Search the archive

HOW DOES AI IMPACT SKILL FORMATION?

Anthropic's latest paper seems to be proof that AI makes people slower and dumber.

GOOGLE TESTS CLAUDE SONNET 4.5 ON GEMINI FOR BUSINESS

Google is testing the inclusion of third-party models like Claude Sonnet 4.5 in its Gemini for Business platform, offering users more model choices and potentially enhancing workflows.

10 CHARTS THAT EXPLAIN THE AI ERA

ChatGPT's rapid adoption, reaching 100M users in two months, highlights AI's unprecedented uptake compared to past technologies like cellphones and the internet.

A PEEK INSIDE PHYSICAL INTELLIGENCE, THE STARTUP BUILDING SILICON VALLEY'S BUZZIEST ROBOT BRAINS

Physical Intelligence is developing advanced robotic intelligence resembling ChatGPT for machines, using data from diverse environments like warehouses and kitchens.

COREWEAVE'S $30 BILLION BET ON GPU MARKET INFRASTRUCTURE

CoreWeave raised over $25 billion, leveraging heavy debt without a forward curve for GPU compute, akin to a 1990s independent power producer model.

SWITCH TO FRIENDLIAI — CLAIM UP TO $50K

🤗 500K open models e.g. Qwen [5] 🦫, DeepSeek [6] 🐋, GLM

💻, Llama [8] 🦙

APPLE ACQUIRES Q.AI

Apple has acquired Q.ai, an Israeli startup that specializes in audio-focused AI, to advance features like whispered speech recognition and audio enhancement for products like AirPods and Vision Pro.

AMAZON COULD INVEST UP TO $50 BILLION IN OPENAI IN COMING WEEKS, SOURCE SAYS

Amazon is negotiating a potential $50 billion investment in OpenAI, highlighting a strategic shift towards AI. Talks involve OpenAI CEO Sam Altman and Amazon's CEO Andy Jassy, with potential agreements including the use of Amazon's AI chips. This comes despite Amazon's prior investments in Anthropic, an OpenAI rival, and forms part of a broader $100 billion fundraising round spearheaded by OpenAI.

👀 INSIDE OPENAI'S UNIT ECONOMICS

The finances of the GPT-5 bundle are worse than what Sam Altman and Dario Amodei suggest. Other labs are probably in a similar boat. However, AI models don't need to be profitable today, as long as companies can convince investors that they will be in the future. The sheer scale of value that AI could create is one reason to be really bullish about AI's long-run profitability.

INSIDE OPENAI'S IN-HOUSE DATA AGENT

OpenAI uses a bespoke in-house AI data agent that explores and reasons over its platform. It lets employees go from question to insight in minutes, allowing them to answer high-impact data questions. The agent was built specifically around OpenAI's data, permissions, and workflows. This post details how it was built to demonstrate how AI can support day-to-day work across teams. The tools used to build the AI data agent are available to all developers.

APPLE'S AI GAME IS MISUNDERSTOOD

Some say Apple is behind on AI, while others say Apple is playing a different game. The gap between these two realities is where investment risk lives. Apple is likely building a moat where inference is free, privacy is structural, and the capex already happened when users bought the hardware. The next three years will tell us whether that's wisdom or complacency.

$281B FROM ONE CUSTOMER

Microsoft's Q2 FY2026 earnings beat expectations across revenue, earnings per share, and Azure growth, yet the stock fell 11% after earnings. Demand is outpacing Microsoft's ability to build, so the company is accelerating its spending. Microsoft's commercial remaining performance obligation surged to $625 billion in Q2, with $281 billion coming from OpenAI, a substantial risk. Microsoft's strength indicates an insatiable demand for inference and increasing customer concentration risk for even the largest businesses in the world.

RL ENVIRONMENTS FOR AGENTIC AI: WHO WILL WIN THE TRAINING & VERIFICATION LAYER BY 2030

The winners in RL environments will be the teams that build real infrastructure. This has to be done in close partnership with frontier labs to industrialize replication training, verification, and long-horizon environment orchestration. There will likely be three to five significant winners. The market may end up resembling data labeling, which already has roughly three over-$1 billion-revenue players.

See why it's worth your time

AGENT TRACE: CAPTURING THE CONTEXT GRAPH OF CODE

SECURE CODEBASE INDEXING

Semantic search can be sped up in large codebases by securely reusing existing indexes across team members. This optimization can reduce time-to-first-query from hours to seconds.

GROK IMAGINE API

Grok Imagine API is a unified bundle of powerful APIs designed for end-to-end creative workflows. Grok imagine features state-of-the-art video generation across quality, cost, and latency. It can bring images to life and even refine flex cinematic sequences. Examples of videos generated with the model are available in the article.

EVOLVING REPRESENTATIONS IN LANGUAGE MODELS

Google researchers reveal how internal model representations shift over conversations, showing that concepts like factuality are fluid and context-dependent, with implications for model steering and interpretability.

BUILDING AIS THAT DO HUMAN-LIKE PHILOSOPHY

Building AIs that can do human-like philosophy could prove an especially subtle and hard-to-evaluate challenge. There are a variety of strategies that are now possible that could make real progress in the area. This essay looks at different angles of attack to help inspire efforts to tackle the change. Philosophy is hard, and solving it is likely to require some very sophisticated philosophical achievement.

RETIRING GPT-4O AND OLDER MODELS

GPT-4o, GPT-4.1, GPT-4.1 mini, and certain other older models will be retired on February 13 to streamline capabilities and transition users toward newer generations like GPT-5 and beyond. GPT-4o in ChatGPT had been restored after earlier deprecation due to user feedback, but the final retirement reflects broader adoption of more capable models. While the change affects the consumer ChatGPT experience, there are currently no simultaneous API changes, and developers relying on older models will be guided to migrate to newer alternatives as part of normal deprecation practices.

FULLY REMOTE)

As the first engineer on TLDR's new Applied AI team, you'll build AI agents and composable Claude Skills to let non-technical teammates create their own AI workflows. Learn more [25].

GOOGLE DEEPMIND'S PROJECT GENIE NOW LIVE

Project Genie, an interactive AI world-building prototype, is now available to US Google AI Ultra subscribers.

FACT CHECKING MORAVEC'S PARADOX

Moravec's paradox, an endlessly repeated aphorism that tasks that are hard for humans are easy for AI and vice versa, is really just a statement on what the AI community finds worthwhile to work on and doesn't have any predictive power about which problems are going to be easy or hard for AI.

BYTEDANCE AND ALIBABA TO RELEASE NEW AI MODELS FOR LUNAR NEW YEAR

ByteDance plans to release three AI models (a large language model, an image generation model, and a video generation model), and Alibaba plans to launch Qwen 3.5, a flagship AI model optimized for complex reasoning tasks.

Get $25 in credit and start building

NVIDIA, OTHERS IN TALKS FOR OPENAI FUNDING, INFORMATION SAYS

Nvidia, Microsoft, and Amazon are in talks to invest as much as $60 million into OpenAI. The total round will contribute as much as $100 billion to OpenAI's funds, which could help alleviate investor concerns about OpenAI's cash burn while continuing the investment frenzy. OpenAI is also attempting to line up funding from top investors in the Middle East in a funding round that could total at least $50 billion. These talks are early, and the amounts could change.

GOOGLE BEGINS ROLLING OUT CHROME'S “AUTO BROWSE” AI AGENT TODAY

Google has started rolling out its autonomous browsing agent to Chrome. The new features are accessible from the AI button, which will now default to a split-screen view. Users can still pop Gemini out into a floating window, but the split-view gives the tool more room to breathe while manipulating a page with AI. Gemini in Chrome can access and edit images with Nano Banana - users just have to open an image from the web and type in the side panel with a description of the edits they want.

META'S AI PLANS INCLUDE SHOPPING AGENTS AND NEW MODELS

Meta announced a major AI product rollout for 2026, emphasizing new models and agentic tools focused on commerce. Mark Zuckerberg pointed to personalized AI shopping assistants and Meta's access to user context as key differentiators in delivering tailored agent experiences.

TRAINING A 67M-PARAMETER TRANSFORMER ON AN M4 MAC MINI

This paper takes a grounded look at what happens when you build a modern language model from first principles, train it on real data, and ask it to do something unforgiving. The author trained a 67-million-parameter transformer end-to-end on an M4 Mac Mini using Apple Silicon MPS - no discrete GPU, just 24 gigabytes of unified memory. The resulting model achieved 93.94% exact-match accuracy on CLI command generation.

ON-DEVICE LLMS IN 2026

Real-time on-device LLMs have become viable due to new deployment techniques and smarter model compression.

RESEARCH AFTER AI: PRINCIPLES FOR ACCELERATED EXPLORATION

AI integration in research transforms it from a tool to an embedded cognitive environment, accelerating exploration but demanding careful application and judgment. Researchers use AI to navigate complex problems, stress-test hypotheses, and facilitate code development and material synthesis, necessitating thoughtful oversight to prevent errors and misinterpretations. While AI speeds exploration, the responsibility for correctness, judgment, and ethical considerations remains with the human authors, addressing challenges of truth, authorship, and ethical implications.

FULLY REMOTE)

As the first engineer on TLDR's new Applied AI team, you'll build AI agents and composable Claude Skills to let non-technical teammates create their own AI workflows. Learn more [13].

TOWARDS A SCIENCE OF SCALING AGENT SYSTEMS: WHEN AND WHY AGENT SYSTEMS WORK

Smarter models accelerate the need for multi-agent systems. However, the 'more agents' approach often hits a ceiling. It can even degrade performance if not aligned with the specific properties of the task. The next generation of AI agents needs to move from heuristics to quantitative principles to become more numerous, smarter, safer, and efficient.

THE SURPRISING ATTENTION ON SPRITES, EXE.DEV, AND SHELLBOX

Several new Linux VPS services have popped up recently, all optimized to get users running from zero to 'machine running Claude Code' in a couple of minutes. They also handle the annoying bits of sharing web services, going beyond just the 'development' part and to the full end-to-end stack. These services are targeting developers who want frictionless access to machines without having to set up VMs, wrestle with containers, or configure certificates to show off prototypes. For many developers, it's nice to have something that just works.

MISTRAL VIBE 2.0

Mistral Vibe 2.0 introduces a revamped terminal-native coding assistant with features like custom subagents, slash-command skills, and workflow configuration. It's powered by the new Devstral 2 model family and available via Le Chat Pro and Team plans.

VISUAL MODELS FOR REASONING LIKE HUMANS

This project introduces a unified multimodal model that uses visual generation as part of chain-of-thought reasoning, aiming to mimic human-like reasoning through both visual and verbal modalities.

YOUTU-VL: A LIGHTWEIGHT VLM

Youtu-VL is a compact 4B-parameter vision-language model that uses a novel training strategy called Vision-Language Unified Autoregressive Supervision (VLUAS). This approach boosts general and vision-centric task performance without relying on task-specific modules.

WHY AI SWARMS CANNOT BUILD ARCHITECTURE

Agents can be individually excellent but collectively incoherent. Two agents with identical training who independently come up with the same answer aren't coordinating, they're sampling from the same distribution. Swarms are parallel, independent, and stateless. They have no propagation mechanism, and scaling makes it worse. Adding global visibility, temporal persistence, and enforcement authority to a swarm would change it into a hierarchical system with stochastic workers, creating an architecture that allows for coordination.

WORLD MODELS

The AI industry is turning its focus to world models, models that predict the next state observations. These models aim to understand the causal laws of an environment like a video game, codebase, or market. Companies have been working on similar models for decades under different names, but the bottleneck to the next step is now almost able to be overcome. The next AI era will require learning from the world directly through execution and outcomes.

MARK ZUCKERBERG SAYS A FUTURE WITHOUT SMART GLASSES IS ‘HARD TO IMAGINE'

Mark Zuckerberg claims AI smart glasses are set to become mainstream.

KEEPING YOUR DATA SAFE WHEN AN AI AGENT CLICKS A LINK

OpenAI's new safety mechanism is designed to reduce the risk of AI agents inadvertently leaking private data when they automatically fetch URLs by only allowing automatic retrieval of links that are publicly indexed on the open web.

GOOGLE TESTS VOICE CLONING ON AI STUDIO POWERED BY GEMINI

Google is testing voice cloning for AI Studio with a hidden "Create Your Voice" feature linked to the Gemini 2.5 Flash model.

Read the guide

OPENAI INTRODUCED PRISM

OpenAI's Prism is a free collaborative scientific writing platform powered by GPT-5.2. It integrates LaTeX-native writing, revision, and publishing workflows in one workspace, aiming to streamline and accelerate research.