras for contextual AI awareness; and Nvidia partnering with startup SPAN to install mini data centers with 16 Blackwell GPUs in new homes, with the host speculating this could evolve into a model where homeowners rent excess compute back to AI companies.
Key Insights
- Mira Murati testified under oath that Sam Altman falsely told her OpenAI's legal department had cleared a new AI model from going through the deployment safety board — a claim Murati said was contradicted when she directly verified with General Counsel Jason Quan.
- The host argues the Anthropic-SpaceX compute deal is likely an 'enemy of my enemy is my friend' arrangement, with Elon Musk willing to help Anthropic gain ground over OpenAI while monetizing XAI's idle compute capacity, which Musk himself claimed is only 11% utilized.
- Anthropic's new 'dreaming' feature for managed agents is described as fundamentally different from standard ChatGPT/Claude memory — instead of passively remembering facts, it runs on a schedule to proactively review sessions, extract patterns, and restructure memory to improve agent performance over time.
- OpenAI's GPT Realtime 2 voice model supports a 'stay quiet' command that allows users to hold a separate conversation while the model listens passively and resumes only when told — a capability the host called the most impressive aspect of the demo.
- The host speculates that Nvidia's partnership to install 16-GPU Blackwell mini data centers in new homes could evolve into a model where homeowners rent excess compute capacity back to AI companies like Anthropic and OpenAI, analogizing it to solar panel energy grid credits or blockchain mining.
Topics
Transcript
[0:00] Here's the AI news you probably missed from this week. There was a ton of news out of OpenAI this week. Let's start with the GPT 5.5 Instant. This is OpenAI's brand new default model inside of ChatGpt. Now, if I'm being totally honest, I don't feel like most people are going to notice a massive difference with this. If you're using Chat GPT, you'll probably notice some fairly minor improvements. They claim it's smarter and more accurate with clearer, more [0:30] concise answers that feel better tailored to you. This isn't like a brand new top-of-the-line model out of OpenAI. This is more of like an incremental improvement to the previous default model. So, here's some side-by-side comparisons…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Matt Wolfe
AI News: Fable Banned, New Open-Source Leader, Midjourney Shocker
This AI news roundup covers the US government forcing Anthropic to shut down its Fable 5 and Mythos 5 models due to a security vulnerability jailbreak, the release of a competitive open-source model GLM 5.2 by ZAI, and MidJourney's surprising pivot into medical imaging technology with a new ultrasound-based body scanner.
AI News: Claude's Massive Leap & Siri Gets Good!?
This AI news roundup covers the release of Claude Fable 5 (a Mythos-tier model from Anthropic) and its controversial safety restrictions, Apple's WWDC AI announcements including a major Siri overhaul, and updates from Google including NotebookLM upgrades and real-time translation. Additional rapid-fire items include OpenAI and SpaceX IPO filings, ChatGPT email sending, and a teased Midjourney hardware device.
Shopping Online Is About To Change Forever
The video introduces 'agentic commerce,' a new AI-driven shopping paradigm where AI agents proactively match users to products before they search. The platform Glance is highlighted as a leading example, using selfies and personal data to generate personalized outfit recommendations with direct purchase links. The creator frames this as a major evolution in e-commerce beyond chat-based AI search.
Microsoft Build Recap in 82 seconds
Microsoft Build in San Francisco featured seven new in-house AI models, including a flagship reasoning model, a coding model, a transcription model, and a voice generation model. Microsoft also entered the AI agent space with Microsoft Scout, giving OpenAI direct access to Microsoft products and Windows management. The announcements signal a major push by Microsoft into competitive AI across multiple domains.
How Big Tech Lies About AI Layoffs
The transcript argues that major tech companies are using AI as a cover story for layoffs that are actually driven by post-COVID over-hiring and a desire to cut costs. Industry leaders like Nvidia's Jensen Huang and DeepMind's Demis Hassabis have publicly called out this practice as lazy and irresponsible.