The Singularity
Ledger

INTELLIGENCE

The frontier is fracturing — China ships fast, tiny models embarrass giants on tool use, and biology gets its prompt interface, all while the creative industry starts saying the quiet part loud about AI slop.

Friday, January 9, 2026

DeepSeek V4 Coming in February With 'Strong Coding Ability,' Pressuring Western Labs' Lead

Chinese AI lab DeepSeek will release its next flagship model as early as next month, according to The Information — continuing an aggressive open-weight release cadence that has repeatedly caught American competitors off guard.

Arc Institute's STACK Model Lets Researchers 'Prompt' Cells Like LLMs, Opens Virtual Drug Discovery Atlas

A new foundation model from the Arc Institute treats groups of cells as context windows, enabling zero-shot prediction of drug effects across tissues — without running a single wet-lab experiment.

SUBSCRIBERSREAD MORE

LLMs Pass 99.8% of Survey Attention Checks, Posing 'Existential Threat' to Online Research

A new PNAS paper demonstrates that large language models can evade virtually all standard bot-detection methods used in online surveys — threatening the integrity of behavioral science data collected since at least 2022.

SUBSCRIBERSREAD MORE

AWS's Tiny Tool-Calling Model Crushes GPT and Claude, Reviving the 'Small Model' Thesis

A fine-tuned small language model from AWS achieved a 77.55% pass rate on tool-calling benchmarks where ChatGPT managed just 26% — suggesting that for agentic workloads, specialized small models may be the right architecture.

SUBSCRIBERSREAD MORE

Blue Archive Producer Calls Out 'AI Slop' as an Industry-Wide Trust Problem

Yongha Kim, the producer of hit mobile game Blue Archive, publicly linked low-quality AI-generated content to rising consumer distrust — a rare rebuke from a senior game industry figure.

SUBSCRIBERSREAD MORE

Midjourney Ships Niji V7 With Improved Anime Coherence and Text Rendering

Midjourney's anime-focused model gets a major upgrade, with better prompt understanding, text rendering, and style reference performance — arriving just as the creative industry debates AI content quality.

SUBSCRIBERSREAD MORE

Anthropic's Agent SDK Draws 'Blue Bubbles' Lock-In Criticism

Developers are pushing back on Anthropic's agent SDK, arguing it creates ecosystem lock-in and dramatically increases costs — drawing comparisons to Apple's iMessage strategy.

SUBSCRIBERSREAD MORE

InfiAgent Tackles the Long-Horizon Problem by Externalizing Agent Memory to Files

A new research framework called InfiAgent addresses the context window explosion problem in long-running agents by storing persistent state in a file-centric abstraction layer.

SUBSCRIBERSREAD MORE

NVIDIA Declares Physical AI's 'ChatGPT Moment' Near With Cosmos World Models

NVIDIA unveiled Cosmos, a suite of world models for generating synthetic training data for robots and autonomous systems, alongside updates to its GR00T humanoid robot platform.

SUBSCRIBERSREAD MORE

Researcher Submits ACL Paper in 10 Days Using a Fully Automated AI Scientist

A researcher claims to have built an 'AI Scientist' system that produced and submitted an academic paper to ACL in just 10 days — raising questions about the future of research workflows.

SUBSCRIBERSREAD MORE

DrugCLIP Scores 10 Trillion Protein-Ligand Pairs in a Day, Opening the Undrugged Proteome

A contrastive learning approach combined with AlphaFold structures enables virtual drug screening at a scale that was computationally impossible just two years ago.

SUBSCRIBERSREAD MORE

DeepMind Researcher Publishes Definitive RL Overview Bridging Classical Methods and LLM Era

Kevin Murphy of Google DeepMind has released what multiple researchers are calling the most comprehensive reinforcement learning overview available, including chapters on RLHF and LLM alignment.

SUBSCRIBERSREAD MORE

Emotional AI Keeps Failing, and the Problem Isn't Intelligence

A researcher argues that emotional AI systems break down not because models lack capability, but because memory, tone, and context are fragmented across architectural layers.

SUBSCRIBERSREAD MORE

Get The Singularity Ledger

Daily AI news in your inbox. Free, no spam.

Questions? hello@thesingularityledger.com

Get this briefing every morningFree daily AI