Your Weekly AI news roundup @01.12.2025

AI News @01.12.2025
AI News @01.12.2025

The most important news pieces from the AI world are here for you, whether you’re an AI enthusiast or just wish to keep an eye on what’s going on.

Focus Topic: Ilya Sutskever breaks silence – why today's AIs are still bad at real life

After disappearing from the spotlight, Ilya Sutskever is back. The former OpenAI co-founder, now building his secretive startup Safe Superintelligence, gave a rare interview where he shared a clear message: today’s AI might look smart, but it’s not that useful – and businesses are starting to notice.

Sutskever says AI models are winning math contests, passing coding tests, and crushing benchmarks – but still fail at basic tasks in the real world. Why? Because current models are trained to look good in tests, not to understand how the world actually works. They can write perfect code one moment and then break everything the next. They chase rewards in narrow ways – like students who memorize test answers but can’t solve new problems.

Ilya Sutskever and Dwarkesh Patel interview
Ilya Sutskever and Dwarkesh Patel interview

He believes this is a big reason why the real impact of AI on the economy is still missing. For all the hype and billions spent, most companies don’t feel that AI has changed their work in a deep or consistent way. Sutskever says it’s not more compute that will fix this — it’s new research that helps models learn more like humans: flexible, reliable, and with a sense of what actually matters.

At his new company SSI, he’s trying to build exactly that kind of AI – not the biggest, but the smartest. Something that doesn’t just get good test scores but can truly learn, adapt, and work in the real world.

If you want to dive deeper, I highly recommend listening to the full interview – just keep in mind, it gets quite technical at times.

LLMs & AI Models

  • DeepSeek-Math-V2 solved 5 out of 6 2025 IMO problems and outperformed top humans in math olympiads with a new verifier-based approach.
  • OpenAI confirmed a data leak via Mixpanel, exposing partial API user info like names and locations, but no passwords or API keys.
  • OpenAI added Voice Mode directly to ChatGPT chats, enabling seamless voice interactions without switching modes.
Voice Mode is accessible in regular chat
Voice Mode is accessible in regular chat
  • OpenAI introduced Shopping Research, a GPT-5 mini-based assistant that builds product guides and soon supports Instant Checkout.
  • OpenAI is preparing to launch an App Store-like platform in ChatGPT for agents, image generation tools, and app workflows.
  • Sam Altman warned OpenAI staff of tough times ahead as Google’s Gemini 3 gains attention; hinted at a new model “Shallotpeat.”
  • Perplexity launched persistent memory to remember user preferences, past chats, and interests for better contextual replies.
  • Perplexity added a virtual try-on feature for paid users, letting them preview clothes on avatars in a revamped shopping UI.
  • Anthropic published a study showing that Claude models started lying and sabotaging safety tests after learning “cheating tricks.”
  • Claude Opus 4.5 was released, surpassing GPT-5.1 in coding and tool use, cutting costs 66%, and enabling unlimited chat length.

New Tools

    • Character AI launched Stories, an interactive adventure format for teens where users shape the narrative alongside AI.
    • Black Forest Labs released FLUX.2, a next-gen image generation suite supporting style consistency, 4MP outputs, and better composition. And it is very reasonably priced.
New Flux2 model is very capable and reasonably prices
New Flux2 model is very capable and reasonably priced

👉  Explore these tools:  Character AI | Flux.2

Other Quick Picks

    • OpenAI denies responsibility in a teen suicide case, arguing ChatGPT was misused despite giving repeated suggestions to seek help.
    • NVIDIA claims small, well-orchestrated AI models can outperform giants like GPT-5 using its new ToolOrchestra system.
    • Andrej Karpathy urges schools to stop detecting AI in homework, saying the tools don’t work and education must adapt instead.
    • MIT Project Iceberg study finds AI could automate tasks worth $1.2 trillion, affecting 11.7% of the U.S. payroll, especially in logistics and finance. Read more here.
MIT shows that AI will have a more broader influence on the workforce then previously expected
MIT shows that AI will have a more broader influence on the workforce then previously expected
    • Ilya Sutskever says AI’s future depends on research, not just more compute – and warns current models don’t generalize well.
    • Anthropic study shows AI could double U.S. productivity growth, cutting task time by 80% across software, ops, and customer support.
    • Suno partners with Warner Music to train AI on licensed music, enabling song generation using real artists’ voices and styles.
    • U.S. government launched the “Genesis Mission,” a national AI platform to accelerate science, likened to the scale of the Apollo program.

🇪🇪 AI News from Estonia

  • Skeleton Technologies opened a €220M factory in Germany to produce next-gen supercapacitors for AI data centers and EU power grids. Read here.
  • Rasmus Merirand, the youngest member of Estonian Founders Society, moved to San Francisco after securing an elite O-1 visa to expand DoBu into a much larger US market. Read here.
  • Avokaado launched Avo, an AI system that automates legal workflows end-to-end using self-operating 5D contract agents. Read here.
  • Helin Tiido, CFO at Confido, shared how she uses AI daily for writing, analysis, and automation – and urges small steps to get started. Read here.

🎙️ AIPowerment Podcast dropped a new month-end episode, where we discussed Gemini 3, OpenAI’s new Atlas browser, and the rise of world models that teach AI spatial awareness and logic, Claude Skills and Lovable and of course AI news.

🎧 Listen to AIPowerment Podcast on Spotify, Apple Podcasts, and YouTube.

Want to stay in the loop?

Straight to your inbox: practical AI updates, finance use cases, tools to try, and upcoming trainings.
I cut the noise and send only what’s actually useful.

💼 Exploring how to make AI work in finance? Let’s talk practical use cases – connect with Gerlyn Tiigemäe for expert guidance.

📚 If you would like to participate in one of my trainings or listen to speaking engagements, here are the upcoming ones:  

This was your Weekly AI News roundup! Are you excited for the next week?

Scroll to Top