TLDR 2025-08-29
gpt-realtime 🤖, Starship economics 🚀, Grok Code 👨💻
Microsoft starts testing AI model that could escalate competition with OpenAI (3 minute read)
Microsoft has started testing its MAI-1-preview model on LMArena. MAI-1-preview is a foundation model trained end-to-end in-house by Microsoft. The model will be rolled out for certain text use cases within Copilot over the coming weeks. Developers can fill in a form to request early access.
Elon Musk reveals when SpaceX will perform first-ever Starship catch (3 minute read)
Elon Musk says that SpaceX will attempt to catch Starship on Flight 13, 14, or 15, depending on how well the V3 flights go. The V3 rocket is more powerful than V2 in terms of initial thrust and booster thrust, and it has a higher payload-to-orbit rating. V4 will likely be the first version to carry humans to space. Musk has previously said that SpaceX hopes to attempt a catch of Starship this year, but the company will need to launch two more flights before the end of the year to get to his expected first catch attempt of Flight 13.
🚀
Science & Futuristic Technology
Starship Will Reduce Bandwidth Launch Cost by up to 50x (12 minute read)
Even if fully expendable, Starship already has an around 40% lower cost per kilogram than a mature Falcon 9. With reuse, Starship's cost per kilogram quickly falls towards a $60 per kilogram floor. Starship introduces a different cost regime that will transform the industry to be supply-rich, price-flexible, and capability-driven. For Starlink, this will mean faster scale, richer service tiers, and room to price for demand and margin.
The world's largest sand battery just went live in Finland (4 minute read)
Finnish Firm Polar Night Energy has built an industrial-scale sand battery in Pornainen, Finland. The battery is set to reduce carbon emissions from the local heating network by as much as 70%. At 42 feet tall and 50 feet wide, the battery can store up to 100 MWh, enough to heat the whole town for a week. The sand battery can hold a charge for months at a time, helping balance the energy grid during periods of high demand. A video of the battery being put together is available in the article.
💻
Programming, Design & Data Science
Grok Code Fast 1 (7 minute read)
grok-code-fast-1 is a speedy and economical reasoning model that excels at agentic coding. xAI built the model from scratch using a brand-new model architecture. The model has mastered the use of common tools like grep, terminal, and file editing. It is free for a limited time for users of select launch partners, including GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf.
Introducing gpt-realtime and Realtime API updates for production voice agents (8 minute read)
OpenAI's Realtime API, powered by gpt-realtime, enables developers to build reliable, production-ready voice agents. It supports remote MCP servers, image inputs, and phone calling through Session Initiation Protocol. OpenAI has released two new voices in the Realtime API, Cedar and Marin. OpenAI's most advanced speech-to-speech model yet, gpt-realtime shows improvements in following complex instructions, calling tools with precision, and production of speech that sounds more natural and expressive.
Will all our drugs come from China? (18 minute read)
The West's big car manufacturers are facing an existential threat from Chinese manufacturers, which relatively recently began pushing the boundaries of performance, cost-efficiency, driver experience, and autonomy. Western carmakers have been content to innovate in a gradual way for far too long, and it could cease to be competitive. This could also happen to the biotech industry. There has been a steady rise in Chinese companies as a source of genuinely new drugs. Chinese companies are now responsible for about a quarter of new trial starts.
Are people's bosses really making them use AI tools? (13 minute read)
Many companies are making employees use AI tools, outsourcing responsibilities to a technology that often gets things completely wrong. AI is completely incapable of doing analytical work that requires creative and soft skills and a lot of experience. Large language models are language processing systems, so it shouldn't be surprising that they can only regurgitate generic advice. Workers stuck using AI tools need to document their use carefully so they are able to demonstrate how the tools cause problems, slowdowns, or other disappointing outcomes.
We're looking for a growth marketing lead to help grow TLDR! (Fully Remote, $150k-$200k)
TLDR is hiring a
Growth Marketing Lead with deep expertise in Meta ads to help us grow our subscriber base.
If you're excited to work with the latest AI tools, discover and scale non-obvious growth channels, and work with a fast-paced fully remote team to grow our newsletters beyond 7 million subscribers, please apply here.
Refer a friend to jobs@tldr.tech and if we hire them we'll send you $5k!
Get the most interesting stories in startups, tech, and programming delivered in a free daily email.
Join 1,600,000 readers for
one daily email