• AI Valley
  • Posts
  • DeepSeek’s R2 can’t escape the lab

DeepSeek’s R2 can’t escape the lab

PLUS: LAPD eyes AI that can pinpoint your location from a photo

Together with

Howdy. It’s Barsee again.

Happy Thursday, AI family, and welcome back to AI Valley.

Today’s climb through the Valley reveals:

  • LAPD eyes AI that can pinpoint your location from a photo

  • DeepSeek’s next AI model delayed by attempt to use Huawei’s chips

  • Apple said to stage big AI comeback

  • ChatGPT’s model picker is back with 4o after GPT-5 anger

  • Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

MOVIEFLO AI

Image: MovieFlo

MovieFlo.AI is where AI video meets professional storytelling: one platform for studios, marketers, and creators who think in scenes, not prompts.

In one smooth workflow, you can go from idea to final video with built-in tools that help you cast actors, plan scenes, and generate cinematic results.

Top video models like Veo, Kling, Wan, and Runway are right at your fingertips: no setup, no guesswork.

Built by veterans from LucasFilm and ILM, MovieFlo.AI delivers professional-grade tools with consistency baked in: from actors and props to branding and animation.

From product ads to short films, this is your production studio in a browser.

🎬 Try it for free, with 10% off annual subscriptions, exclusively for AI Valley readers → https://bit.ly/MovieFloAI

*This is sponsored

THROUGH THE VALLEY

The LAPD (Los Angeles Police Department) is exploring GeoSpy, an AI tool that can identify where a photo was taken in seconds by examining soil, buildings, and other visual details, sometimes narrowing it down to a specific address. Emails show interest in its $5,000-a-year plan with 350 searches. It’s part of a broader AI wave in U.S. law enforcement and intelligence: the CIA uses its own model, Osiris, for analysis; the NSA applies AI for translation and pattern detection, and local police use predictive policing and facial recognition tools. Despite some federal limits on “rights-impacting” AI, exemptions and political momentum suggest adoption will keep growing.

DeepSeek’s upcoming R2 AI model has faced repeated delays despite initial expectations for an August 2025 launch. Designed to outperform its R1 predecessor with faster reasoning, better accuracy, and open-source access, R2 is being developed and trained on Huawei’s Ascend 910B chips instead of Nvidia’s, but training has proven difficult. Reports say DeepSeek still uses Nvidia for training while relying on Huawei for inference, and issues like low-quality domestic training data, GPU shortages, and underwhelming performance have slowed progress. CEO Liang Wenfeng remains unsatisfied, and no official release date has been announced.

Apple is reportedly developing four AI-powered smart home devices set for 2026–2027, including a tabletop robot, smart display, and AI security cameras. The robot, codenamed “Pixar Lamp,” could feature a motorized arm with a 7-inch screen that follows users, an upgraded Siri with a personality called “Bubbles,” and iPhone controls. A mid-2026 smart display will use the new “Charismatic” OS for home control, music, and browsing. AI-powered cameras will automate tasks like lighting and music. Apple is also rebuilding Siri under “Linwood” and testing Anthropic’s Claude (“Glenwood”) as a backup AI system.

OpenAI is making changes to ChatGPT after backlash over the GPT-5 launch. The popular GPT-4o model is returning for all paid users, and Altman says there will be plenty of notice if it’s ever removed. GPT-5’s weekly limit for advanced reasoning has been raised from 200 to 3,000 queries, and users can now choose “Auto,” “Fast,” or “Thinking” modes to avoid being routed to the wrong model. Altman also confirmed a personality update is coming, noting the need for more customization after many users voiced strong attachment to older models.

Anthropic will make its Claude AI model available to US government agencies for just $1, aiming to secure federal contracts and strengthen ties with public institutions. The move comes shortly after Claude, OpenAI’s ChatGPT, and Google’s Gemini were all approved as official AI vendors for the government. CEO Dario Amodei said access to secure, advanced AI is key to maintaining US leadership. The offer mirrors OpenAI’s recent deal, which also made ChatGPT Enterprise available to federal agencies for $1 per agency for the next year.

TRENDING TOOLS

  • Profound - Helps your brand appear more in AI search results and shows exactly what ChatGPT says about you

  • Gemini - Now remembers your past chats (on by default) and offers Temporary Chats that auto-delete after 72 hours

  • Conductor - Runs multiple Claude Code agents on your codebase at once (one can fix bugs, another write tests, and another add features, all in separate workspaces)

  • Autumn - Like Stripe, but built for AI startups

  • Reclaim - An AI-powered calendar for both work and personal life

  • Spielwerk - TikTok-style endless feed of mini games, all made inside the app by other users

  • Open Lovable - Open-source AI tool that turns any website URL into an editable working clone you can build on instantly

THINK PIECES / BRAIN BOOST

THE VALLEY GEMS

What’s trending on social today:

Presented by Limitless:

Your Memory: Now instantly searchable

Limitless is the wearable AI that instantly captures, transcribes, and remembers every meeting, conversation, and brilliant thought you have - securely, seamlessly, and privately. Transform how you work, live and never miss a beat again.

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.