TLDR 2024-09-13

Attend live or get free access to the recording

OPENAI RELEASES O1, ITS FIRST MODEL WITH ‘REASONING' ABILITIES

OpenAI has released o1 and o1-mini, the first in a series of reasoning models that have been trained to answer more complex questions faster than a human can. The model is better at writing code and solving multistep problems than previous models, but it is more expensive for developers and slower to use than GPT-4o. The release is still in preview to indicate how nascent it is. ChatGPT Plus and Team users should already have access to the model, while Enterprise and Edu users will get access early next week. OpenAI plans to bring o1-mini access to all free users, but it hasn't set a release date yet.

SPACEX ASTRONAUTS CONDUCT SPACEWALK, PUTTING NEW SPACESUITS TO TEST

The private-citizen astronauts traveling with SpaceX have completed the first-ever commercial spacewalk. The spacewalk kicked off shortly after 6 am ET on Thursday. The livestream showed tech industry billionaire Jared Isaacman leaving the SpaceX Crew Dragon and conducting a series of mobility tests before coming back after around an hour. The entire spacecraft was exposed to the vacuum of space during the spacewalk. The crew members were wearing SpaceX-designed and manufactured suits that were previously untested in orbit. A clip from the livestream is available in the article.

FACE TO FACE WITH FIGURE'S NEW HUMANOID ROBOT

Figure's humanoid robots were put to work at BMW's Spartanburg, South Carolina plant for a couple of weeks earlier this year - they will be returning for good in January. The automotive industry has been eager to adopt the humanoid robot form factor, which is in line with its deployment of robots on factory floors. Figure aims to hit a sub-$20,000 price point, but it will take some time. In the meanwhile, the company is testing its robot's efficacy in helping out with tasks around the home. This article features a video of Figure's second-generation robot up close.

GOOGLE DEEPMIND TEACHES A ROBOT TO AUTONOMOUSLY TIE ITS SHOES AND FIX FELLOW ROBOTS

Google DeepMind researchers recently released a paper showcasing a method for teaching robots to perform a range of dexterous tasks. It used a new learning platform called ALOHA Unleashed paired with a simulation program, DemoStart, to teach robots by watching humans. The research showcased how systems can learn to perform complex, dexterous tasks from a visual demonstration. Videos of robots trained with the method performing tasks are available in the article.

YOUR NEXT STORE

Your Next Store is a modern ecommerce site template in Node.js that uses Stripe as the backend. It gets all of the products, prices, descriptions, and categories from Stripe, making it easy for users who know Stripe already to use. Your Next Store supports simple product variants, Stripe Webhooks, Stripe Tax, and more. A demo site and video are available.

HOW GITHUB TAUGHT ME TO MICROMANAGE

Good feedback cultivates quality work and professional growth - bad feedback degrades quality and erodes relationships. Bad feedback makes critiques but doesn't offer a better direction. OK feedback provides direction. Good feedback educates to produce a better result. All of this depends on context - detailed feedback may be welcomed in circles where people are trying to improve others around them, but it can be toxic in environments where few people care about craft.

THE SUCCESS OF BLACK MYTH: WUKONG AND WHAT IT MEANS FOR MEDIA

Black Myth Wukong is one of the fastest-selling games of all time. The game cost $70 million to make and has grossed close to $1 billion so far. The rise of regionally rooted AAA video games has been slower than in film due to their comparatively high production cost, the rate of technological, creative, and change, and a smaller market. The growing capability and ease of use of game engines as well as the affordability and size of asset store libraries will result in a growing number of hits that target non-Western regions, countries, and cultures, some of which will go far beyond those markets.

THE BUTTON PROBLEM OF AI

Productivity software tends to buttonify AI into small, discrete tasks because most models are currently incapable of building and following complex plans as independent agents. Project builders are forced to bind models to tasks that are small and easily definable because they aren't able to handle knowledge-labor workflows in their totality. Simply replacing small portions of a workflow with a button isn't enough to meaningfully change the work itself. The buttonification of AI is a reminder that we are still in the 'training wheels' phase of the supposed AI revolution.

WHY NASA IS STICKING WITH BOEING

Dumping Boeing would mean that NASA will only be left with SpaceX.

WHY IS LANGUAGE DOCUMENTATION STILL SO TERRIBLE?

It would help a lot for language documentation to be designed to be read and navigated instead of the mess that most languages currently have.

SILICON VALLEY'S Y COMBINATOR TO DOUBLE NUMBER OF COHORTS PER YEAR

Y Combinator is expanding to four cohorts a year, meaning the accelerator will be in session for almost the entire year.

THE BRAIN AGED MORE SLOWLY IN MONKEYS GIVEN A CHEAP DIABETES DRUG

Monkeys who received metformin daily showed slower age-associated brain decline than those who didn't in a study that tracked the animals for more than three years.

SOMETHING NEW: ON OPENAI'S "STRAWBERRY" AND REASONING

A hands-on look at OpenAI's new o1-preview model, which can 'think through' problems before solving them, allowing it to address very hard problems that require planning and iteration.

MICROSOFT TO CUT 650 JOBS AT ITS XBOX GAMING UNIT

The job cuts are aimed at aligning Microsoft's post-acquisition team structure following its acquisition of Activision Blizzard.