Chrome turns 15 and is getting a big redesign (5 minute read)
Chrome is getting a big redesign to celebrate its 15th birthday. The redesign will bring Google's 'Material You' design language to Chrome. Other updates from the latest version of Chrome include a more comprehensive menu, an upgraded Safe Browsing blocklist, and AI in the sidebar. Screenshots of the redesign are available in the article.
ChatGPT traffic slips again for third month in a row (3 minute read)
ChatGPT's traffic has declined for the third month. There are signs that the decline is coming to an end. School coming back into session in September may have boosted ChatGPT's traffic and usage. ChatGPT is still one of the top 30 websites in the world.
Every Software Project is a Startup That Will Probably Fail (3 minute read)
Most software will never produce more value than it cost to develop.
An article about Mustafa Suleyman, one of the co-founders of DeepMind, and his views on how AI will affect our lives in the future.
At least 5 people have been cured of HIV. Is the AIDS pandemic ending? (10 minute read)
The life expectancy of someone in the US infected with HIV is currently about the same as the rest of the population. This is largely due to antiretroviral therapy, a daily treatment regimen that decreases the level of HIV in a person's blood to the point it's no longer detectable or transmittable. There have been a handful of people over the last couple of decades that have been able to reach undetectable HIV levels while off treatment, essentially meaning they've been cured of the disease. New universal cures are now on the horizon.
Bun is an all-in-one toolkit for running, building, and debugging JavaScript and TypeScript. Version 1.0 is now available. Bun is a drop-in replacement for Node.js and a JavaScript bundler with an esbuild-compatible plugin API. It also functions as an npm-compatible package manager and a Jest-compatible test runner. Bun can run virtually any Node.js application in the wild.
Nvidia Says New Software Will Double LLM Inference Speed On H100 GPU (7 minute read)
Nvidia says that TensorRT-LLM, an open source software library, will double the H100's performance for running inference on leading large language models. The software is now available in early access. It will be integrated into the Nvidia NeMo LLM framework as part of the Nvidia AI Enterprise software suite when it is released to the public next month. TensorRT-LLM will also support Nvidia's A100, L4, L40, L40, and Grace Hopper Superchip GPUs.
7 simple habits of the top 1% of engineers (7 minute read)
This article looks at the habits of successful engineers and explains how these habits make those engineers more successful. Examples include sticking to a consistent standard and style of coding to make code easier to read and understand, deep domain knowledge in at least one field, and good self-marketing. The article delves deep into coding habits and how to write good code.
Why Nvidia’s AI Supremacy is Only Temporary (9 minute read)
Nvidia's top spot in the AI world is far from secure. It is currently winning as almost nobody is running large machine learning apps, the competition hasn't matured, researchers have the purchasing power, and because of the focus on training models. In the near future, inference with CPUs will dominate over training with GPUs, engineers will have more purchasing power, and the focus will be on application costs. The winners of this shift will likely be traditional CPU platforms like x86 and Arm.
Death by a thousand microservices (15 minute read)
There is no need for every service to be 'micro'. Some startups do fine with just 'services'. Starting with a monolith is fine. Services can be branched from the monolith if they take care of a clearly-identifiable and separately-scalable load. A simple approach is often a more appropriate choice.
China AI & Semiconductors Rise: US Sanctions Have Failed (21 minute read)
US sanctions aimed at limiting Chinese firms' ability to manufacture high-end chips have clearly failed as China is rapidly developing exascale supercomputing capabilities - the country has announced its intent to become the world leader in AI by 2030.
SpaceX’s Starlink made $1.4 billion last year (2 minute read)
Starlink generated $1.4 billion in revenue last year, up $222 million from 2021 but $11 billion short of its original projections. The company is no longer incurring losses from producing satellite antennas. Elon Musk initially predicted in 2015 that Starlink would have 20 million subscribers by the end of 2022. The service reported 1.5 million users in May 2023. Starlink was losing about $20 million a month maintaining its services in October 2022.
Moderna CEO on AI’s potential to cure cancer (9 minute read)
Moderna was using AI to create new drugs years before ChatGPT came out. The technology was used to create drugs like its Covid-19 vaccine. Moderna's experimental mRNA flu vaccine did well enough in a Phase 3 study that it could potentially gain regulatory approval. It is now planning a Phase 3 trial of an mRNA-based therapy for a certain type of lung cancer. This article contains an interview with Moderna's CEO where they discuss AI, management challenges that the tech poses, and how artificial intelligence is being used to help find a cure for cancer.