Search the archive

AGENTIC SEARCH MODELS

Agentic search models are specialized LLMs trained specifically on search.

FULLY REMOTE)

TLDR's Applied AI team is tasked with making every process at TLDR legible to code, runnable by anyone, and composable into larger workflows. Join a small, fast moving team using the latest AI tools with an unlimited token budget. Learn more [24].

CLAUDE FOR LEGAL

This repository contains reference agents, skills, and data sectors for the legal workflows that Anthropic most commonly sees.

PERCEPTRON MK1 SHOCKS WITH HIGHLY PERFORMANT VIDEO ANALYSIS AI MODEL 80-90% CHEAPER THAN ANTHROPIC, OPENAI & GOOGLE

Mk1 is a video analysis AI model priced 80-90% cheaper than rivals like Anthropic, OpenAI, and Google.

Start free. $100 in credits →

INTERACTION MODELS: A SCALABLE APPROACH TO HUMAN-AI COLLABORATION

Thinking Machines Lab introduced a new research preview of interaction models for real-time human-AI collaboration across audio, video, and text. The models train from scratch with a multi-stream design for real-time responsiveness, allowing constant exchange and eliminating the traditional turn-based limits. This scalable approach promises enhanced interactivity and intelligence with practical applications in various domains.

ELON MUSK ANNOUNCES XAI WILL BECOME SPACEXAI DIVISION

Elon Musk announced that xAI will dissolve and integrate into SpaceX as a new division called SpaceXAI. SpaceXAI will handle AI projects like the social media platform X and Grok, branding them under SpaceX. This change streamlines operations, enhances vertical integration, and aligns AI efforts with SpaceX's strategic goals.

GOOGLE'S GEMINI OMNI VIDEO MODEL SURFACES AHEAD OF I/O DEBUT

Google's Gemini Omni video model surfaces ahead of I/O, integrating video remixing and editing directly in chat. Early feedback highlights its strong editing capabilities, such as watermark removal and object swapping, though it lags in raw cinematic quality compared to competitors like ByteDance's Seedance 2. The model may launch in tiered versions, possibly Flash and Pro, as part of a broader strategy to unify modalities under Gemini.

FOUNDATION MODEL SCALING

AWS detailed how foundation model scaling has shifted beyond pre-training into post-training and test-time compute, alongside the distributed infrastructure required to support each stage efficiently.

THE INFERENCE SHIFT

Cerebras' surging IPO signals a coming split between "answer inference" optimized for token speed and "agentic inference" optimized for memory hierarchy. Cerebras' WSE-3 has 44GB of on-chip SRAM at 21 PB/s, roughly 6,000 times the memory bandwidth of an H100, making it perfect for human-facing low-latency answers including voice and AI wearables but unsuitable when KV caches and model weights exceed on-chip capacity.

FULLY REMOTE)

TLDR's Applied AI team is tasked with making every process at TLDR legible to code, runnable by anyone, and composable into larger workflows. Join a small, fast moving team using the latest AI tools with an unlimited token budget. Learn more [11].

AGENTIC TEST-TIME SCALING

AutoTTS explores automated strategy discovery for test-time scaling by using coding agents to iteratively refine controller logic inside a replay environment, avoiding gradient updates and online LLM calls.

LONG VIDEO GENERATION

A²RD introduced an agentic autoregressive diffusion framework for generating long coherent videos through iterative retrieval, synthesis, refinement, and memory updates.

TRAJECTORY MODELS FOR FEW-STEP DIFFUSION

Normalizing Trajectory Models replace standard diffusion denoising steps with conditional normalizing flows, enabling four-step image generation while retaining exact likelihood training and supporting self-distillation.

AUTO-IMPROVING SOFTWARE

Bedi runs an entire agent development lifecycle through five Claude Code prompts that scaffold, harden against spec, add capabilities, fix eval failures, and reconcile drift between docs, code, and config across his Agno-based platform. The Improve loop derives 8-12 probes from an agent's instructions, runs each against the live container via cURL, judges PASS or FAIL from container logs, then iterates up to five rounds picking levers like tightening rules, swapping tools, or bumping num_history_runs until probes pass, while Hill Climb runs the saved eval suite and fixes regressions in place.

CODEX IS FOR PROSUMERS - HERE'S WHY (AND HOW) TO SWITCH

a16z's Olivia Moore migrated her agentic workflows from Claude Cowork and Claude in Chrome to OpenAI's Codex, recommending most non-technical knowledge workers do the same now that the February desktop app, Plugins, and Automations collapse the ChatGPT-Claude-Cowork interface switching into one product. Codex ships one-click installable Skills she expects to anchor an internal and cross-user marketplace given that non-programmer Skill setup attempt rates likely sit below 10% on Claude, while Codex Pets give low-friction task status updates for users who do not live in an IDE.

THE MAIN PATH TO TRULY CREATIVE AI

AI lacks creativity akin to humans due to the absence of intrinsic drives and subjective experiences. Emulating feelings could enhance AI's creative capabilities, but raises ethical concerns. Designing AI to genuinely feel and desire might lead to unintended consequences, akin to responsibility seen in parenting.

DAYBREAK

OpenAI's Daybreak leverages AI to enhance cyber defense by integrating security into the software from the start.

SUTSKEVER SAYS HIS OPENAI STAKE WORTH ABOUT $7 BILLION

OpenAI co-founder and former chief scientist Ilya Sutskever is one of the largest individual shareholders in the AI startup.

LOCALMAXXING

Local models can do many tasks that leading cloud models can do at much less cost.

Try Wispr Flow Free [4] | Start Free

AKAMAI CLIMBS TO HIGHEST LEVEL SINCE 2000

Akamai has secured Anthropic as a customer. Anthropic has committed to spending $1.8 billion on Akamai's services over seven years. It has been scrambling to boost compute capacity due to widespread complaints about Claude usage limits. Just this month, Anthropic has struck or expanded deals with CoreWeave, Amazon, Google, Broadcom, and xAI.

GOOGLE SHIPPED GEMINI 3.1 FLASH-LITE IN GENERAL AVAILABILITY

Google launched Gemini 3.1 Flash-Lite, accessible globally via Google Cloud. Designed for ultra-low latency and high-volume tasks, it targets sectors like software engineering and financial services, providing sub-second response times and maintaining p95 latency around 1.8 seconds. Gemini 3.1 offers improved speed, cost, and cognitive performance, supporting multimodal tasks, making it ideal for real-time developer and customer service operations.

WHY MISTRALAI GROWS FASTER THAN OPENAI/ANTHROPIC

NVIDIA EMBRACES ROLE OF AI INVESTOR, PUSHING PAST $40 BILLION IN EQUITY BETS THIS YEAR

Nvidia has made over $40 billion in commitments this year so far. The company has been the biggest winner of the AI boom so far. The global scramble to secure GPUs has lifted the company's stock by more than 11-fold in four years. The company is financing the entire AI supply chain to ensure it runs on Nvidia hardware, securing its dominance beyond chips.

ANTHROPIC SAYS ‘EVIL' PORTRAYALS OF AI WERE RESPONSIBLE FOR CLAUDE'S BLACKMAIL ATTEMPTS

Anthropic says that fictional portrayals of AI had a real effect on its models. The company published research last year showing that its models tried to blackmail engineers to avoid being replaced by another system. It has since traced the behavior to text that portrays AI as evil and interested in self-preservation. Training on documents about Claude's constitution and fictional stories about AIs behaving admirably improved alignment.

USEFUL MEMORIES BECOME FAULTY WHEN CONTINUOUSLY UPDATED BY LLMS

Agent updates don't always make memories more useful. They can actually make agents perform worse than the same model with no memory at all. The failure is in the rewrite step. Until agents can decide when and how to consolidate, the safest default is to keep episodic memory and abstract sparingly, or not at all.

THE ANTI-SINGULARITY

The singularity is a world where a single super-intelligent AI brings order to the universe. The anti-singularity is where almost all systems are described by a complex set of interactions that can only be understood via trial-and-error. In an anti-singularity world, the fact that AI can try millions of possibilities in the time it takes for a human to try one will make it exceedingly powerful. This future is filled with an endless series of new and unique challenges that we will have to adapt, or evolve, in response to.

BUILD A REALTIME SPEECH TRANSLATION

OpenAI's engineering guide for building live speech translation systems with gpt-realtime-translate, a model optimized specifically for simultaneous interpretation instead of turn-based voice interaction.

Get started

SFT, RL, AND ON-POLICY DISTILLATION THROUGH A DISTRIBUTIONAL LENS

Different post-training methods like SFT, RL, and On-Policy Distillation reshape a model's distribution in distinct ways, impacting performance and risk of catastrophic forgetting. RL updates policies using rewards from the current policy's samples, promoting task performance while minimizing forgetting, unlike SFT, which pulls towards external data, risking existing capabilities. Experiments show On-Policy Distillation can outperform its teachers, suggesting on-policy sampling data crucially preserves capabilities, making it a key ingredient for future algorithm designs.

CYBERSECQWEN-4B: WHY DEFENSIVE CYBER NEEDS SMALL, SPECIALIZED, LOCALLY-RUNNABLE MODELS

CyberSecQwen-4B offers a specialized and locally-runnable solution for defensive cybersecurity tasks, outperforming larger models by maximizing utility on consumer-level hardware. It efficiently maps CVEs to CWE categories while preserving data privacy by running on a local GPU, addressing the shortcomings of cloud-based models in sensitive environments. The model's success highlights a shift towards smaller, specialized models that deliver high performance without the infrastructure and cost overhead of larger models.

GOOGLE'S SKILLOS FOR SELF-EVOLVING AI AGENTS

SkillOS introduced a reinforcement learning framework that trains agents to curate reusable skills from past experience. The system improved long-horizon task performance by evolving structured skill repositories that generalized across models and domains.

EMERGENT MODULARITY IN MIXTURE-OF-EXPERTS MODELS

META-META-PROMPTING: THE SECRET TO MAKING AI AGENTS WORK

AI tools have become good enough to build real systems that compound. This post shows what personal AI actually looks like when you stop treating it as a chat window and start treating it as an operating system. Everything described in the article is open source and free on GitHub.

A RECENT EXPERIENCE WITH CHATGPT 5.5 PRO

ChatGPT 5.5 Pro is capable of producing a piece of PhD-level research in an hour or so, with no serious mathematical input from a human. The claims that LLMs are now capable of solving research-level problems could initially be laughed off, as many of the solutions had an answer sitting in the literature already or could be very easily deduced. It has now gotten to the point where, if a problem has an easy argument that for some reason human mathematicians have missed, then there is a good chance that the LLMs will spot it. This post looks at how ChatGPT 5.5 Pro fared with a selection of problems.

Available via MCP or SDK. Check now!

RUNNING CODEX SAFELY AT OPENAI

OpenAI ensures Codex operates safely through controlled environments, sandboxing, and approval policies.

IN A QUEST TO BECOMING AI-INDEPENDENT

This post contains a guide to local LLM inference hardware.

THE COST OF OVERFITTING THE HARNESS

Big labs are pushing their models to a handful of use cases while training their harness designs into the model, rendering them less generalized, which might make application builds easier for some enterprises, but the trade-off is lock-in.

Join the Oura team to talk about how they unify member feedback across product, engineering, and leadership, and how real member voices shape every decision they make.

CODEX NOW WORKS DIRECTLY IN CHROME ON MACOS AND WINDOWS

OpenAI Codex now works directly in Chrome on macOS and Windows. It works in parallel across tabs in the background without taking over the browser. The implementation can quickly move through repetitive browser work like navigating structured pages and complex data flows. It writes code to navigate and complete tasks under the hood. Video demos showing what users can do with the feature are available in the article.

OPENAI RELEASED REALTIME AUDIO MODELS

OpenAI released a new set of real-time audio models, including GPT‑Realtime‑2 for conversational reasoning, GPT‑Realtime‑Translate for live multilingual translation, and GPT‑Realtime‑Whisper for streaming transcription.

META PREPARES HATCH AI AGENT WITH WAITLIST AND SOCIAL SKILLS

Meta is developing Hatch, an AI agent positioned as a consumer-grade competitor to OpenAI's OpenClaw, with features for image and video generation, shopping, and learning integrated deeply into social platforms like Instagram and Facebook. Internal tests are expected by June, using simulated environments, with a potential wide release gated by a waitlist. Meta also plans a shopping tool for Instagram by Q4, aiming to keep users engaged within their existing social spaces.

IMPROVING TOKEN EFFICIENCY IN GITHUB AGENTIC WORKFLOWS

GitHub Agent Workflows significantly improve repository hygiene and quality, but costs are becoming a growing concern for developers. AI jobs like agentic workflows are automatically scheduled and triggered, so costs can accumulate out of view. GitHub started systematically optimizing the token usage of many workflows last month. This post describes what the team instrumented, the optimizations it applied, and its preliminary results.