This week was one of those weeks where the AI landscape just shifted under everyone’s feet. And the hosts of The Future Live caught all of it in a single Friday session.
Matt Berman, along with guest host Hiten Shah and special guest Jonathan Siddharth (founder and CEO of Turing), broke down the biggest stories shaping AI right now. The conversation went deep, and I walked away with a completely different perspective on where the real money in AI is moving.
Here’s what you need to know.
🔥 Sora is officially gone
OpenAI shut down Sora completely. Not just the social network layer, but the video model itself. No more API access. Hiten Shah put it bluntly: if a consumer app doesn’t grow, it gets killed. Add copyright nightmares and expensive inference costs on top? Done. Matt pointed out something interesting though. They’re not even keeping the model available. That’s a full retreat, not a pivot.
📊 OpenAI vs Anthropic: two very different playbooks
The show dug into the strategic split between these two companies. OpenAI is chasing consumer reach, planning to double headcount, and running ads in ChatGPT. Problem is, those ads reportedly aren’t working for advertisers. Hiten connected the dots: if consumer ads fail and direct monetization is already maxed out with ChatGPT subscriptions, what’s left?
Anthropic, meanwhile, quietly went all-in on enterprise. And it’s working. The hosts pointed out that Anthropic has built an internal flywheel: they deploy their own AI tools internally, ship faster because of it, then feed those learnings back into the product. The result is a shipping pace that Hiten (a 20-year Silicon Valley veteran who’s invested in over a thousand companies) said he’s never seen before.
🧠 Human data isn’t going anywhere
Jonathan Siddharth shared a take that cuts against the “synthetic data will replace everything” narrative. Turing builds RL environments for enterprise workflows, and even they use tons of synthetic data. But here’s the key insight: to push frontier models forward, you still need human data. Domain experts designing prompts, verifiers, and realistic seed data. His prediction? The human data business will be alive and well for the next 20 years.
That’s a bold call, but he backed it up. Distilling from stronger models to weaker ones works for some things. Moving the frontier? That still takes humans in the loop.
⚡ The adoption gap is widening, and that might be fine
One of the most interesting threads was about the growing gap between frontier AI companies and the enterprises trying to adopt their tools. Anthropic ships updates weekly. Enterprise customers can barely digest monthly changes. Hiten raised a classic product problem: enterprise users hate interface changes. They need change management, rollback options, hand-holding.
Matt’s takeaway was practical. Adaptability is becoming the most valuable skill, for companies and individuals. The gap will keep widening. The companies that learn to consume and deploy AI updates faster will win.
🛠 Pro tips from the conversation
- Enterprise AI adoption isn’t about having the best model. It’s about deployment speed and internal workflows that create compounding advantages.
- Synthetic data is useful for distillation but won’t replace domain-expert-generated training data for frontier work anytime soon.
- Consumer AI monetization is harder than it looks. Even OpenAI’s ChatGPT ad experiment stumbled.
- Adaptability beats raw technical skill. Teach yourself (and your team) to embrace rapid change.
The full episode goes way deeper into Turing’s RL environment approach, the IPO positioning game between OpenAI and Anthropic, and a fun segment about Opus 4.6 vs GPT 5.4 preferences. Definitely worth watching the whole thing if you want the complete picture.