OpenAI's IMO Gold 🥇, Zuckerberg's recruiting 💰, against AI agents 👨‍💻

An experimental reasoning model from OpenAI has achieved gold medal-level performance on the International Math Olympiad (IMO) ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With TLDR 2025-07-21 Claude Code is the terminal-native coding agent for devs who want to ship faster (Sponsor) Serious work happens in the terminal. Claude Code lets you unleash Anthropic's most powerful models right from your command line. 4 ways you can use Claude Code right now (see examples): > Natural language → features: Claude writes the code, tests the code, and ships. > Autonomous debugging: Let Claude analyze your codebase, identify the problem, and implement a fix. > Codebase onboarding: Claude searches, navigates, and explains entire codebases in a few seconds. > Automate the tedious: — fiddly lint issues, merge conflicts, release notes. Add it to your CI pipelines for peak automation. Try Claude Code on a Max subscription → 📱 Big Tech & Startups OpenAI experimental reasoning LLM has achieved a longstanding grand challenge in AI (4 minute read) An experimental reasoning model from OpenAI has achieved gold medal-level performance on the International Math Olympiad (IMO). IMO problems demand a new level of sustained creative thinking compared to past benchmarks. Submissions for the challenge are hard-to-verify, multi-page proofs. The experimental model solved five of the six problems on the 2025 IMO. OpenAI plans to release GPT-5 soon, but it won't be releasing anything with the experimental model's level of math capability for several months. The Epic Battle for AI Talent—With Exploding Offers, Secret Deals and Tears (14 minute read) The war among some of the biggest companies on the planet for talent is playing out in an unprecedented frenzy of talent raids, secret deals, and betrayals. AI researchers whose minds have never been so highly valued are set to become as rich as NBA players and Hollywood stars. Some researchers are being offered pay packages worth more than $300 million, and even that kind of money isn't always enough to win them over. The social contract that once united founders and employees may be unraveling. 🚀 Science & Futuristic Technology A New Frontier in Fusion Technology (4 minute read) Marathon Fusion has announced a solution to the transmutation of gold using a method that is massively scalable, pragmatically achievable, and economically irresistible. Its approach can generate 5,000 kilograms of gold per year per gigawatt of energy generation without compromise to fuel self-sufficiency or power output. It uses the neutrons that drive the multiplication reactions in deuterium-tritium fusion on mercury-198 to produce mercury-197, which decays in a few days to the only stable isotope of gold. The method could also be used to create other precious metals, radically transforming the economics of fusion and energy. Cancer DNA is detectable in blood years before diagnosis (3 minute read) It is possible to spot tumor DNA more than three years before a cancer diagnosis. These telltale traces could be a powerful tool in early cancer screening efforts. The technology could help doctors detect cancers before any other signs or symptoms of the disease appear and dramatically change outcomes for patients. 💻 Programming, Design & Data Science 😘 Kiss bugs goodbye with fully automated end-to-end test coverage (Sponsor) QA Wolf's AI-native service gets web and mobile apps to 80% automated test coverage in less than 4 months. They create and maintain your test suite in open-source Playwright. Plus, they provide unlimited parallel test runs on their infrastructure (24-hour maintenance included). The result? Salesloft saves $750k/year in QA engineering + executes 300+ tests in parallel on every PR in minutes. ⭐ Rated 4.8/5 on G2. Trusted by Cohere, AutoTrader, Mailchimp, and many others. Schedule a demo to learn more Coding with LLMs in the summer of 2025 (an update) (6 minute read) Software developers can maximize their impact by using large language models (LLMs) in an explicit way while staying in the loop. This allows them to do things that are otherwise at the borders of their knowledge/expertise while learning in the process. It is wise to test out the capabilities of agents from time to time, as the technology will improve and eventually many coding tasks will be better served by AI alone. Those who avoid using LLMs due to ideological or psychological refusal will be at a disadvantage as they fail to develop the set of skills needed to work with the technology. Rethinking CLI interfaces for AI (6 minute read) Every command-line interface (CLI) can be improved to provide extra context to large language models (LLMs). Doing this reduces tool calls and optimizes context windows. Agents may benefit from training on tools available within their agents. Developers may benefit from a whole set of AI-enhanced CLI tools or a custom LLM shell. 🎁 Miscellaneous How YouTube Won the Battle for TV Viewers (13 minute read) YouTube became the most-watched video provider on televisions in the US earlier this year. People now watch YouTube on TV sets more than on their phones or any other device. In response, creators are making longer, higher-quality videos that appeal to families and groups of friends watching in their living rooms. This article looks at the history of the platform and the secrets to its success. Why Banks Are on High Alert About Stablecoins (5 minute read) The House recently voted to pass a bill that spells out some ground rules for stablecoins. Stablecoins have the potential to lure away customer deposits from banks. The Genius Act creates a regulatory framework where stablecoins are supposed to maintain a 1:1 ratio with the US dollar or other fiat currencies so that they can easily be used for payments. This could make stablecoins more attractive to use, especially for cross-border payments, which can take days to settle and are subject to interchange and other fees. ⚡ Quick Links In recent layoffs, AI's role may be bigger than companies are letting on (9 minute read) Firms are limiting their explanations to terms like reorganization, restructuring, and optimization, but that terminology could be AI in disguise. It's rude to show AI output to people (5 minute read) AI output should only be relayed if it's either adopted as your own or there is explicit consent from the receiving party. Netflix's first show with generative AI is a sign of what's to come in TV, film (5 minute read) The Eternaut, which premiered in April, used generative AI footage during a scene that showed a building collapsing. Amazon's AWS cloud computing unit cuts at least hundreds of jobs (2 minute read) Amazon CEO Andy Jassy is reducing what he describes as an excess of bureaucracy at the company. Why I'm Betting Against AI Agents in 2025 (Despite Building Them) (10 minute read) Error rates compound exponentially in multi-step workflows, context windows create quadratic token costs, and designing tools and feedback systems that agents can actually use effectively is a huge challenge. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/7c97ce4b/ Track your referrals here. Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of tech executives, decision-makers and engineers, you may want to advertise with us. Want to work at TLDR? 💼 Apply here or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! If you have any comments or feedback, just respond to this email! Thanks for reading, Dan Ni & Stephen Flanders Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR isn't for you, please unsubscribe.