I spent way too long this week trying to keep up with AI news, and my feed felt like a firehose. Every tiny update gets blown up into “the biggest thing ever.” So I went looking for someone who actually filters the noise.
That someone is Matt Wolfe, the creator behind this weekly AI roundup. He records on Thursdays, ships on Fridays, and only covers what matters to the most people. I love that approach, so here’s my breakdown of what the original poster surfaced this week.
🚀 What’s New
The headline drop is Anthropic’s Claude Opus 4.8. According to the author, it’s a modest bump over 4.7. Slightly better coding, reasoning, and computer use, same pricing. The standout he flags is honesty: the model is now better at flagging uncertainty and avoiding claims it can’t back up.
Then the money news. Anthropic raised $65 billion in their Series H and got valued near $965 billion, making them the most valuable startup ever, ahead of OpenAI. The creator’s take? These numbers stopped feeling real.
🔀 The Twist
Here’s the part that surprised me. Anthropic also shipped dynamic workflows inside Claude Code. The expert explains it like this: Claude plans from your prompt, breaks the job into subtasks, and fans them out to subagents running in parallel. Some agents solve the problem, others try to refute the findings, and the run keeps iterating until the answers converge. A self-checking agent team that argues until it agrees. That’s a real shift in how coding agents work.
🛠️ The Rapid-Fire Roundup
The person who shared it packed in a ton more. A few worth your time:
- Microsoft’s MAI Image 2.5 jumped to number three on the Arena leaderboard, especially strong at text rendering and branding.
- Microsoft 365 Copilot got a redesign that pulls from your emails, files, and meetings inline. Perplexity also landed inside Word, Excel, PowerPoint, and Outlook.
- Leonardo AI added image-to-3D. The creator tested it on a wolf image and got a usable model, though faces still get a little cursed.
- ElevenLabs dropped Music V2 (trained on licensed data) and Dubbing V2 that keeps your own voice across languages.
🎬 My Favorite Find
The Gemini Omni examples blew me away. The original poster showed how one user drew a route on a Google Maps screenshot, then asked Omni to generate a first-person taxi drive along that exact path. Another creator sketched a flight path and got drone POV footage that followed it, under a bridge and all. If you make short films and need an establishing shot without a drone, this is huge.
💡 Worth Knowing
A few signals the contributor flagged:
- YouTube is moving AI labels to a more prominent spot and rolling out automatic AI detection starting in May.
- Sam Altman is walking back his jobs-apocalypse warnings, admitting AI hasn’t cut white-collar roles as fast as he feared.
- Nvidia’s Jensen Huang called the “AI made us do layoffs” excuse lazy and irresponsible.
- The Pope put out a rare official letter on AI, comparing it to nuclear weapons and saying it needs to be “disarmed,” joined by an Anthropic co-founder.
And yes, China now has a $118 pet translator claiming 95% accuracy and sub-dollar robot barbers. The creator says he’ll keep his human barber for now. Same.
✨ Pro Tip
If you only test one thing, try the Gemini Omni map-to-video trick. Upload a screenshot, draw your path, and prompt for a POV drive or drone shot. It’s the most creative use case here.
Watch the full video for the live demos and the haircut robot footage. The Omni clips alone are worth it. 🎥