AI Nears 94% to AGI

PLUS: OpenAI’s New Structure Faces Pushback, Grammarly Raises $1B for Expansion and more.

In partnership with

Receive Honest News Today

Join over 4 million Americans who start their day with 1440 – your daily digest for unbiased, fact-centric news. From politics to sports, we cover it all by analyzing over 100 sources. Our concise, 5-minute read lands in your inbox each morning at no cost. Experience news without the noise; let 1440 help you make up your own mind. Sign up now and invite your friends and family to be part of the informed.

Today:

  • AI Nears 94% to AGI

  • DeepSeek R1-0528 Challenge OpenAI o3, Gemini 2.5 Pro

  • Perplexity Launches AI Productivity Suite

  • OpenAI’s New Structure Faces Pushback

  • Grammarly Raises $1B for Expansion

AI Researcher SHOCKING "Singularity in 2025 Prediction"

AI analyst Alan D. Thompson claims the singularity will hit in mid-2025. His latest Memo says humanity is 94 percent of the way to general-purpose AI and tracks fifty milestones toward super-intelligence, none finished yet but several taking shape. 

Early hints include Microsoft’s AI finding a safer battery coolant, Google’s Alpha Evolve redesigning chips and code, and talk of governments writing AI laws. Ilia Sutskever even suggests a bunker before releasing AGI, showing the high stakes.

Chinese startup DeepSeek has upgraded its free model to R1-0528, matching the paid giants from OpenAI and Google. The update answers tougher reasoning tests, scoring 87.5% on AIME math and 73.3% on LiveCodeBench coding. New perks include JSON output, function calls, fewer made-up answers, and no awkward system tag. An 8-billion-parameter slim version fits one gaming GPU, keeping advanced language power open to everyone at no extra cost.

Why this matters

  1. Open tech matches leaders – A freely licensed model now rivals the best paid systems, proving open tools can keep pace.

  2. Lower barrier to entry – Small, affordable hardware can run a powerful variant, letting startups, students, and researchers experiment without huge budgets.

  3. Faster innovation pressure – Big companies must keep improving when open-source options close the gap, speeding progress for the whole AI field.

Perplexity has unveiled Labs, a workspace for its $20-a-month Pro subscribers that builds full reports, spreadsheets, dashboards (visual summaries) and even web apps (interactive pages). The tool spends up to ten minutes gathering facts through live search, writing code, making charts and packaging every file for download. Labs works on the web and mobile now, with Mac and Windows versions coming, showing Perplexity’s push beyond search into practical productivity services.

Why this matters

  1. AI tackles real office work – Labs shows that a search chatbot can now create finished documents, code, and graphics with little human help.

  2. Cuts barriers for non-experts – Anyone can turn raw information into clear tables and charts without knowing coding or data tools.

  3. Heats up platform race – As Perplexity joins rivals adding “agent” features, competition will speed up progress across the AI field.

OpenAI dropped its plan to spin off its money-making arm. It now wants to make it a public-benefit company still overseen by the charity that owns OpenAI. Activist groups say the charity’s rumored 25% stake is unfair and urge prosecutors in California and Delaware, who must approve the deal, to demand better terms and changes. Failure could cancel SoftBank’s planned $20 billion investment.

Why this matters

  1. Who steers AI’s future – The restructuring will decide who ultimately controls ChatGPT and other powerful models.

  2. Money for research – A pending $20 billion cash injection may vanish if the deal stalls, slowing OpenAI’s progress.

  3. Rulebook for others – Regulators’ verdict will set an example for how mission-driven AI nonprofits balance public good and profit.

🧠RESEARCH

The paper explores why AI models trained with reinforcement learning often stop exploring too soon. It finds that this is caused by a drop in randomness, or entropy, during training. The authors propose two simple fixes that limit overconfident actions, helping the models keep exploring and perform better in reasoning tasks.

SWE-rebench is a system that pulls real software engineering tasks from GitHub to better train and test AI coding agents. It solves two big problems: lack of real-world training data and outdated benchmarks. SWE-rebench offers over 21,000 clean, interactive Python tasks, improving model evaluation and reducing test result bias.

R2R is a method that smartly switches between small and large AI models during reasoning tasks. It identifies key moments where the small model struggles and routes only those parts to the larger, more powerful model. This boosts performance while saving time and resources, achieving faster, smarter results.

🛠️TOP TOOLS

AI Code Converter - AI-powered tool that offers code conversion, translation, and generation capabilities across over 50 programming languages.

Magical AI - AI writing assistant that integrates seamlessly with over 10 million apps, allowing users to draft emails, messages, and automate repetitive tasks directly from their browser.

DeepBrain AI - AI-powered video creation, featuring realistic AI avatars, natural text-to-speech capabilities, and advanced editing tools.

Claid AI - AI-powered photo enhancement platform designed specifically for e-commerce businesses to improve user-generated content and product imagery. 

InstantArt - AI-powered platform that allows users to generate original artwork using over 25 fine-tuned stable diffusion models.

📲SOCIAL MEDIA

🗞️MORE NEWS

  • Grammarly raised $1 billion from General Catalyst to expand its AI tools and grow into a full productivity platform. The deal avoids giving up ownership and supports future acquisitions and a possible IPO.

  • Mistral released Codestral Embed, a new AI model for code search and retrieval that beats OpenAI and Cohere on real-world benchmarks. It supports fast, low-cost retrieval for tasks like code search, analytics, and clustering.

  • YouTube is adding Google Lens to Shorts, letting users tap, circle, or highlight objects in videos to instantly search for them. This new feature helps identify landmarks, animals, or products without leaving the video.

  • Black Forest Labs launched FLUX.1 Kontext, an advanced AI image model that edits and generates images using text or reference visuals. It supports precise edits, consistent styles, and fast performance for enterprise workflows.

  • Delaware’s attorney general has hired an investment bank to independently assess OpenAI’s restructuring plan, potentially delaying its for-profit conversion. The review focuses on the nonprofit’s equity stake, possibly influenced by Elon Musk’s $97.4B bid.

  • Meta is partnering with defense startup Anduril to develop AI- and AR-powered military tools. The goal is to give soldiers real-time battlefield intelligence. Meta’s tech and Anduril’s autonomous weapons aim to improve decision-making and safety.

  • Hugging Face has unveiled two open-source humanoid robots, HopeJR and Reachy Mini, aimed at making robotics more accessible. HopeJR is full-sized with 66 movements; Reachy Mini is a desktop unit for testing AI apps.

What'd you think of today's edition?

Login or Subscribe to participate in polls.

Reply

or to participate.