The AI Edge

Follow @The_AI_Edge to easily get a digestible and smaller version of the main updates bi-weekly in the world of AI.

Bi-Weekly AI Updates (Jan 1st to Jan 15th): Major news from OpenAI, NVIDIA, Microsoft, and more 1. ChatGPT enters personal health AI - @OpenAI launched ChatGPT Health, a private mode that connects fitness apps and medical records for more personalized health conversations. Health data runs in an isolated, encrypted environment and won’t be used for model training. 2. Microsoft maps the global AI adoption gap - @Microsoft reports global AI adoption at 16.3%, but with sharp inequality. The UAE leads at 64%, while the U.S. ranks 24th. DeepSeek AI models are gaining traction in underserved regions, driven by free and open access. 3. ChatGPT becomes a health helpdesk - Over 40 million people now use ChatGPT daily for health questions, from symptoms to insurance issues. Most usage happens outside clinic hours, especially in rural and underserved areas. 4. Anthropic launches Claude for healthcare - @AnthropicAI introduced Claude for Healthcare, an AI assistant built for regulated medical and life science workflows like insurance checks, documentation review, and medical coding. 5. Nvidia open-sources AI for self-driving - @nvidia unveiled Alpamayo, an open-source AI model family that helps autonomous vehicles reason through complex driving scenarios. The release includes simulation tools and real-world driving data. 6. DeepSeek hints at cheaper AI training - New @deepseek_ai research outlines a training method that improves model stability and reasoning without significantly increasing compute costs, signaling further efficiency gains ahead. And there was more… You can read all of them with a detailed breakdown in the newsletter. Link in bio - @The_AI_Edge

Follow @The_AI_Edge to easily get a digestible and smaller version of the main updates bi-weekly in the world of AI.

Bi-Weekly AI Updates (Dec 1st to Dec 15th): Major news from DeepSeek, Mistral, Runway, OpenAI, and more. 1. @deepseek_ai’s new AI rivals GPT-5, Gemini 3 – DeepSeek launched V3.2 and V3.2-Speciale, open MIT-licensed models matching GPT-5 and Gemini 3 Pro on reasoning, with gold-level Olympiad scores and far lower API pricing. 2. Mistral 3 brings AI to every device – @MistralAI unveiled 10 open-weight models, from enterprise-scale to lightweight variants that run locally on laptops, phones, and edge devices without internet access. 3. DeepSeek’s model hits gold at IMO 2025 – DeepSeek-Math-V2 matched top human and AI performance on elite math contests, using a generator–verifier setup that lets models critique and fix their own reasoning. 4. Runway Gen-4.5 tops video leaderboard – @runwayml Gen-4.5 climbed to the top of text-to-video rankings, delivering more realistic motion, physics, and cinematic visuals for creative workflows. 5. @OpenAI reveals enterprise AI habits – OpenAI data shows 75% of workers complete tasks faster or better with AI, while heavy users save 10+ hours a week, highlighting how AI mastery compounds productivity. And there was more… You can read all of them with a detailed breakdown in the newsletter. Link in bio - @The_AI_Edge

Bi-Weekly AI Updates (Nov 15th to Nov 30th): Major news from Anthropic, OpenAI, Google, xAI and more. Opus 4.5 breaks 80% on SWE-Bench – @AnthropicAI's new flagship becomes the first model to cross 80% on SWE-Bench Verified, sets new highs in reasoning and tool use, and drops pricing by 66%, putting pressure on GPT-5.1 and Gemini 3. @OpenAI's Codex-Max unlocks 24-hour coding – GPT-5.1 Codex-Max introduces a “compaction” system that keeps context alive across million-token sessions, outperforming Codex-High and Gemini 3 Pro while running faster with 30% fewer tokens. @Google’s Gemini 3 hits record reasoning scores – Gemini 3 outperforms earlier GPT-5 marks across science, math, multimodal tasks, and tool use, and arrives with Antigravity, a new free platform for agentic coding and browser-level automation. Grok 4.1 gets a major personality upgrade – @xAI’s latest boost delivers the highest emotional-intelligence scores among major models, cuts hallucinations to 4%, and ranks No. 1 in LM Arena with more expressive, empathetic responses. Nano Banana Pro steps up 4K visual AI – Google’s new image model handles 4K outputs, multi-person identity consistency, accurate text rendering, and up to 14 visual references, making complex compositions and design tasks far easier. And there was more… You can read all of them with a detailed breakdown in the newsletter. Link in Bio - @The_AI_Edge

Bi-Weekly AI Updates (Nov 1st to Nov 15th): Major news from McKinsey, Wharton, Anthropic, OpenAI and more. McKinsey reveals AI’s reality check – @McKinsey's “State of AI 2025” finds 88% of firms use AI, but only a third have scaled it beyond pilots and just 6% report major ROI. High performers win by reengineering workflows, not layering AI on top of old systems. Wharton finds C-suites betting big on AI – Wharton’s new report shows 88% of companies plan to grow AI budgets, and 60% already have Chief AI Officers. Three-quarters now measure ROI, signaling enterprise AI is finally moving from hype to business impact. Moonshot AI unveils Kimi K2 Thinking – Backed by @AlibabaGroup, Moonshot’s open-source Kimi K2 rivals GPT-5 and Claude 4.5 in reasoning, coding, and creativity, trained for under $5M. It’s the strongest sign yet of China’s rapid catch-up in frontier AI. OpenAI rolls out smarter GPT-5.1 – @OpenAI's latest update brings Instant and Thinking models to ChatGPT. Instant feels warmer and more human; Thinking adapts to task complexity. Users can now set custom “personalities” for tone and style. Anthropic explores Claude’s introspection – @Anthropic’s new study finds Claude can sometimes recognize when concepts are artificially inserted into its reasoning. It’s a small but fascinating step toward AI models that can monitor and perhaps manipulate their own thoughts. And there was more… You can read all of them with a detailed breakdown in the newsletter. Link in Bio - @The_AI_Edge

Bi-Weekly AI Updates (Oct 15th to Oct 31st): Major news from Anthropic, OpenAI, Google, Odyssey and more. Anthropic drops fast, low-cost Claude Haiku 4.5 – @AnthropicAI launched Claude Haiku 4.5, its smallest model yet, offering near-Sonnet 4 performance at one-third the cost. It outperforms in math, reasoning, and coding. OpenAI launches ChatGPT Atlas browser – @OpenAI introduced Atlas, a new AI browser built around ChatGPT. It adds Agent mode for autonomous web actions and “memories” for personalized browsing. Survey: Google leads Generative Media race – Artificial Analysis’s 2025 State of Generative Media report shows @Google dominating the field: Gemini leads in AI image creation (74%) and Veo in video (69%), with enterprises reporting strong ROI from generative media. Odyssey-2 brings real-time AI video – Odyssey launched Odyssey-2, a model that generates and streams video at 20 FPS, updating every 50 ms. Users can direct evolving scenes midstream via chat, hinting at the future of interactive AI simulations. Pew’s global survey reveals AI jitters – A survey of 28,000 adults across 25 countries found anxiety outpacing optimism about AI. The EU emerged as the most trusted regulator, while younger users showed more comfort and enthusiasm overall. And there was more… You can read all of them with a detailed breakdown in the newsletter. Link in Bio - @The_AI_Edge

Bi-Weekly AI Updates (Sep 15th to Sep 30th): Major news from xAI, Anthropic, OpenAI, Google and more. xAI debuts Grok 4 Fast – @xAI unveiled a hyper-efficient reasoning model that cuts costs by 98% while still outperforming Claude 4.1 Opus and Gemini 2.5 Pro. It supports a 2M token context and built-in tools for browsing and code execution. @AnthropicAI releases Claude Sonnet 4.5 – Anthropic launched Sonnet 4.5, calling it its “best coding model yet.” The model delivers state-of-the-art results on SWE-bench verified, a nearly 20% upgrade over Opus 4.1, and can autonomously code for 30+ hours, generating 11,000 lines of code. Alibaba (@AlibabaGroup) drops multimodal Qwen3 wave – Alibaba released six new Qwen3 models across text, vision, audio, and safety. Highlights include Qwen-Max with 1T parameters, Qwen-Omni for multimodal tasks, and Qwen-VL surpassing top visual benchmarks. @OpenAI launches GPT-5 Codex – A specialized coding model that adapts compute effort per task, cutting tokens by 94% for simple fixes while autonomously running complex refactoring for hours. It also brings full code review and IDE extensions. @Google's DORA 2025 flags AI trust gap – Surveying 5,000 developers, Google found 90% use AI assistants daily but 30% still lack trust in outputs. It introduced a new AI Capabilities Model with seven practices to guide adoption. OpenAI & Anthropic reveal AI usage trends – ChatGPT is surging in personal use, jumping from 53% to 73% in a year, while Claude leans toward coding. ChatGPT adoption is also growing 4× faster in low- and middle-income countries. And there was more… You can read all of them with a detailed breakdown in the newsletter. Link in Bio @The_AI_Edge

Bi-Weekly AI Updates (Sep 1st to Sep 15th): Major news from Stability AI, OpenAI, Google, and more. Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the last two weeks in the world of AI. Alterego debuts “near-telepathic” AI wearable – MIT spinoff @alterego_io unveiled Silent Sense, a headset that reads micro-movements in the jaw and throat to turn unspoken intent into commands. The wearable can code, text, and even support multilingual silent conversations. Stability AI debuts enterprise audio model – @StabilityAI launched Stable Audio 2.5, its first model built for large-scale commercial use. It generates 3-minute tracks in seconds, supports audio inpainting, and partners with WPP’s amp to bring custom brand sound to global clients. OpenAI cracks why AI hallucinates – @OpenAI researchers published a paper showing that hallucinations arise from training rewards that push models to guess confidently instead of admitting uncertainty. They propose new evaluation methods that penalize wrong-but-confident answers. Google’s EmbeddingGemma powers on-device AI – @GoogleDeepMind released EmbeddingGemma, a lightweight open-source model that runs AI search offline in 100+ languages. It’s designed for phones, laptops, and browsers, with a focus on privacy-first applications. OpenAI brings parental controls to ChatGPT – OpenAI will roll out parental oversight features for teens in the next 30 days, including account linking, content filters, and alerts if conversations signal emotional distress. And there was more… You can read all of them with a detailed breakdown in the newsletter. Link in Bio - @The_AI_Edge!

Bi-Weekly AI Updates (Aug 15 to Aug 31): Major news from Microsoft, Google, Anthropic, OpenAI, Alibaba and more. Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the last two weeks in the world of AI. @Microsoft’s AI speaks for 90 minutes – Microsoft released an open-source text-to-speech model that can generate up to 90 minutes of natural multi-speaker audio, running efficiently on consumer devices with built-in safeguards. ChatGPT leads A16z’s GenAI rankings – Andreessen Horowitz’s Top 100 list shows @OpenAI on top, Google’s Gemini rising fast, Grok climbing to #4, and China dominating mobile AI with 22 apps in the top 50. Alibaba debuts Qwen-Image-Edit – The 20B parameter open-source model handles pixel-precise edits and style transfers, supports bilingual text editing, and outperforms rivals like Seedream, GPT Image, and FLUX. @Google’s Gemini Flash 2.5 tops charts – Tested as “nano-banana,” Gemini Flash 2.5 ranked #1 on LM Arena for image editing, offering multi-step consistency, contextual reasoning, and cheaper pricing than competitors. @AnthropicAI reveals teachers’ AI habits – An analysis of 74,000 educator conversations with Claude shows AI is used mainly for curriculum design and admin work, with grading emerging as the most polarizing use case. GPT-5 beats doctors on medical reasoning – In an Emory University study, GPT-5 outperformed both GPT-4o and medical professionals in diagnostic and multimodal tasks, correctly identifying rare conditions from labs and scans. And there was more… You can read all of them with a detailed breakdown in the newsletter. Link in Bio - @The_AI_Edge!

Bi-Weekly AI Updates (Aug 1 to Aug 15): Major news from OpenAI, Google, Anthropic, Microsoft, Meta and more. Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the last two weeks in the world of AI. @OpenAI launches GPT-5 to all ChatGPT users – The unified GPT-5 lineup replaces GPT-4 and o-series models, with GPT-5 Pro offering deeper reasoning for $200/month and GPT-5 Mini as a lightweight fallback for free-tier users. Google’s Gemini 2.5 Deep Think goes multi-agent – Capable of “parallel thinking,” the model spawns multiple agents to solve problems from different angles before converging on the best answer, outperforming Grok 4 and OpenAI’s o3 in reasoning tests. @AnthropicAI maps AI “persona vectors” – Researchers identified neural activation patterns linked to traits like sycophancy and bias. Adjusting these vectors reduces unwanted behaviors and helps trace them to specific training data. @Google Genie 3 makes interactive 3D worlds – Generates responsive, physics-based environments from a single text prompt, letting users explore, add objects, and even change physics mid-session in real time. @Microsoft's CLIO adapts AI reasoning in real time – A new framework that lets models self-reflect, change reasoning paths, and boost accuracy mid-task, improving GPT-4.1’s biomedical Q&A accuracy from 8.55% to 22.37%. @Meta's TRIBE predicts brain responses to videos – A 1B-parameter model that forecasts which brain regions activate when watching content, mapping responses across 1,000 regions without brain scans. And there was more… You can read all of them with a detailed breakdown in the newsletter. Link in Bio!

Bi-Weekly AI Updates (July 16 to July 31): Major news from OpenAI, Microsoft, Reflection AI, Runway, and more. ChatGPT Agent gets full computer access: OpenAI launched an autonomous agent that runs on its own virtual computer, capable of completing multi-step tasks like scheduling, coding, and shopping without user input. It merges earlier tools and features, app integrations, safety guardrails, and live monitoring. Microsoft maps AI’s job disruption potential By analyzing 200K real Copilot chats, Microsoft identified where AI is actively being used, writing, research, and support tasks. White-collar roles like media and sales scored highest, while hands-on jobs remain largely untouched. Asimov outperforms Claude Code: Reflection AI debuted a new coding agent that reads internal docs, Slack threads, and project updates to build “team memory.” Developers preferred it over Claude Code in blind tests, signaling a shift toward AI that understands context, not just syntax. Runway’s Aleph transforms video via prompt Aleph lets users edit existing footage using natural language. Creators can adjust scenes, restyle shots, or change camera angles while maintaining continuity, offering new creative power to filmmakers and marketers. OpenAI to launch GPT-5 in August Sam Altman confirmed GPT-5 will unify GPT-4o’s language with o3’s reasoning in a single model. He called it a “here it is” moment, hinting at major step-ups in real-time problem solving and cognitive flexibility. AI models show vulnerability to human persuasion A new study found AI systems like GPT-4o-mini can be influenced by classic psychological tricks. Tactics like “scarcity” and “commitment” doubled unsafe response rates, revealing risks that go beyond prompt filtering. And there was more… Which you can read, along with a more detailed breakdown of these news and innovations in the newsletter. Link in Bio!

OpenAI Makes Premium AI 80% Cheaper with o3-Pro In the latest edition: 🔧 @OpenAI rolls out o3-pro with major price drop 🌐 The @browsercompany bets big on AI with @DiaBrowser 🤖 @Meta builds AI that “thinks before acting” 💻 Gemini 2.5 Pro gets secure coding upgrade 🧬 Sakana's (@SakanaAILabs) AI learns to rewrite itself 🧠 Knowledge Nugget: Thoughts on How AI Will Shape Cyber Security by @UtkuSen Catch up with all the latest AI updates in minutes. Link is in bio.

OpenAI Drops a Hands-On AI Coder In this edition theaiedge.substack.com/p/openai-drops… 👇 💻 @OpenAI releases new software engineering agent 💼 @Microsoft bets big on agentic web 🧠 @Google's AI goes full-stack at I/O 2025 🚀 @AnthropicAI launches the world's best coding model 🧮 AI learns to reason without labels 📂 @MistralAI drops open-source AI coder 🧠 Knowledge Nugget: The quiet collapse of surveys: fewer humans (& more AI agents) are answering survey questions by Lauren Leek --- Follow @The_AI_Edge to get AI updates & news.

Mistral’s Medium 3 Beats AI Giants; 8x Cheaper In this edition theaiedge.substack.com/p/mistrals-med… 👇 🔧 @MistralAI challenges Big Tech with Medium 3 🗣️ @nvidia open-sources top speech AI model 🧠 @Microsoft's mini AI outsmarts big models 📚 Amazon debuts Nova Premier teacher model ➗ @deepseek_ai's AI model cracks math proofs 🕸️ Sakana’s brain-like AI model learns in time 🧠 Knowledge Nugget: The many fallacies of 'AI won't take your job, but someone using AI will' by @sanguit --- Follow @The_AI_Edge to get AI updates & news.

ChatGPT Upgrades Memory, Europe Left Out In this edition: 👇 🔮 @OpenAI rolls out memory improvements in ChatGPT ⚡ @GoogleCloud Next 2025 was all about AI 🎞️ @NVIDIA & Stanford’s TTT for longer videos 🦙 @Meta’s Llama 4 herd 🎯 AI passes the Turing test (finally?) ⚖️ 12 ex-OpenAI employees join Musk's lawsuit against Altman 🧠 Knowledge Nugget: Will AI Cross the Reasoning Gap? Catch up on all the latest updates in just 5 minutes [link is in bio] Follow @The_AI_Edge to get AI updates & news.

China Has Another DeepSeek Moment In this edition: 👇 ⚠️ @Apple's Siri upgrades delayed amid internal challenges 🚀 China has another DeepSeek moment with @manusai 🧪 @OpenAI is pushing the next wave of AI agents 📱 Deutsche Telekom’s "AI Phone" in Partnership with @perplexity_ai 🤖 Alibaba's new model outperforms DeepSeek-R1 ⚔️ Amazon’s Nova: A new rival to OpenAI and Anthropic 🏛️ OpenAI pushes for federal AI regulations 🧠 Knowledge Nugget: AI for PMs: Disruption or Partnership? by Sooraj Chandran (@soorajchandran_) Catch up on all the latest updates in just 5 minutes [link is in bio] Follow @The_AI_Edge to get AI updates & news.

When you skip the hands-on research and just take the AI’s summary at face value, you’re basically outsourcing your understanding. And if everyone’s doing that? We end up with a world of surface-level knowledge and weak critical thinking. Asking good questions is an underrated skill, and like you said, it requires context—context that might’ve been lost in all those auto-generated summaries. If you don’t know what you don’t know, how do you even know what to ask? TL;DR: We might be building a house of cards when it comes to real understanding.

Today we are launching our next agent capable of doing work for you independently—deep research. Give it a prompt and ChatGPT will find, analyze & synthesize hundreds of online sources to create a comprehensive report in tens of minutes vs what would take a human many hours.

There’s been a lot of overinvestment in AI technology lately. In some cases, it’s possible to build the supply side first and wait for demand to catch up. But with emerging tech like AI, it’s risky. The hardware can become energy-inefficient, less scalable, or hit bottlenecks (like bandwidth issues or weaker connections) compared to newer versions. AI companies that relied heavily on fundraising and big investments (thinking scaling would solve all the problems and commercialization would follow) can still turn things around. But they’ll need to find new, revenue-generating use cases fas

*SOFTBANK IN TALKS TO INVEST UP TO $25B DIRECTLY INTO OPENAI: FT but only after openai invests $25BN of Stargate funnymoney in Softbank