ChatGPT Atlas (4 minute read)
OpenAI introduced ChatGPT Atlas, an AI-powered web browser that integrates ChatGPT into the browsing experience. Atlas includes a sidecar chatbot panel with contextual awareness of what's on screen and tracks browsing history to personalize responses. |
Anthropic, Google in Talks on Multibillion-Dollar Cloud Deal (2 minute read)
Anthropic is in early-stage talks with Google about a deal that would provide it with additional computing power valued in the high tens of billions of dollars. The plan involves Google providing cloud computing services to Anthropic, giving Anthropic access to Google's tensor processing units, chips custom-designed to accelerate machine learning workloads. Google is a previous investor in and a cloud provider for Anthropic. |
BlackRock Among Biggest Investors in Meta's Giant Data-Center Debt Deal (2 minute read)
Meta and private-credit firm Blue Owl raised $27 billion to finance the buildout of a data center in Louisiana. BlackRock was one of the biggest investors in the deal. It purchased more than $3 billion in bonds last week to finance the deal. The project is 80% owned by Blue Owl, with Meta owning the remaining stake. The deal is the largest private-debt offering so far. |
|
On Dwarkesh Patel's Podcast With Andrej Karpathy (50 minute read)
Andrej Karpathy recently appeared on a podcast where he discussed a wide range of topics, including AGI, self-driving, the evolution of intelligence and culture, and more. This article breaks down some of the key points made and adds commentary. A link to other reactions to the podcast is provided. A video of the original podcast is available. |
The Continual Learning Problem (22 minute read)
Memory layers are a natural architecture for models that are always training and continually learning from experience over time. Finetuning memory layers enables learning without forgetting much more effectively than LoRA. Continual learning and memory is still a huge and rich space to explore. We are only just starting to reach the point where we can imagine what it might look like to have models that learn online in the real world. |
Mini Models Battle: Claude Haiku 4.5 vs GLM-4.6 vs GPT-5 Mini (8 minute read)
Claude Haiku 4.5, GLM-4.6, and GPT-5 Mini all fall within a similar price bracket. They are mid-tier models optimized for cost and speed rather than maximum capability. GPT-5 Mini delivers the strongest concurrency safeguards, Haiku 4.5 is the fastest with the most features, and GLM-4.6 produces multi-file architecture but requires disabling reasoning mode for tool calls. Each model makes different trade-offs. |
|
Google revamps AI Studio with new features for vibe coding (2 minute read)
Google has revamped its user experience for AI Studio. The update, targeted at developers, startups, and non-coders interested in rapid AI prototyping, helps users build AI-powered applications with less friction from prompt to production. Apps can be easily deployed to Google Cloud Run via a single click. An animated preview of the new interface is available in the article. |
BOLT - How Mura wrote an in-house LLM Eval Framework (11 minute read)
BOLT is an evaluation system built by Mura. It has become critical infrastructure for how the company ships improvements, migrates between models, and maintains accuracy as it grows. Mura is a seed-stage startup that helps commercial HVAC providers get paid faster without hiring more back-office staff. It turns messy inputs into structured data using AI and orchestration, and then presents suggestions for users to approve. This article talks about how Mura built BOLT. |
The Not-so Bitter Lesson (8 minute read)
Maybe AGI will take over everything, but we're clearly not there yet and probably won't be anytime soon. Software engineers still have a lot of time before they become obsolete. Right now, developers should focus on building scalable systems that leverage search and compute, build harnesses instead of solutions, and define problems instead of hard-coding solutions. |
|
OpenAI has an army of ex-investment bankers making financial models to train ChatGPT (1 minute read)
OpenAI has hired over 100 former investment bankers to help teach OpenAI's models. They are getting paid $150 an hour to do the grueling work of junior bankers, including tweaking PowerPoint slides and billing financial models in Excel. This shows that OpenAI is looking for opportunities in the business world. The company's consumer subscriptions currently account for 70% of its $13 billion annual recurring revenue. |
Airbnb CEO says ChatGPT isn't ready (3 minute read)
Airbnb's CEO, Brian Chesky, said that he didn't integrate Airbnb's travel app with OpenAI's ChatGPT because ChatGPT's connective tools aren't quite ready yet. Airbnb will monitor the development of ChatGPT's app integrations and may consider adding them in the future. Chesky, who is close friends with OpenAI CEO Sam Altman, said he advised OpenAI on its new capability for third-party developers to make their apps available within ChatGPT. Airbnb on Tuesday announced new AI tools to let customers take more actions without the need for a live representative. The company's AI customer service agent now displays action buttons and links that can help people complete reservation changes and cancellations. |
|
| Love TLDR? Tell your friends and get rewards! | | Share your referral link below with friends to get free TLDR swag! | | | |
| Track your referrals here. |
| Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of AI professionals and decision makers, 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, Andrew Tan, Ali Aminian, & Jacob Turner
|
|
| |
0 Comments