AI News Archive — Page 11

Learning: Best AI Guides and YouTube Channels

OpenAI, Google, and Anthropic have released official AI learning guides, plus a curated list of the top 10 YouTube channels for beginners.

Jun 16, 2025

AI Builders and Tools: "Cursor of X" Era and Dia Browser

A wave of "Cursor of X" startups—AI coding tools modeled on Cursor's success—are emerging, with Dia Browser now available as the latest entrant.

Jun 16, 2025

Apple's "The Illusion of Thinking" and the Rebuttal

Apple's research paper argues reasoning models like o1 and o3 aren't truly thinking but failing due to context limits, though a rebuttal shows o3 solves Tower of Hanoi when prompted to be concise.

Jun 16, 2025

MEMOIR: Scalable Lifelong Model Editing Framework

Researchers introduce MEMOIR, a framework that edits AI models to reliably update thousands of facts sequentially using a residual memory module.

Jun 16, 2025

SEAL: LLMs That Update Their Own Weights

SEAL lets large language models generate their own training data to update their weights in response to new inputs, with self-editing learned through reinforcement learning.

Jun 16, 2025

Reasoning Features Learnt by SAEs Transfer Across Models

Sparse autoencoders trained to identify reasoning features on one language model transfer directly to other models and datasets, suggesting reasoning capabilities may rely on shared underlying structures.

Jun 16, 2025

Sakana AI: Text-to-LoRA Hypernetwork

Sakana AI introduced Text-to-LoRA, a hypernetwork that generates task-specific LLM adapters from text descriptions, presenting the work at ICML 2025.

Jun 16, 2025

DeepSeek Drops Best End-to-End Large Model Training Paper

DeepSeek released a comprehensive paper on end-to-end large language model training methodology.

Jun 16, 2025

deedydas: The Best 340-Page AI Report

A 340-page AI report from deedydas summarizes key findings across the field's latest developments.

Jun 16, 2025

Terence Tao on Hard Math and AI

Terence Tao discusses how AI could help mathematicians tackle the century's hardest unsolved problems.

Jun 16, 2025

Commercial AI Videos Enter the Mainstream

https://x.com/PJaccetturo/status/1932893260399456513 https://x.com/ROHKI/status/1931081752992477285

Jun 16, 2025

AI Dinner 10.0 at Greycroft, May 21st

AI Dinner 10.0 at Greycroft on May 21st will discuss A2A, Ilya Sutskever's top 30 papers, Alpha Evolve, and the latest AI developments through Socratic dialogue.

May 20, 2025

Two major June AI conferences: AI Engineer SF vs AI Tech Week NY

A comparison of AI Engineer SF (June 3-5) and AI Tech Week NY (June 2-8) finds the former concentrates top speakers while the latter spreads content across multiple venues with more noise.

May 20, 2025

Google I/O 2025: Gemini agents, AI-first Android, and new hardware

Google unveiled Gemini agents across Search, Android, and new hardware, positioning AI to handle tasks autonomously while users point and direct.

May 20, 2025

DeepMind AlphaEvolve: evolutionary coding agent discovers new algorithms

DeepMind's AlphaEvolve, revealed May 14, 2025, uses evolutionary algorithms powered by Gemini-2 to discover new code that beats 56-year-old methods for matrix multiplication, recovers 0.7% of Google's compute fleet, and speeds up Gemini kernels by 23%.

May 20, 2025

OpenAI Codex: cloud SWE agent built on codex-1 reasoning model

OpenAI launches Codex, a cloud-based software engineering agent built on its o3 reasoning model that executes coding tasks like feature writing, bug fixes, and tests in parallel within sandboxed environments, completing work in 1–30 minutes with…

May 20, 2025

Pope Leo XIV chose his name because of AI

Pope Leo XIV chose his regnal name because of artificial intelligence.

May 20, 2025

Sam Altman and Jony Ive hint at a new personal AI product

Sam Altman and Jony Ive are hinting at a new personal AI product, though details remain scarce.

May 20, 2025

LLM Models Vibe Check & Benchmarks: OpenRouter, lmarena, and IQ

Gemini 2.5 climbs OpenRouter rankings while Claude 3.7 declines, as companies increasingly optimize models specifically for benchmark performance rather than general capability.

May 20, 2025

Absolute Zero: Reinforced Self-Play Reasoning with Zero Data

An AI system learns to reason by generating and solving its own Python coding problems through reinforcement learning, without any human-labeled training data.

May 20, 2025

Flow-GRPO

May 20, 2025

ZeroSearch: incentivizing search in LLMs without searching

ZeroSearch uses reinforcement learning to teach LLMs when to retrieve information by training them on self-generated documents rather than external search results.

May 20, 2025

Sakana's Continuous Thought Machines (CTM) architecture

Sakana proposes Continuous Thought Machines, a neural architecture that uses neural dynamics as its core representation and can dynamically reduce computation once patterns are recognized, similar to human thinking.

May 20, 2025

OpenAI acquires Windsurf for $3B

OpenAI acquires Windsurf for $3 billion, consolidating control over AI application layers and reducing reliance on competing models.

May 20, 2025

Why the AI wave is different: Cursor's rise from $100M to $300M ARR

Cursor tripled ARR to $300M in months, outpacing GitHub Copilot despite its data advantages, suggesting AI products succeed through pull demand and UX rather than traditional SaaS sales cycles.

May 20, 2025

Zeki Data Report: US to become a net exporter of AI talent in 2025

US AI talent net outflows reverse to inflows as domestic layoffs exceed hiring, positioning the country as a net exporter of AI workers by 2025.

May 20, 2025

How LLMs do arithmetic

Large language models struggle with basic arithmetic because they process numbers as text tokens rather than mathematical operations.

May 20, 2025

Rich Sutton on AI alignment and Decentralization [15 min video]

Rich Sutton argues AI safety researchers ask the wrong question—rather than controlling AI goals, we should ask how to build a good future without that control, similar to how societies function without controlling individuals' goals.

May 20, 2025

Random thought tweet from @goyal__pramod

https://x.com/goyal\\pramod/status/1921944575842644206

May 20, 2025

OpenAI publishes a guide on when to use which model

OpenAI published a guide explaining which model to use for different tasks, covering performance, cost, and capability tradeoffs.

May 20, 2025

Vesuvius Challenge finds a scroll title for the first time

Vesuvius Challenge successfully identified the title of a charred scroll for the first time: "On Vices, Book 1" by the Stoic philosopher Philodemus.

May 20, 2025

The Intelligence Curse: exploring how to avoid an AGI disaster

Researchers explore strategies for preventing catastrophic outcomes when artificial general intelligence arrives.

May 20, 2025

GPT model stopped learning Croatian due to downvoting users

OpenAI's GPT model stopped improving on Croatian because users in that language community downvoted responses more frequently, reducing the feedback signal for learning.

May 20, 2025

TikTok, Google, Meta can run human experiments at scale

TikTok, Google, and Meta can run psychological experiments on billions of users, dwarfing the sample sizes of famous studies that used dozens of people.

May 20, 2025

JSON uses many more tokens than alternative formats

JSON consumes significantly more tokens than alternative data formats like MessagePack or Protocol Buffers, making it less efficient for large-scale AI applications.

May 20, 2025

Loops you can take home to your mother

replace the loops with downloading the videos https://x.com/kentskooking/status/1922570670132604967 https://x.com/kentskooking/status/1921464932119286053

May 20, 2025

AI Dinner 9.0 — Ghibli Style (sponsored by Greycroft)

NYC's monthly AI Dinner 9.0 gathers researchers, engineers, and founders for Socratic discussions on AI-2027 and agent-to-agent systems, sponsored by Greycroft.

Apr 19, 2025

AI 2027: Month-by-Month Intelligence Explosion Forecast

Daniel Kokotajlo and Scott Alexander forecast an intelligence explosion by early 2028 if AI agents achieve autonomous R&D capabilities by 2027, with geopolitical tensions over AI control and safety risks shaping the outcome.

Apr 19, 2025

Google launches A2A agent-to-agent protocol with 50 partners

Google launched A2A, an agent-to-agent protocol enabling multimodal communication between AI agents via JSON cards, with 50 partners at launch and built on HTTP, SSE, and JSON-RPC for interoperability.

Apr 19, 2025

OpenAI releases o3 and o4-mini reasoning models

OpenAI released o3 and o3-mini reasoning models, with o3 scoring 87.7% on GPQA Diamond and 71.7% on SWE-Bench Verified, while o3-mini offers lower-cost performance for coding tasks.

Apr 19, 2025

OpenAI drops GPT-4.1 with Mini and Nano variants

OpenAI released GPT-4.1 with Mini and Nano variants, offering a 1M token context window, 55% score on SWE-Bench Verified coding tasks, and pricing from $0.40/$1.60 per 1M tokens for the smaller models.

Apr 19, 2025

Google Gemini 2.5 Pro launches with strong benchmarks

Google's Gemini 2.5 Pro launches with 1M token context window, 63.8% score on SWE-Bench Verified, and multimodal reasoning capabilities across text, audio, images, and video.

Apr 19, 2025

Karpathy: Building micrograd — intro to neural networks and backprop

Andrej Karpathy walks through building micrograd, a minimal neural network framework, to teach backpropagation and how neural networks learn from scratch.

Apr 19, 2025

Joscha Bach: Only a simulation can be conscious

Cognitive scientist Joscha Bach argues that consciousness can only exist within simulations, not in physical systems, because the brain itself creates a virtual reality model where consciousness emerges as a simulated property.

Apr 19, 2025

Robin Hanson on Futarchy and Grabby Aliens

Economist Robin Hanson explores whether advanced aliens a million years ahead of us are monitoring Earth, and whether they share a common origin with humanity.

Apr 19, 2025

Fun Fact: "please and thank you" are costing OpenAI millions

https://x.com/sama/status/1912646035979239430 If you're curious to know how the sausage is made 🥩 + 🥖 = 🌭 1\. Bookmark tweets, videos and blog posts over a month 2\. Combine them into a sheet https

Apr 19, 2025

AI Dinner 8.0 — MCP Study Group

AI Dinner 8.0 on Mar 19th will focus on Model Context Protocol (MCP) and AI engineering fundamentals.

Mar 16, 2025

Anthropic's Model Context Protocol (MCP) 🔥

Anthropic's Model Context Protocol lets LLMs connect to external applications through a standardized server/client interface, with adoption growing among players like Perplexity AI and a new registry called OpenTools emerging to catalog available…

Mar 16, 2025

Anthropic Claude Sonnet 3.7 (Aurora) 🛸

Anthropic released Claude Sonnet 3.7 (Aurora) on March 10, 2025, achieving a ChatEval score of 0.85 and an XGLUE score of 0.88, with the company also introducing prompt caching to reduce inference costs.

Mar 16, 2025

OpenAI releases GPT-4.5 (Orion) 🚀

OpenAI launched GPT-4.5 (codenamed Orion) on February 27, 2025, emphasizing natural conversation and reduced hallucinations with improved factual accuracy across benchmarks, though it lags specialized models in complex reasoning.

Mar 16, 2025

Learning: Best AI Guides and YouTube Channels

OpenAI, Google, and Anthropic have released official AI learning guides, plus a curated list of the top 10 YouTube channels for beginners.

Jun 16, 2025

We value your privacy

All news

Learning: Best AI Guides and YouTube Channels

AI Builders and Tools: "Cursor of X" Era and Dia Browser

Apple's "The Illusion of Thinking" and the Rebuttal