The Next Wave – NotebookLM Just Killed After Effects? + ChatGPT vs Anthropic War

Exploding AI Model Updates: What’s New and Why It Matters

The world of AI models is charging forward faster than ever. OpenAI released two new big updates—GPT-5.3 Instant and GPT-5.4, while Google rolled out Gemini 3.1 Flash. Also mentioned were updates from Grok (Grok 4.20), Microsoft’s new Phi-4, and Qwin 3.5. The flood of new models might sound overwhelming, but focusing on the main three—GPT-5.3, GPT-5.4, and Gemini 3.1 Flash—gives the clearest view of the whole scene.

GPT-5.3 Instant stands out by cutting the “cringe” out of ChatGPT’s replies. Now, it’s less chatty, skips the extra pep talks or overly safe warnings, and jumps right into giving the answer. As Sam Altman put it, it’s “a little less cringe.”

OpenAI didn’t have to start over from scratch for this update. Training an AI model has three main steps: pre-training (scraping the internet for patterns), fine-tuning (teaching the model to be helpful), and RLHF—a phase where humans give it feedback. These quick model changes mostly focus on fine-tuning and RLHF, which makes the experience smoother and faster for all of us.

GPT-5.4 is more of a tool for developers and those building apps. The million-token context window is a huge deal—it means the model can remember a lot more information in one conversation. That helps people working with giant codebases or having long, complex chats with AI. The model’s new “computer use” agent (sometimes called agent mode) can draft emails, organize your inbox, or even click around your computer for you. The “tool search” update makes it faster and cheaper by teaching the AI how to only load the information it actually needs for the job.

For Google’s Gemini 3.1 Flash, speed and savings are the main benefits. It’s not the deepest or smartest AI but works really well for fast jobs—translation, content moderation, or simple analysis—while costing much less to use. These distilled versions copy outputs from the main model to stay quick and cost-effective.

Key Takeaways:

• GPT-5.3 Instant prioritizes natural answers that get to the point.
• GPT-5.4 is ideal for developers with large projects and “agent” tasks.
• Gemini 3.1 Flash is best for fast, inexpensive tasks like content reviewing.
• Regular users benefit from friendlier, smoother conversations.
• Developers can build bigger, smarter apps without losing context.

ChatGPT, Claude, and the Big User Shift

Over the past week, Claude, made by Anthropic, has seen a wave of new users leaving ChatGPT. This “exodus” is driven by a mix of features, ease of switching, and some drama around military contracts that made some users uneasy.

Claude launched a super-simple migration path for people moving from ChatGPT. You can bring over your memory—your past conversations and context—so you don’t have to start from scratch.

A key catalyst was the news that the Pentagon blacklisted Anthropic after disagreements on red lines in AI contracts. The government didn’t like the restrictions Anthropic wanted. Soon after, OpenAI grabbed the contract, reassuring people they kept similar rules. Still, many lost trust and started looking elsewhere.

Statistics back it up: 295% more people uninstalled ChatGPT, and Claude shot up to become the number one most downloaded app in the App Store (from previously nowhere near the top ten).

Claude also made their memory feature free (previously paid), encouraging even more users to try it. There’s a downside on the free Claude plan: rate limits hit faster compared to ChatGPT, so power users sometimes hit walls after 20 messages.

Key Takeaways:

• Claude is seeing a massive influx of new users.
• Migrating from ChatGPT to Claude is now point-and-click simple.
• Free Claude users get memory but face some usage limits.
• The Pentagon-Anthropic drama highlighted the tech/policy crossroads.
• OpenAI is still #1 with regular consumers, but Anthropic is catching up with businesses.

NotebookLM’s Massive Video Update

NotebookLM, one of our favorite AI tools, rolled out an upgrade for creating cinematic, animated videos—if you’re on the $250/month Ultra plan. Before, video generation felt like watching a boring slideshow. Now, users get something closer to a real video, complete with animation and consistent characters.

Matt demonstrated using a transcript to generate a full, multi-minute video. For example, giving it several news stories about the “Birds Aren’t Real” joke movement produced videos with animated characters and real news images (even showing the sources used). The videos aren’t Pixar-level yet, but resemble basic After Effects work and could be dropped right into a YouTube explainer. At the current pace, quality keeps jumping fast—six months from now, expect broadcast-level video.

NotebookLM also lets users ground their videos in multiple sources, which boosts accuracy. The tool uses Google’s top AI models, lets you generate images without watermarks, and will always have the freshest features first.

Key Takeaways:

• AI can generate quick, animated, custom explainer videos from just an outline or notes.
• The output keeps improving and is getting much closer to what pro editors make.
• Ultra plan costs $250/month but can replace editors or freelancers for basic animations.
• All videos and images can be managed with no visible watermark.
• The tool is evolving rapidly and is one of the main creative AI tools for personal and business use.

Hollywood, Netflix & AI: How Ben Affleck’s Company Changes the Game

Netflix made big headlines by acquiring Interpositive, Ben Affleck’s AI-powered creative company. Instead of generating movies from scratch, Interpositive trains on film dailies so editors can streamline post-production. The tool lets creators quickly clean up mistakes, remove wires from stunts, adjust lighting, enhance the backgrounds, and fix shots in ways that used to require big teams and lots of time.

Affleck stresses it’s about helping actors and directors create with less technical struggle. The tool only works with what the actors already filmed, supporting—not replacing—them. Still, as Matt pointed out, this puts many behind-the-scenes jobs—like lighting and VFX—under threat, even while “keeping actors safe.”

Hollywood has resisted some AI adoption, but pressure to cut costs and speed up production is too big to ignore. Competition from streaming means studios need more content, faster. The tech lets studios try more ideas (“more shots on goal”), adjust and test, and possibly find more hits with less wasted money.

Key Takeaways:

• Netflix is using AI to cut post-production time and costs.
• Interpositive keeps the creative focus on the actors and directors.
• Some jobs behind the scenes—editing, VFX, lighting—may dry up.
• The rise of streaming means studios need ways to churn out more, faster.
• Hollywood is starting to adopt AI, even after initial pushback.

Privacy, Meta Glasses, and Real AI in Daily Life

Meta (Facebook’s parent) found itself in privacy trouble after users discovered that videos captured with Meta smart glasses are sometimes seen by human reviewers in Kenya. Meta’s terms allow them to review your videos for AI training, but most people didn’t realize how that works. Recordings can include sensitive moments—people getting dressed, entering passwords, or showing credit cards.

Switching off the recording isn’t obvious (the power switch is hidden!), and privacy settings aren’t clear. Lawsuits from New Jersey and California now accuse Meta of false advertising about privacy. While AI needs lots of real-world data, humans reviewing sensitive personal moments raises big questions. As more tools like this roll out, knowing how your data is handled is crucial.

Key Takeaways:

• Meta smart glasses recordings could be seen by real people, not just AI.
• Sensitive daily moments might be captured without clear user consent.
• Meta faces lawsuits over privacy promises and unclear controls.
• Make sure to check, and if possible, limit privacy settings when using AI gadgets.