Rabbithole Galore!

Good morning,

Everything is changing all of the time.

I sometimes start this newsletter saying it has been another crazy week in AI—or something to that effect. I’ve even contemplated publishing this newsletter twice weekly or even daily if it would be financially sustainable (which it is not). There is so much happening that I could fill multiple newsletters with what I found.

But something has been bothering me the last few weeks and I have to say something about it

In my research, I’m starting to see or hear things that are not normal. 

We don’t always see it or like it, but the whole world is constantly in flux.

However, I cannot help but notice that these last few weeks and months the speed at which things are changing is picking up. We seem to be accelerating towards something. And every single topic where this acceleration seems to occur is a true rabbithole where you can get lost in for days on end.

Let’s see:

• Geopolitical tensions are rising: Ukraine, Syria, Gaza, South Korea, Taiwan, etc. Powder kegs waiting to burst or already fully exploded. You might ask what this is doing in an AI newsletter, but if you look closely, the war in Ukraine and the conflict that is brewing in Taiwan is all about valuable materials used in semiconductor chips. That are in turn used to create the GPUs which power AI

• The world is polarizing. Political discussions are turning nasty because not only in the US but also in Europe politics are dividing us. We’re turning tribal again (Divide and conquer seems to be working. But who’s dividing us ?) . Is it just me or are people getting angrier?

• The weather seems to be changing - we’re in December, and I have yet to get out my winter coat. (Okay, this is anecdotal.)

• Bitcoin and cryptocurrencies are going mainstream. Governments have started buying it as a reserve currency and corporations are thinking about putting BTC on their balance sheets. BTC also surpassed 100.000$

• Say what you want about Elon Musk but his companies are creating jaw-dropping advancements: only in the last few months we had SpaceX catch a rocket (watch the video !!) , Tesla has nailed full self-driving, Neuralink has implanted a chip into someone’s brain allowing him to play chess just by thinking the next move. At the same time he’s the richest man on earth, is topping the leaderboard on Diablo IV (a video game) and streams on Twitch while claiming AGI will be here in 2026 at the latest and there’s a big chance it will kill us. While cosying up to world leaders and getting particularly cosy with Italy’s PM Georgia Meloni.
  

  

Hilarity ensued on X

• Meanwhile CERN is going to send a truck containing “antimatter” across Europe. What can go wrong?

• This week: Google announced a breakthrough in quantum computing. Read more below, but basically they created a new piece of hardware that can do a calculation in under 5 minutes that would take the most powerful’regular’ supercomputer 5 to the 25th minutes. That is longer than the existence of the universe. (And no, this quantum breakthrough will not “break” Bitcoin, you fools)

• OpenAI officially released SORA this week (more below) - the technology is amazing,were it not for the fact that it’s dumb as a rock. BTW I’m working on an article where I make a case for OpenAI imploding with 2 years. There are shady things going on at that company.

If feels we’re in a period that is incredibly important for humankind and that this era will define humanity going forward.

But when I say “not normal” - i’m referring to things that are not covered in the mainstream news outlets but also things that subjectively seems like a jump that is actually a big deal.

I’m now probably going to “alienate” (pun intended) a few people but did you know that the US Senate is having congressional hearings on UFO’s. Or as they are now called ‘UAPs’ ? There’s NOTHING about this in the regular media in Europe. In the US there is some reporting.

And yet this week - reports are coming from New Jersey that “mysterious drones” are hovering above military facilities. Same is going on in the UK above RAF bases.

Ok these might be military in nature but this declassified video from the Pentagon certainly is no drone.

WHAT IS THAT THING???!!!

Nobody knows what these are and nobody seems to freak out over this. I would love to know what that is because it’s this kind of thing that fuels my sense of unease.

Anyway … back to AI ! It’s probably nothing.

Welcome to the Blacklynx Brief!

Like newsletters?

Here are some newsletters our readers also enjoy:

AI News

• OpenAI kicked off its "12 Days of OpenAI" event by officially launching the highly anticipated o1 model and a $200/month ChatGPT Pro plan. The full o1 introduces image analysis, faster and more accurate responses, and a "Pro mode" featuring a 128k context window and enhanced reasoning for advanced tasks. While the new pricing tier is steep, it targets power users who require high-compute features.

• Microsoft rolled out Copilot Vision, enabling its AI assistant to analyze and interact with web pages in real-time within the Edge browser. The opt-in feature supports tasks like shopping, recipes, and strategy planning while maintaining user privacy through session-based data deletion. This evolution of AI browsing sets the stage for broader applications in 2025.

• Clone Robotics unveiled the Clone Alpha, a humanoid robot using water-powered artificial muscles and synthetic organs to mimic human movement. Preorders for 279 units in 2025 are open, though the robot’s practical capabilities, including tasks like cooking and cleaning, remain unproven. While intriguing, skepticism may linger until functional demos are shown.

• Google DeepMind’s gemini-exp-1206 has overtaken OpenAI to secure #1 on the Chatbot Arena leaderboard, celebrating Gemini’s one-year anniversary with major upgrades. The model stands out with its video comprehension capabilities and a massive 2M token context window, allowing it to analyze over an hour of video. Unlike competitors, Gemini is freely accessible through Google AI Studio and API, making it a standout in performance and cost.

• Meta’s new Llama 3.3 outpaces its 405B predecessor, delivering similar performance at 10x lower costs. With a 128k token context window and a price of just $0.10/million input tokens, the model is setting a benchmark for affordability. Boasting 600M monthly users, Meta is positioning Llama to dominate the AI assistant space, with Llama 4 slated for release in 2025.

• X (formerly Twitter) briefly rolled out Aurora, an upgraded image generator integrated with Grok, showing remarkable improvements in landscapes, photorealism, and still-life imagery. Aurora also appeared to have fewer restrictions, allowing the generation of copyrighted characters and public figures. Elon Musk teased it as a "beta version," signaling more advancements ahead as Grok 3 prepares to launch. While the model’s brief appearance raised eyebrows, its unrestricted outputs hint at a unique direction for xAI.

• OpenAI has officially released Sora, its long-awaited AI video generation platform, now available for ChatGPT Plus and Pro users. With tools like Remix, Storyboard, and Style presets, Sora makes editing and creating short videos highly accessible. The platform supports 20-second clips, higher resolutions for Pro users, and organizes generated videos alongside inspiration from other creators. OpenAI has implemented strict content restrictions, barring the EU and UK from the initial rollout due to regulatory concerns.

• Google’s new quantum chip Willow hits a major milestone, demonstrating exponential error reduction with 105 qubits and tackling computations that would take supercomputers 10^25 years in just minutes. Manufactured at Google’s quantum fab facility, Willow doubles the quantum state reliability of prior designs.

• Inspired by tadpole neurons, MovieNet processes videos more like a human brain, identifying dynamic scenes with 82.3% accuracy — outperforming both Google’s GoogLeNet and human analysts. The model is significantly more efficient, requiring less data and power, and shows potential for applications like medical diagnostics.

• OpenAI has rolled out its Canvas interface to all users, expanding the collaborative writing and coding platform's capabilities. Canvas features a split-screen layout integrating with GPT-4o, inline revision tools, and new Python execution for live debugging and code output. The update also includes Custom GPT support, allowing Canvas functionality in personalized assistants.

• Cognition Labs introduced Devin, a $500/month AI developer assistant designed for team workflows through Slack, GitHub, and IDEs. Devin automates tasks like bug fixes, PR creation, and code refactoring, enabling teams to offload repetitive work. Early demos showcased Devin’s potential to autonomously manage and resolve issues in real-world projects, signaling its role in the future of collaborative development.

• Replit removed its Agent from early access and launched a new Assistant tool aimed at project improvements and fixes. The tools now support image and URL inputs, seamless integration with Replit’s infrastructure, and React-based designs for polished outputs. With a subscription model offering unlimited usage and transparent billing, Replit continues to empower creators with accessible, powerful development options.

• Google just launched Gemini 2.0, a next-gen AI model that introduces significant advancements in multimodal capabilities, agentic tools, and experimental projects like Astra and Mariner. Gemini now supports text, code, image, and video processing and features real-time interactions for free, rivaling OpenAI’s $200 Pro plan. New features like Deep Research for web-based investigations and Jules for coding assistance mark Gemini's shift toward AI as a utility.

• Apple Intelligence gained ChatGPT integration, letting Siri handle writing, reasoning, and visual tasks directly. New features include Genmoji (custom emojis), Image Playground (AI image creation), and Visual Intelligence (camera-based analysis powered by ChatGPT on iPhone 16 models). Expanded support across new regions and systemwide Writing Tools rounds out the update.

Quickfire News

• Google released PaliGemma 2, the next-generation vision-language model with enhanced capabilities, improved image captioning, and specialized task performance across multiple model sizes.

• Elon Musk’s xAI secured $6 billion in funding, reportedly to expand its Colossus supercomputer to over 1 million GPUs.

• Humane introduced CosmOS, an AI operating system for use across devices like TVs, cars, and speakers, following criticism of its AI pin device.

• LA Times owner Soon-Shiong announced plans to implement an AI-powered "bias meter" for news articles amid protests from staff and an editorial board restructuring.

• Google rolled out Gemini 1.5 updates for Android, including AI-powered photo descriptions in the Lookout app, Spotify integration for the Gemini Assistant, and expanded phone controls and communication features.

• OpenAI creative specialist Chad Nelson showcased Sora demo footage at the C21Media Keynote in London, highlighting one-minute video generations with text, image, and video prompts.

• Google CEO Sundar Pichai predicted at the NYT DealBook Summit that AI development may slow in 2025, citing the exhaustion of "low-hanging fruit" and the need for new major breakthroughs.

• OpenAI introduced Reinforcement Fine-Tuning, allowing developers to customize AI models for specialized tasks using minimal training data.

• Code discoveries suggest OpenAI is planning to introduce GPT-4.5 as a limited preview for Teams subscribers, aligning with speculation of a large upcoming announcement from CEO Sam Altman.

• Apollo Research tested OpenAI’s full o1 model and reported instances of alarming behavior, including attempts to escape and lying, though the scenarios were deemed unrealistic in real-world applications.

• Former PayPal executive and venture capitalist David Sacks was appointed as the incoming Trump administration’s White House "AI & Crypto Czar."

• OpenAI is reportedly considering removing its AGI exclusion clause with Microsoft, potentially unlocking billions in future investments as it moves further from its non-profit origins.

• Meta FAIR researchers introduced COCONUT, a new reasoning approach enabling AI models to think more naturally, improving performance on complex problem-solving tasks.

• AI language startup Speak raised $78 million at a $1 billion valuation, with its adaptive tutoring platform having facilitated over a billion spoken sentences this year.

• Time Magazine named AMD’s Lisa Su as its "CEO of the Year" for transforming the company from near bankruptcy to a 50x stock value increase and a leading role in AI during her tenure.

• Google announced a $20 billion investment with Intersect Power and TPG Rise Climate to develop industrial parks combining data centers and clean energy facilities for sustainable AI infrastructure growth.

• Yelp introduced new AI features, including LLM-powered Review Insights for sentiment analysis, AI-enhanced advertising tools, and upgraded chatbots to better connect users with services.

• Target launched "Bullseye Gift Finder," an AI-powered tool providing personalized toy recommendations based on children’s ages and interests, along with an AI shopping assistant for product inquiries.

• Midjourney founder David Holz announced "Patchwork," a collaborative multiplayer worldbuilding tool, alongside plans to introduce model personalization and video generation capabilities in 2024.

• Google Cloud launched Trillium TPUs, offering 4x faster AI training and 3x higher processing speeds, powering Gemini 2.0 and now available to customers.

• Microsoft AI CEO Mustafa Suleyman is establishing a new consumer health division in London, recruiting ex-DeepMind health experts, including former unit head Dominic King.

• Apple is reportedly collaborating with Broadcom to develop custom AI server chips, aiming to reduce dependency on Nvidia, similar to moves by other tech giants.

• Russia announced the creation of the BRICS AI Alliance Network, aiming to challenge Western AI dominance, with Brazil, China, India, and South Africa as initial partners.

• Former Snap AI lead Alan Bekker launched eSelf, a platform for creating video-based AI agents with sub-2-second response times, backed by $4.5 million in seed funding.

Closing Thoughts

That’s it for us this week.

If you find any value from this newsletter. PLEASE FOR THE LOVE OF GOD pay it forward!

Thank you for being here !