TLDR 2023-07-20

Tesla is in early discussions to license its Full Self-Driving (FSD) technology to another major automaker that has not been named. Elon Musk says that licensing the software has always been part of the plan. Tesla's FSD works on local roads with traffic signals, intersections, and vulnerable road users. It requires drivers to pay attention to the road and take control of the vehicle when asked. There are currently 16 crashes by Tesla vehicles using Autopilot being investigated by the National Highway Traffic Safety Administration, out of which only one caused a fatality.

Apple is readying its own large language model, Ajax. It is recruiting more engineers to work on generative AI. Ajax is solely for internal use at Apple. Apple staff are using it for real work like prototyping products and summarizing text. The company does not currently intend to launch the chatbot, but it is trying to find a consumer angle for it. Apple has already deeply embedded AI and machine learning technology into everything it's doing.

Botanists have been storing plant specimens from extinct species for centuries. They are now starting to realize their potential to resurrect life forms long believed lost forever. Libraries of data from herbarium specimens of extinct plants are now available in browsable digital databases. Attempts to revive extinct animals will only result in hybrid genetically-engineered animals, but successful recovery of a plant will result in the resurrection of the actual species.

Getting hearing aids could cut the risk of cognitive decline in half. Cognitive decline and dementia have become a growing problem as people are living longer. Hearing loss is one of the biggest risk factors for developing dementia. There are several reasons why hearing loss might contribute to cognitive decline, such as the brain having to work harder to decode audio signals, brain atrophy, or a lack of social activity.

Retake provides an infrastructure for vector data streams. It features out-of-the-box data sync, real-time updates, and an extensible Python development kit. Retake is stable and efficient and can be deployed anywhere. It ensures consistency between inputs, sources of truth, and vectors without the time, complexity, and cost of maintaining pipelines.

Fury is a multi-language serialization framework. It features cross-language out-of-band serialization, a highly-extensible JIT framework, multiple binary protocols, and more. New protocols can be easily added to Fury and the optimization for one protocol can be reused for another protocol.

This article presents an in-depth analysis on why Meta made Llama 2's weights commercially available. Llama 2's license provides hints to Meta's goals - it prevents Llama Materials from being used to improve other models, competitors with more than 700 million monthly active users must request a license from Meta, and there are no trademark licenses. This suggests Llama 2 was open-sourced to reduce competitor moats, to attract more users and developers, and as a marketing tool.

Hey everyone, TLDR is looking for a resourceful, tech-savvy senior sponsorship sales director to join our fully remote team of six and work with our head of sales Danny Miller (who previously started the ads business at StackOverflow) to lead the next phase of our revenue growth.

We are growing at a rate of 10% month over month and we're looking for someone who is highly entrepreneurial and thrives in a fast-paced environment.

Total compensation including commission will be between $150,000 - $200,000.

If you know someone who would be a good fit, please refer them to us and we will send you $1000 if we hire them.

To apply, please send your resume or LinkedIn profile to jobs@tldr.tech along with a few sentences on why you would be a good fit.

A paper published by researchers from Stanford University and the University of California shows that GPT-4's outputs have declined in quality over the past few months. While some experts aren't convinced by the paper, the lack of certainty around whether GPT-4's quality has declined points to a larger issue with how OpenAI handles its model releases. Popular theories about GPT-4's decline in performance include OpenAI reducing computational overhead, unintended effects from fine-tuning, and OpenAI reducing its abilities so more people will pay GitHub Copilot. OpenAI has consistently declined any claims that its models have decreased in capability.

There are many companies currently developing or shipping products with large language models - this Hacker News thread looks at how developers are iterating, evaluating, and improving their AI apps.

The ActivityPub community is trying to figure out how to respond to Meta joining the decentralized movement.

There have been thousands of open-source hardware designs released over the 15 or so years, but more and more open-source hardware companies are becoming closed-source, delaying the release of files/source code, or requiring NDAs to obtain software.

Tesla's Dojo supercomputer, which Elon Musk claims will be capable of an exaflop floating-point operations per second, uses chips designed by Tesla.

Rewrites can be expensive and slow, there will be a lot of doublework during the rewrite, and the team will continue to find bugs and missed use cases after the rewrite ships.

The Basic plan will no longer be available for new or rejoining members, but existing members will be able to remain on the plan until they change plans or cancel their accounts.