Test-Time Training (TTT) is now on Video and not just a 5-second video. U can generate a full 1-min video!

TTT module is an RNN module that provides an explicit and efficient memory mechanism. It models the hidden state of an RNN with a machine learning model, which is updated via gradient descent.

Google announced their new TPU 'Ironwood'. Each individual chip has a peak compute of 4,614 TFLOPs. Google is calling it 'a monumental leap in AI capability'.

OKX has partnered with Standard Chartered to launch a world-leading collateral mirroring program.

Teaming up with Brevan Howard and Franklin Templeton, this initiative lets institutional clients use crypto and tokenized money market funds as secure off-exchange collateral.

Backed by a top-tier G-SIB custodian and regulated under Dubai’s VARA framework, it’s a game-changer for safety and capital efficiency in digital asset trading.

Anthropic published a new quickstart - a minimal implementation of an LLM agent with MCP tools, loops, context management.

Very helpful if you're trying to learn what an agent actually looks like in code.

OpenAI has rolled out a significant enhancement to ChatGPT's memory capabilities.

The AI can now reference your entire chat history to deliver more personalized responses based on your preferences and interests.

Unlike before, where ChatGPT only used specifically saved memories, it can now automatically analyze patterns across all your past conversations to provide more relevant assistance for writing, advice, learning, and more.

Memory isn't just another product feature. It signals a shift from episodic interactions (think a call center) to evolving ones (more like a colleague or friend).

Users maintain full control - you can opt out of this feature in settings or use temporary chats for private conversations. Currently available to Plus and Pro users (except in certain European regions).

Google finally made the best model - not just in benchmarks but real use.

Veo 2, Google’s SOTA video model, is rolling out to Gemini Advanced + Whisk

You can create 8s, high-res videos from text prompts fluid character movement + lifelike scenes across a range of styles.

Tip: the more detailed your description, the better.

Plus, you can try Veo 2 using Whisk from Google labs.

Just input images, blend them together, and – now –  “animate” to bring your creation to life. Available for all Google One AI Premium subscribers today.

OpenAI released o3/o4-mini. The eval numbers are SOTA (2700 Elo is among the top 200 competition coders)

OpenAI’s team expect o3/o4-mini will aid scientists in their research.

And the secret trick is to talk to the models in images.

Firecrawl just launched FIRE-1, a new agent-powered web-scraper.

It navigates complex websites, interacts with dynamic content, and fills forms to scrape the data you need.

🚀 Top 100 AI apps 2025: trends and insights!

🔥 ChatGPT is still the king: from 100 million to 400 million active users in a year!

🌟 DeepSeek is tearing up the market: 10 million users in 20 days, #2 in the world!

🎥 AI video is on the rise: Hailuo and Kling have surpassed Sora in visits.

💻 For developers: Cursor and Bolt (2 million users in 2 months) are heating up.

💡 Interesting: popularity ≠ profit. Apps for plants (PictureThis), nutrition (Cal AI) and languages ​​(Loora) bring in millions, even without the top audience.

🧠 My opinion: the AI ​​market is a field of opportunities! Newcomers like DeepSeek and niche products prove that there is a place for everyone.

Yale and GoogleDeepMind introduced C2S‑Scale a family open-source LLMs trained to “read” and “write” biological data at the single-cell level.

Google dropped Gemini 2.5 Flash, a reasoning model matching o4-mini in preview

It's 'thinking budget' (up 24k tokens), can balance between answer quality, cost, and speed

The model performs particularly well on reasoning, STEM, and visual reasoning

Nvidia presents Eagle 2.5

- A family of frontier VLMs for long-context multimodal learning
- Eagle 2.5-8B matches the results of GPT-4o and Qwen2.5-VL-72B on long-video understanding.
Paper.

OpenAI now projects $125B in revenue in 2029, with $25B of that from new products not yet announced

The Information reports OpenAI forecasts revenue reaching $125 billion in 2029 and $174 billion in 2030, mainly from AI agents, subscriptions, monetizing free users, and potentially affiliate fees.

According to internal documents seen by The Information, OpenAI expects revenue in 2029 to include $29 billion from AI agents, $50 billion from ChatGPT subscriptions, $22 billion from API access, and $25 billion from monetizing free users and other new, unspecified products.

CEO Sam Altman mentioned recently affiliate fees or taking a percentage of sales generated through user searches as possible revenue sources, while CFO Sarah Friar told the Financial Times there are “no active plans” for selling traditional advertising.

If they hit it, the current valuation ($300B) will be a steal; Google does ~$400B in revenue and is worth $2T.

Tesla has updated its app to include a ridesharing feature 🕑

This is likely a test feature that is part of the preparatory work for the launch of Robotaxi.

OpenAI is introducing shopping features in ChatGPT today, powered by GPT-4o, for ChatGPT Pro, Plus, and Free users, as well as logged-out users worldwide.

Meta is releasing a standalone mobile app for its ChatGPT competitor, Meta AI.

There's a Discover feed that shows interactions that others (including your IG/FB friends) are having with the assistant. Meta tells the idea is to demystify AI and show “people what they can do with it."

OpenAI is working on a similar feed for ChatGPT.

Xiaomi MiMo-7B: a 7B reasoning model series trained from scratch, outperforms 32B+ baselines on math and code via dense RL

- pretrained on 25T tokens w/ multi-token prediction
- RL rewards from rule-verifiable math/code tasks
- cold-start RL model (MiMo-7B-RL-Zero) hits 93.6% MATH-500, 49.1% LCB v5
- SFT→RL variant matches OpenAI o1-mini
- also open: base + SFT checkpoints
- seamless rollout engine: 2.29× faster RL training
- vLLM + MTP inference ready
- strong AIME 2025 (55.4%) and LCB v6 (49.3%) results

Microsoft is getting ready to host Elon Musk’s Grok AI model. Microsoft has been in discussions with xAI to make Grok AI available on Azure's AI Foundry service.

In recent weeks Microsoft has been in discussions with xAI to host the Grok AI model and make it available to customers and Microsoft’s own product teams through the Azure cloud service.

The move could prove controversial internally and further inflame tensions with Microsoft’s partner OpenAI.

Microsoft Introduced Phi-4-reasoning, adding reasoning models to the Phi family of SLMs.

The model is trained with both supervised finetuning (using a carefully curated dataset of reasoning demonstration) and Reinforcement Learning.

- Competitive results on reasoning benchmarks with much larger top-tier models up to DeepSeek R1.

- Strong performance on new tests released after data collection (AIME 2025, HMMT).

- Reasoning transfers/generalizes well to new domains even with only SFT (e.g. k-SAT, Mae Solving, Calendar Planning, etc.)

- Retains and often significantly improves general-purpose capabilities (e.g. instruction following).

HuggingFace Phi-4-reasoning
HF Phi-4-reasoning-plus
Hf Phi-4-mini-reasoning