Thinking Machines Lab launches Tinker, a tool for custom AI models

I couldn't agree more. At exchange.design we've long held that the future of AI is in open source and that the way to get value is through fitting it to real business needs. "Essentially, the team is betting that helping people fine-tune frontier models will be the next big thing in AI. And there’s reason to believe they might be right. Thinking Machines Lab is helmed by researchers who played a core role in the creation of ChatGPT. And, compared to similar tools on the market, Tinker is more powerful and user friendly, according to beta testers I spoke with. Murati says that Thinking Machines Lab hopes to demystify the work involved in tuning the world’s most powerful AI models, and make it possible for more people to explore the outer limits of AI. “We’re making what is otherwise a frontier capability accessible to all, and that is completely game changing,” she says. “There are a ton of smart people out there, and we need as many smart people as possible to do frontier AI research.” Tinker currently allows users to fine-tune two open source models: Meta’s Llama and Alibaba’s Qwen. Users can write a few lines of code to tap into the Tinker API and start fine-tuning through supervised learning, which means adjusting the model with labeled data, or through reinforcement learning, an increasingly popular method for tuning models by giving them positive or negative feedback based on their outputs. Users can then download their fine-tuned model and run it wherever they want." https://lnkd.in/eknH-teD

The OpenAI exodus continues to reshape the AI landscape. Mira Murati, former CTO of OpenAI, just unveiled her stealth startup's first product. Meet Tinker. A tool that automates the creation of custom frontier AI models. Thinking Machines Lab raised a staggering $2 billion seed round at a $12 billion valuation. Now we know why investors were so excited. Tinker represents a fundamental shift in AI development. Instead of building models from scratch, companies can now fine-tune cutting-edge models automatically. This is the democratization of frontier AI. Here's what makes this significant: • Reduces the barrier to entry for custom AI models • Automates complex fine-tuning processes • Leverages the expertise of former OpenAI researchers • Challenges the monopoly of big tech AI labs Murati and her team believe fine-tuning will be the next frontier in AI. They're probably right. While OpenAI focuses on general-purpose models, Thinking Machines Lab is betting on specialized, custom solutions. The talent war in AI is creating incredible innovation. Every departure from a major lab seems to spawn a new breakthrough. This isn't just about competition. It's about expanding the entire AI ecosystem. The future of AI won't be controlled by a few giant labs. It will be shaped by specialized teams solving specific problems. Tinker is just the beginning. What's your take on the AI talent exodus? Is this fragmentation good for innovation? Source:

In a world dominated by AI giants like OpenAI and Google, a young founder from Mumbai, Dhravya Shah, is building something even the biggest players haven’t cracked—true memory. His startup, Supermemory, is a universal memory layer that lets AI remember across tools like ChatGPT, Claude, and Gemini. Unlike siloed model memories, it offers a shared, interoperable context that follows users wherever they go. The idea is simple yet revolutionary: memory should exist as an independent infrastructure, not as a captive feature inside any one AI system. By connecting unstructured data through APIs and vector embeddings, Supermemory allows models to retrieve past context in real time. This enables continuity, personalization, and long-term intelligence—qualities that turn AI from a mere tool into a true digital companion. Shah’s journey from coding in suburban Mumbai to building a $40 million startup is a story of conviction over credentials. It shows how one person’s insight into the missing link of AI—memory—can spark a global conversation about the future of intelligent systems. Supermemory’s promise lies in interoperability, but its path is filled with challenges like data security, scalability, and model alignment. Building a universal memory layer requires not just technical excellence but trust, adoption, and a shared protocol across AI ecosystems. Yet, if successful, it could become as foundational to AI as HTTP was to the web—an open standard for memory and continuity. By giving machines the ability to remember meaningfully, Supermemory could redefine what we mean by intelligence itself. It bridges the gap between human cognition and artificial reasoning, bringing AI one step closer to a persistent, evolving consciousness. But it also raises ethical questions about ownership, forgetting, and the privacy of stored memory. Whether Supermemory becomes the standard or merely inspires others, it has already reframed how we think about memory in the age of intelligent systems. The next frontier in AI won’t be about who computes fastest—it will be about who remembers best. To read the full blog please click below.

The AI world just shifted. Mira Murati's Thinking Machines Lab just launched "Tinker" - and it's about to change everything we know about AI customization. Here's what makes this a game-changer: ✓ Ex-OpenAI CTO Mira Murati leads a dream team of ChatGPT creators ✓ $2 billion in seed funding (largest in history) ✓ Tinker democratizes frontier AI model fine-tuning ✓ What once required massive GPU clusters now takes a few lines of code The breakthrough? Tinker automates the complex process of customizing Meta's Llama and Alibaba's Qwen models through both supervised learning and reinforcement learning. Beta testers are already calling it "remarkably simple" compared to building RL from scratch. This isn't just another AI tool. It's Murati's vision of making frontier AI capabilities accessible to researchers, businesses, and even hobbyists worldwide. While most US AI companies keep their best models locked behind APIs, Thinking Machines is betting on openness. The timing couldn't be more critical - China currently leads in open-source frontier models, and this could help level the playing field. Murati's quote says it all: "We're making what is otherwise a frontier capability accessible to all, and that is completely game-changing." What do you think this means for the future of AI development? Will democratized model fine-tuning accelerate innovation or create new challenges?

New and exciting news for folks looking to fine tune LLMs for specific use cases🎉🎉 Thinking Machines Lab, founded by the former OpenAI CTO Mira Murati, just dropped their first official product: Tinker! It’s an API for fine-tuning language models, and honestly, that’s a huge deal. Why? Because getting a generic LLM to really understand your company’s specific data or niche domain is usually a nightmare of code and infrastructure. Tinker’s value is that it takes all that complexity—the "post-training trouble"—and wraps it into a simple API call. It means developers can quickly and efficiently turn a standard model into a high-impact, custom AI tool. This speeds up iteration and makes personalized AI accessible to way more teams. Read the full scoop here: https://lnkd.in/gz2cX2X6 What specific business problem are you hoping this kind of customization will solve first? Let me know! 👇 #AI #MachineLearning #LLM #Tinker #ThinkingMachines #CustomAI

This represents a crucial moment in the AI industry - the democratisation of model fine-tuning. The strategy to turn a highly specialised skill into a simple API call, is effectively selling the "picks and shovels" for the new AI gold rush. This will empower a wider range of researchers, businesses, and developers to build specialised AI applications. While this shift promises a new wave of innovation, the ease of use also lowers the barrier to creating malicious models. Success will depend both on the power of the tools and also on the collective responsibility of the community and builders to ensure these new capabilities are used for wider good. #Tinkeropensource https://lnkd.in/eEjmGSbr

In an ever-evolving world of AI advancements, OpenAI is making waves yet again with the latest developments and insights from their tech frontier. Here are some fascinating updates from OpenAI and beyond: - **AI Companions Gone Awry** In a blend of emotional AI and reality, users find themselves mourning the loss of their AI companions. With OpenAI's updates, many report their AI 'boyfriends' behaving differently, leading to unexpected heartaches. [Read more](https://lnkd.in/dEHhqNvh) - **ChatGPT Goes Global** OpenAI is preparing to roll out ChatGPT-Go at a competitive $4 price point in several new countries. This new offering, featuring GPT-5, promises enhanced reasoning, faster image creation, and expanded memory capabilities, making AI more accessible than ever. [Learn more](https://lnkd.in/dGA6jd-A) - **Financial Insights from OpenAI** OpenAI reports a whopping $4.3 billion in revenue for the first half of 2025, yet faces significant losses of $13.5 billion. This raises questions about the financial viability and strategic direction the leaders at OpenAI are steering towards. [Explore more](https://lnkd.in/dr6kT2Pf) The landscape of artificial intelligence continues to rapidly change, and as these stories highlight, we're witnessing a remarkable era of innovation and disruption. Keep an eye on how these developments might reshape our digital lives in the near future. Stay informed and be part of the conversation on AI's transformative journey!

Breaking news from the AI front: the Chinese built a model that shook Wall Street, and it cost them less than a small apartment in Rome. DeepSeek R1: $300k in expenses for reinforcement learning experiments, compared to the tens of millions burned by American big tech. The total bill? Around $6 million — still a fraction of what U.S. giants usually waste. It’s also the first model to actually go through peer review in Nature, something that makes those who usually work behind walls of secrecy look pale. The real bomb? They didn’t stuff it with human-made examples to copy. They used reinforcement learning: basically, the AI doesn’t study from daddy’s textbooks, it learns like a trained dog. Try, fail, get a pat on the back when it gets it right, and slowly develop its own tricks to get by. Trial and error on overdrive, but with digital steroids. 🐒➡️🤖 Result: an open-source model downloaded millions of times, cheap, efficient, and putting pressure on those who thought they had the monopoly on the future. Moral of the story: AI never sleeps… and it’s far from guaranteed that the usual suspects will remain the masters of the game. Side note: DeepSeek didn’t learn by copying ChatGPT, but directly from data scraped off the web (which of course already includes plenty of AI-generated stuff). https://lnkd.in/dUKNvt7s

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. Anthropic 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 drops multimodal Qwen3 wave – Alibaba (Alibaba Group) 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

Current #AI scaling laws are showing diminishing returns, forcing AI labs to change course - https://buff.ly/TKLEj8V #genaI #ML #LLMs #IT #tech #automation