Destination Superintelligence

Good morning AI enthousiasts,

2025 has barely started, and already the hypetrain is barreling forward like a bullet.

A few months ago, I detailed in this newsletter that there’s two different schools of thought.

1. AI is going to evolve into AGI and then ASI and life as we know it will never be the same

2. AI is a bubble, and all these tech companies will go bust once the money runs out.

With the release of o3 only a few weeks ago, we have come very close to AGI. And depending on your definition of AGI, we might already be there.

If these advancements continue, we might be in for a wild year, and school of thought number 2’s bubble will pop.

In the meantime, the usual suspects from the different AI development firms are firing up the hype train again.

Sam Altman, in particular, has now said that AGI is going to be established no matter what, and he and his company are looking at Superintelligence going forward.

Notice the relatively low view count on this post. But here’s Sam Altman hinting not at AGI but at the thing beyond: superintelligence, or ASI.

In a blog titled Reflections, Sam Altman declared OpenAI is now confident it knows how to build AGI, predicting AI agents will join the workforce by 2025. OpenAI’s focus is shifting toward superintelligence, which Altman believes could transform industries and drive global prosperity.

The cynic would suggest he needs some more money.

Hinting at superintelligence seems reckless in early 2025. Why is he skipping these steps. What do they know that we don’t ?

Ilya Sutskever, the AI scientist that has left OpenAI last year to found his own company, SSI (Safe Super Intelligence) is doing a straight shot a superintelligence in his new venture.

He’s not alone. Here’s a tweet by Logan Kilpatrick, product lead AI at Google.

To be honest, I’m a bit confused. We don’t have proper AGI yet, and these people are talking about the next thing already.

Here’s another AI researcher from OpenAI. “Short timelines”, indicating we’re on the verge of a breakthrough.

Remember, 2025 is going to be the year of the agents. Where are our agents?
According to the same researcher, these agents are going to be crazy. For example, an agent that is good at making money online:

Elo refers to the scoring system used in chess. The best player in the world - Magnus Carlsen is 2882 ELO and nobody ever went higher.

Don’t get me wrong- you know my stance on this : a benign superintelligent being that helps us overcome our challenges would be most welcome.

Once again, it’s difficult to discern between hype and reality.

But whatever happens, buckle up because 2025 is gearing up to be quite the ride.

Welcome to the Blacklynx Brief!

Like newsletters?

Here are some newsletters our readers also enjoy:

AI News

• Samsung is investing $181M to acquire a controlling stake in Rainbow Robotics, advancing its humanoid robotics ambitions under a new Future Robotics division led by Dr. Jun-Ho Oh. This partnership combines Samsung’s AI technology with Rainbow’s expertise, such as its groundbreaking Hubo robot, with plans to integrate robotics into Samsung’s manufacturing facilities and consumer offerings.

• ByteDance unveiled 1.58-bit FLUX, a revolutionary compression technique for AI models that reduces storage by 8x and memory use by 5x without sacrificing output quality. By representing data as just three values (positive, negative, or zero) and relying on self-supervision, FLUX enables faster image generation at lower computational costs.

• Rubik’s AI launched Sonus-1, a model family with four versions tailored to speed, everyday use, complex tasks, and advanced reasoning. The Reasoning model achieved 97% on GSM-8k and 91.8% on advanced math tests, while Pro with Reasoning scored 90.15% on MMLU, rivaling top-tier models.

• Sam Altman sparked speculation with a cryptic tweet suggesting humanity might be nearing or past the technological singularity, the point where AI growth becomes uncontrollable. The tweet coincides with the release of OpenAI’s o3 model, which set new performance benchmarks, and a colleague’s comment about knowing how to create superintelligence.

• Microsoft announced an $80B investment in AI data centers, over half of which will be U.S.-based, as part of its strategy for American leadership in AI. The company’s three-pronged plan includes advancing AI tech, expanding workforce AI skills, and advocating for balanced regulation.

• NeuroXess reported advancements in brain-computer interface (BCI) tech, including 71% accuracy in translating Chinese thoughts into text and real-time device control through brain signals. Patients used BCIs to operate smartphones, robotic arms, and digital avatars, marking the first "mind-to-AI large model" interaction.

• The statement comes amid growing internal confidence at AI labs, potentially signaling a faster-than-expected timeline for groundbreaking advancements.

• Samsung introduced its “AI for All” initiative, embedding AI across its products, including TVs with real-time translation and content summaries, Galaxy Book5 AI PCs, and SmartThings-powered appliances. The company announced Microsoft Copilot integration for TVs and hinted at a potential partnership with Google. With AI infiltrating every corner of its ecosystem, Samsung is positioning itself as a leader in the AI-powered home of the future.

• A Harvard study revealed that AI systems can execute fully automated phishing campaigns with 54% success rates, rivaling human experts and far exceeding traditional spam. AI systems like Claude 3.5 and GPT-4o automated tasks like profiling and email creation, reducing costs by 50x. The findings highlight the urgent need for better defenses as AI-powered scams become more effective and scalable than ever before.

• Nvidia CEO Jensen Huang revealed the RTX Blackwell GPU family, including the $2,000 5090 chip with 2x the performance of its predecessor, and Project Digits, a $3,000 supercomputer-level PC powered by the GB10 Superchip. The company also introduced Cosmos, an open platform of world models for robotics, and Nemotron AI models for agentic applications, alongside a major partnership with Toyota for autonomous vehicles. Nvidia’s focus on agentic AI signals its preparation for the next wave of AI-driven innovation.

• Panasonic announced ‘Panasonic Go’, its AI-first transformation plan, aiming to generate 30% of revenue from AI by 2035. The company is launching Umi, an AI wellness coach built on Anthropic’s Claude, set to debut in the U.S. in 2025 with integrations for wellness services like Calm and Blue Apron. Internally, Panasonic will use Claude to modernize its operations, marking a bold shift for the century-old electronics company.

• A German study involving over 460,000 screenings found AI-assisted radiologists detected 17.6% more breast cancers than traditional methods, with improved biopsy accuracy and 43% faster reading times. The study highlights AI’s transformative role in healthcare, enabling faster, more accurate diagnoses and better patient outcomes. AI is becoming an essential tool in modern medicine, improving both efficiency and care quality.

• Omi, an AI wearable priced at $89, offers always-on listening for tasks like meeting notes and real-time summaries while introducing early brain-interface capabilities to detect user intent. Designed as a necklace or temple attachment, it boasts a 3-day battery life and integrates with open-source apps and AI models from OpenAI and Meta. While promising, privacy concerns and adoption challenges could hinder its path to success, especially as the company envisions thought-reading features within two years.

• TransPixar, a new AI system by Adobe and HKUST, generates transparent visual effects like smoke, reflections, and portals for videos, blending them seamlessly into scenes. The model achieves realistic effects with minimal additional training data, enabling applications in movies, gaming, and beyond. This breakthrough could redefine visual effects production, as AI increasingly handles complex creative tasks.

• Microsoft’s Phi-4, a 14B parameter language model known for outperforming larger competitors in math and reasoning, is now fully open-source on Hugging Face under an MIT license. Trained on synthetic, high-quality data, Phi-4 allows developers to build commercial applications with its weights freely available. This move reinforces the growing momentum and accessibility of powerful open-source AI models in the development community.

Quickfire News

• LG introduced its 2025 line of AI-powered gram laptops, featuring both on-device and cloud-based AI capabilities, including new Pro models powered by Intel’s next-gen processors and Microsoft’s Copilot+.

• Samsung partnered with Instacart to integrate grocery ordering into its 2025 Bespoke refrigerator screens, using "AI Vision Inside" technology to recognize and reorder food items for same-day delivery.

• X missed its end-of-2024 release target for Grok-3, with recently discovered code suggesting a possible Grok-2.5 release instead.

• AI startup Rembrand raised $23 million in a Series A funding round to expand its AI-powered virtual product placement technology from social media to connected TV, with plans for self-service and professional offerings.

• Google is reportedly using Anthropic’s Claude to benchmark and evaluate Gemini’s performance, with internal documents detailing response comparisons between the two models.

• An AI-powered humanoid robot named Captcha taught a high school class in Germany, delivering lectures and moderating debates, marking the first instance of a humanoid teaching students.

• Nvidia robotics VP Deepu Talla predicted that the "ChatGPT moment for physical AI and robotics is right around the corner," as the company announced its next-gen Jetson Thor computers for humanoid robots, set to launch in 2025.

• Meta removed its AI-generated social media character profiles after criticism of problematic chatbot responses, implementing platform-wide search restrictions in response.

• Elon Musk teased the upcoming release of Grok 3, stating that pretraining is complete and the model features 10x the compute power of Grok 2.

• Google published a whitepaper on AI agents, introducing a new architecture enabling models to autonomously use external tools and APIs while maintaining real-time decision-making control.

• Samsung unveiled Samsung Food, a new TV feature that uses AI to recognize food items on-screen and suggest recipes based on them.

• New benchmark tests showed AI agents autonomously completing 24% of real-world workplace tasks in software environments, with Claude 3.5 Sonnet excelling in admin, coding, and project management tasks.

• Former Sora lead Tim Brooks posted two job openings at Google DeepMind to expand his team focused on AI world simulation, targeting applications in visual reasoning, embodied agents, and interactive entertainment.

• Apple announced plans to update its AI-powered notification summaries for iPhones after the BBC reported multiple incidents of false news summaries, including fabricated stories about athlete wins and celebrity announcements.

• The U.S. Food and Drug Administration released its first comprehensive draft guidance for AI-enabled medical devices, covering development and deployment recommendations, with over 1,000 devices already authorized.

• OpenAI CEO Sam Altman disclosed that the company is losing money on its $200/month o1 pro subscription model due to higher-than-expected user activity.

• Google introduced an AI-powered TV system using Gemini to generate automated news summaries from online sources and YouTube content.

• HubSpot acquired Frame AI, planning to integrate its conversation intelligence technology into the Breeze AI platform to transform customer interactions into actionable data insights.

• Anthropic is reportedly close to securing a $2 billion funding round at a $60 billion valuation, a significant increase from its $18 billion valuation last year.

• OpenAI is aiming to release its "Operator" autonomous AI agent this month, after delays caused by prompt injection security concerns.

• U.S. President-elect Donald Trump announced a $20 billion AI investment from UAE real estate tycoon Hussain Sajwani to build data centers across eight U.S. states over the next several years.

• NASA highlighted its extensive AI deployments in 2024 through a new blog, showcasing use cases ranging from Mars rover navigation to environmental monitoring.

• Adobe research revealed a 1,300% increase in retail web traffic during the holidays driven by AI assistants like ChatGPT, as consumers increasingly use them for recommendations and price comparisons.

• Microsoft committed $3 billion to expand AI and cloud infrastructure in India, with plans to train 10 million people in AI technologies by 2030.

• Microsoft announced a reversion to the older DALL-E model after reports of quality issues and user complaints with the newer DALL-E PR16 released in December.

• NVIDIA unveiled its next-gen ACE autonomous game characters at CES 2025, introducing human-like AI NPCs powered by small language models to major video game titles.

• Chinese robotics firm EngineAI shared footage of its SE01 humanoid robot walking outside its office, showcasing highly human-like movement capabilities.

• Delta launched Delta Concierge, an AI-powered multimodal assistant designed to provide travelers with a more personalized and intuitive experience.

• Insilico Medicine reported positive Phase I results for ISM5411, an AI-designed drug for inflammatory bowel disease, paving the way for Phase II trials in late 2025.

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter, please pay it forward !

Thank you for being here !