TLDR 2024-02-01

Google Bard Advanced 🧠, ChatGPT interview cheating πŸ‘¨β€πŸ’», Crypto + AI applications πŸͺ™

πŸ“±
Big Tech & Startups

Google says you'll need a subscription to use Bard Advanced (2 minute read)

Bard Advanced, which will be powered by the Gemini Ultra architecture, will be available to users as a subscription. Google has yet to reveal how much the subscription will cost or when it will be released. It will likely be available to early testers before being widely rolled out. Alphabet recently revealed that the company earned more than $15 billion in 2023 from subscription revenue.

Mistral CEO confirms β€˜leak’ of new open source AI model nearing GPT-4 performance (4 minute read)

Files for a quantized version of Mistral were leaked on HuggingFace on around January 28. Quantization is a technique that makes it possible to run certain AI models on less powerful computers and chips. The model surpasses every other LLM in the world except GPT-4 on EQ-Bench. Mistral appears to still be working on the model, so the final version may be even more capable.
πŸš€
Science & Futuristic Technology

Add bacteria to the list of things that can run Doom (1 minute read)

An MIT biotechnology PhD student has created a rudimentary screen capable of displaying Doom using a simulation of E. coli bacteria. It takes roughly 70 minutes to display a single frame of Doom on the bioluminescent bacteria. A full reset takes eight hours and 20 minutes. A video about the project is available in the article.

Reinventing the eel: first lab-grown eel meat revealed (3 minute read)

Forsea Foods in Israel is the first company to produce lab-grown freshwater eel meat. It aims to have its product on sale in about two years. The company's goal is to match the price of wild-caught eel. Forsea Foods' strategy is to target species at risk of extinction that also command high prices in restaurants and shops. It has raised $5.2 million in investments, with more expected to be announced soon.
πŸ’»
Programming, Design & Data Science

WhisperKit (GitHub Repo)

WhisperKit integrates OpenAI's Whisper speech recognition model with Apple's CoreML framework for efficient local inference on Apple devices with Swift. It can be integrated with Swift projects using the Swift Package Manager. WhisperKit automatically downloads recommended models, but it can also create and deploy fine-tuned versions. A link to a demo app is available in the repository.

The Performance Inequality Gap, 2024 (35 minute read)

Performance is a deep and nuanced domain. There's a lot that can go wrong and how sites manage resources after-load can have huge impacts on perceived performance. Global baselines matter as many teams have low performance management maturity and many popular frameworks fail to ward against catastrophic results. This article looks at current web technologies to provide an up-to-date understanding for working web developers. It looks at what's changed since last year and what technologies new projects should consider.
🎁
Miscellaneous

How hard is it to cheat with ChatGPT in technical interviews? We ran an experiment (21 minute read)

ChatGPT can assist people during their technical interviews. An experiment on how much the tool can help and how easy it is to cheat uncovered an urgent need for companies to immediately change the types of interview questions they're asking. Interviewees in the study were able to trick professional interviewers and pass interviews without the interviewers suspecting them of cheating. Not only were interviewees not caught cheating, the interviewers also rated their confidence in their hiring decisions as high.

Why 2023 Was AI Video’s Breakout Year, and What to Expect in 2024 (11 minute read)

There were no public text-to-video models available at the beginning of 2023, but now there are dozens in active use with millions of users worldwide. While many of these products are still limited, the progress in the past year suggests that the industry is about to massively transform. This article looks at the biggest developments so far in AI video generation, companies to keep an eye on, and the remaining underlying questions in the space.
⚑
Quick Links

Lies, damned lies, and benchmarks (6 minute read)

Benchmarks are useful tools, but they're not the best indicators of real-world utility due to their narrow scopes, overfitting, contamination, reproducibility issues, and lack of scope.

The promise and challenges of crypto + AI applications (25 minute read)

Crypto and AI are two major technology trends from the last decade - while the technologies can be combined, there aren't that many useful applications where the two fields intersect.

Microsoft LASERs away LLM inaccuracies (2 minute read)

Layer-Selective Rank Reduction (LASER) is a technique that involves replacing one weight matrix with an approximate smaller one, making it so the model relies less on it - the technique makes large language models more accurate.

Starlink's Laser System Is Beaming 42 Million GB of Data Per Day (3 minute read)

Starlink passes terabits per second of data every day across 9,000 lasers that can sustain a 100Gbps connection per link.

The first version of Infinite Craft, an endless crafting game, is out now! (1 minute read)

Infinite Craft is a game where you combine two elements together to create new elements, which you can then use to combine with other elements to create new elements, which you can then use to combine with other elements to...

The end of my childhood (30 minute read)

A discussion on change by Vitalik Buterin, one of Ethereum's co-founders.
Get the most important tech, science, & coding news in a free daily email. Read by +1,250,000 software engineers and tech workers.
Join 1,250,000 readers for