A new family of late-interaction models for text and image retrieval, with a shared embedding space across model sizes.

d1-3B and d1-omni-600M, open-weight decision models from Liquid AI for text, vision, and audio that run in milliseconds from NVIDIA DGX to Jetson

OpenAI is expanding Codex with reusable cloud development environments, a revamped CLI with voice controls, new code review tools, and a security-focused product for scanning repositories and preparin...

TypeSafe’s “System One” model returns a typed value with a probability attached and never writes a word of text. That single constraint…

Google and the UN system have launched the UN System Data Commons, a new open platform making global statistics accessible and easy to search.

Atoms just work slower than bits

Mozilla's Ajit Varma explains why a redesign suits the fight for the open web.

A small, strange discovery in machine learning called grokking

What the L1–L5 framework can teach us about human supervision, authority and governance for AI

Line-by-line review is going away. Whatever replaces it has to earn the trust reading used to provide.

Atoms just work slower than bits

OpenAI keeps uncovering incidents of its models behaving in ‘unexpected or concerning’ ways.

Nemotron-3-Diarization is an open-weight speaker diarization model from NVIDIA that identifies who spoke when in real-world audio.

I tested TypeSafe AI’s Jev on 3,080 classification tasks to see how its accuracy, latency, calibration, and confidence compare with LLMs — and whether it works as a practical decision layer for AI sys...

The reliability mechanisms we add to LLM pipelines are often the ones that make them confidently wrong.

The space company wants to rapidly boost revenue over the next few years, betting its huge rockets will make it a powerful industry player.

Do AI agents have parallels of emotions? How can sentience, consciousness or affect be used for alignment against rogue or autonomous agents, adding to RLHF?

We are in the midst of a preference cascade about existential risk from AI.

...

I maintain and comaintain several open-source libraries. Some of them are widely used: ada parses URLs in Node.js, fast_float parses numbers in GCC's standard library and in Chromium, simdjson parses ...

a look at the changing economics of AI labs

For nearly 25 years, Palantir has built a unique business that few understand, inside or outside of Silicon Valley. Now, the Frontier Labs want a piece of the action.

Scienceblogs.de, a German science blogging portal, includes a relatively famous list of 50 unsolved ciphers, which range from cryptograms published by serial killers to the famous Voynich manuscript.

For enterprise teams, the larger lesson extends beyond one vulnerable image decoder. AI agents are increasingly being given credentials and connectors that span source code, communications and corporate data.

The new MCP server lets AI agents handle WhatsApp phone onboarding, message templates, webhooks, and test messages.

Why long-running agents silently drop compliance rules

I built a system that automatically discovers, verifies, and applies relevant requirements from earlier interactions without asking the user where they came from.

From a chocolate bar with no price tag to a marketing mix model in PyMC, and the 200-year-old integral that stood in between.

How to apply the latest context engineering guidelines to your day-to-day data science work

A new study shows Claude Code, Cursor, and Codex pick wildly different tools for the same job - and vendors are racing to influence the choice.

Meta’s new personal AI agent can shop and transact on consumers’ behalf, drawing on the company’s greatest advantage: years of social data. But will consumers trust it enough to hand over the reins?

Compare Chain of Thought and Tree of Thoughts reasoning to understand which approach best fits your AI agent.

MCP’s authorization overhaul exposes the real risk in vibe-coded tools: personal tokens, missing scopes, and integrations nobody is monitoring.

Economics graduates also appear to be affected as demand for them falls in well-paid roles in finance

AI agents need durable execution, recovery, and provenance. DBOS may offer a practical database-backed runtime inside agent sandboxes.

Most AI agent builders are solving the N x M integration problem from scratch. MCP already solves it. Here is what it does and when to use it.

Discover how an enterprise MCP gateway secures and governs AI agents, eliminates shadow IT, and enforces zero-secret, identity-bound access control.

The internet fights over anthropomorphism around the Hugging Face hack.

Anthropic’s Model Hardware Standard could become the MCP for labs, factories and robotics.

A practical framework for reducing uncertainty before agents accelerate implementation

Three conditions that must hold before splitting prefill from decode pays off, and why chunked prefill is the right default below that threshold.

AI has made code, tests, policy, documentation, and organisational machinery almost free to produce. It has not made any of them free to own.

Route AI inference locally across RTX, DGX, and Mac.

The AI era has completely broken enterprise buying patterns, and startups haven't yet figured out how to navigate.

I received the two emails below earlier in the month. They’re vaguely coherent. I suppose I shouldn’t be surprised that the corpus that AIs are training on contain data suggesting that I am someone to...

The number of applications a person uses to do their job is quickly collapsing to roughly one, sooner than everyone realizes. If that’s right, there are only three places to stand, and many companies

A knowledge-intensive company should keep durable organisational knowledge outside the model, and let AI do what it is unusually good at: reason across it.

Claude’s new watermark shows how AI provenance works, where metadata removal ends, and why statistical text marks are much harder to verify or defeat.

The convergence of technology, life, and creativity

Enterprise Document Intelligence [Vol.1 #B00] - Retrieval answers one kind of question. Classifying a request, matching free text to a reference list, reading a table, cleaning OCR noise: each has a c...