Search the archive

Read the documentation and see Paragraph code examples

4 SOFTWARE DESIGN PRINCIPLES I LEARNED THE HARD WAY

This author built a large-scale service and found certain principles reappearing throughout the implementation. It's useful to prioritize a single source of truth and minimize mutable state when building something from scratch. Developers should also make sure not to abstract things prematurely and not to overuse mocks when writing tests for their code.

DEBUGGING INFINITE LOOPS

In object-oriented programs, infinite loops can happen unknowingly. To understand and fix these issues, you should successively inline the methods on the stack until you have a self-referencing method. This gets to the root of the issue compared to just applying a code-level hack.

MY FIRST PROMO REJECTION, A BLESSING IN DISGUISE

This author was rejected for their first promotion at Google after two years in their role for not displaying enough impact. From that rejection, they learned the importance of taking initiative and strategically selecting projects that demonstrated impact.

THE TRAP OF OVER-ENGINEERING AND OVER-DESIGN

Developers often over-engineer and over-design software by abstracting things excessively and focusing on hypothetical future scalability. Instead, it's better to prioritize core functionality first, keep the scope reasonable, and have a “good” solution over a “perfect” one.

DROP EVERYTHING AND REVIEW

Software development has multiple feedback loops: the inner loop (immediate feedback from tools), the middle loop (code reviews), and the outer loop (deployment and system-level testing). Code review often gets delayed, unnecessarily inflating the middle loop which leads to slower deployment. This author advocates for code reviews to be the top priority for teams after any urgent production issues so that teams can keep shipping at a smooth pace.

FRANKEN/UI

Franken UI is a collection of copy-pastable components that are HTML-first and framework agnostic.

FIRECRAWL

Crawl and convert any website into LLM-ready Markdown. It can be self-hosted and comes with Python and Node SDKs.

DEVELOPING RAPIDLY WITH GENERATIVE AI

Discord builds products with generative AI quickly by identifying suitable use cases early and then prototyping with off-the-shelf LLMs. It evaluates prompts deeply before deploying at scale. Discord leans towards using off-the-shelf LLMs before thinking about using self-hosted ones because it saves engineering resources. Often, if the target problem can't be solved with GPT-4, then it probably can't be solved yet with generative AI at all.

HOW SHORT CAN GIT ABBREVIATE?

The default 7-character length of Git hashes can lead to collisions in large repos like the Linux kernel. This article analyzes the distribution of abbreviation lengths, finding that while most commits can use short abbreviations, a large number require longer ones (up to 11 currently). It recommends using at least 12 characters for large repositories and referencing full Git hashes if possible.

ON GENERATING IDEAS - LEADERSHIP & WORK

People should generate ideas independently and quietly first before bringing them to group settings.

THE DANGERS OF SINGLE LINE REGULAR EXPRESSIONS

This article describes a vulnerability within an application where a poorly constructed regular expression allowed for multiline input, allowing arbitrary Ruby code execution.

GOOD IDEAS IN COMPUTER SCIENCE

A list of Computer Science victories, like arrays, hashmaps, processes, version control, GUIs, and more.

EFFICIENT ENGINEERING TEAMS (X POST)

Efficient engineering teams often have a Directly Responsible Individual for each part of their systems.

✅ 3 minutes to set up

✅ Get your data off 180+ data brokers and people-search sites

✅ 30-day money back guarantee.

TLDR55 →

TESLA SPENDS WEEKEND CUTTING PRICES OF CARS AND FSD SOFTWARE

Tesla has cut the prices of its models across China, Europe, and the US. Full Self-Driving is now $8,000 in the US. In China, the revamped Model 3 is now 241,900 yuan (~$32,000), down from 245,900 yuan previously and the Model Y was discounted to 249,900 yuan (~$34,500) from 263,900 yuan. The cheapest Model Y in the US is now $42,990. Tesla is facing several challenges, including a recall of almost 3,900 Cybertrucks, sales drops, interrupted production schedules, and more.

GOOGLE'S TOP SECRET CAMERA LAB IS LIKE AN IKEA FOR PIXEL TESTING

Google's Real World Testing Lab is an elaborate environment where Google tests the cameras in its Pixel phones. The lab looks more like a cluster of Ikea displays than a testing room. It allows Google to test its devices in realistic environments that simulate where customers will actually use them. This article takes a look inside the lab and the kind of testing that Google does to make sure its devices capture moments in just the right way.

THE GMO TOOTH MICROBE THAT IS SUPPOSED TO PREVENT CAVITIES

Lumina is a one-time treatment that can present cavities. It can be applied either at home or by a dentist. The treatment, which can be pre-ordered for $250 (before taxes and shipping fees), is expected to be released by mid-June. Marketed as a cosmetic product, the treatment has yet to successfully move through human clinical trials, leading to safety and ethical concerns.

HOW MANY MORE OZEMPIC-LIKE DRUGS ARE IN THE PIPELINE — HERE'S WHAT RESEARCHERS ARE WORKING ON

Semaglutide has proved so successful that its manufacturer is unable to keep up with demand. There is currently a tsunami of clinical trials for new drugs that also operate on the gut-brain axis and re-create the action of other intestinal and pancreatic hormones. This article discusses the state of the industry and new developments on the horizon. These drugs remain inaccessible for many people due to cost.

HOW DOES CHATGPT WORK? AS EXPLAINED BY THE CHATGPT TEAM

When users ask ChatGPT a question, OpenAI chunks the question into tokens, creates embeddings, multiplies the embeddings by model weights, and then samples a prediction. This article explains this process in detail. It covers pretraining and inference as well as the system's limitations. Links to other resources that also explain how ChatGPT works are available in the article.

THE POSTGRESQL COMMUNITY DEBATES ALTER SYSTEM

The PostgreSQL community recently resolved the question of whether to merge a brief patch adding a new configuration parameter. This article tells the story of the decision process, which involved one of the longest-running development discussions in recent PostgreSQL history. The success of the effort was due to the persistence of a small number of developers who saw it through months of opposition and 'helpful' implementation suggestions.

MEASURING PERSONAL GROWTH

Measuring things can often help to make life better. This article presents three metrics for measuring personal growth: rate of change, time to solve problems, and number of future options. Rate of change looks at how long it takes to become a new person. Time to solve problems looks at what big problems have been solved and what needs to be solved. Number of future options looks at options gained and lost.

SAMSUNG SHIFTS EXECUTIVES TO SIX-DAY WORKWEEKS TO ‘INJECT A SENSE OF CRISIS'

Samsung has introduced a company-wide six-day workweek policy for executives. The company had disappointing financial results last year and is now trying to tackle ongoing business uncertainty. The new workweek policy aims to inject a sense of crisis as part of an all-out effort to overcome its current challenges. Employees under the executive level are not expected to follow suit. Samsung is facing increased competition in the production of high-bandwidth memory chips alongside wider economic concerns like rising borrowing costs and oil prices and a rapid depreciation of Korean currency.

session

HAS LLAMA-3 JUST KILLED PROPRIETARY AI MODELS?

Benchmarks show that Llama-3 70B matches GPT-4 and Claude Opus in most tasks.

THE DANGERS OF “DECENTRALIZED” ID SYSTEMS

Decentralized ID systems cannot protect users against state surveillance and exclusion as seen in current government ID systems unless they allow users to participate without linking government ID, allow pseudonyms and throwaways, and keep regular login methods for people who prefer it.

LOOKING FOR AI USE CASES

New use cases will be things that are still automated one at a time, but that could not have been automated before, or that would have needed more software and capital to automate.

EUROPEAN REGULATORS WILL SOON APPROVE APPLE'S PLAN TO OPEN UP TAP-TO-PAY TO BANKS AND PAYMENT PROVIDERS

Europe's new rules mean that Apple will open up access to iPhone NFC chips and allow for non-Apple tap-to-pay payment options.

U.S.-CHINA INTERNET WAR INTENSIFIES AS HOUSE PASSES TIKTOK BAN

The US House has passed a bill, tied to a sweeping aid package for Israel and Ukraine, that may force ByteDance to sell TikTok if the Senate votes to also pass it in the coming days.

FOLLOW CURIOSITY

Curiosity knows more than you do about what's worth paying attention to.

* Explore the future of AI and the open web

* Learn how to mitigate the risks of AI tools * Get expert guidance on how AI is changing ecommerce * Discover why composability [4] is the foundation for AI transformation

See the full agenda—including guest speakers—and join for free

UPDATES FROM GOOGLE DEEPMIND ALIGNMENT RESEARCH

Following Anthropic, GDM has released some work from its alignment efforts. The main insightful piece here is using sparse autoencoders on Gemini Ultra. This is a substantial jump in interpretation size.

NVIDIA TO COLLABORATE WITH JAPAN ON THEIR CUTTING-EDGE ABCI-Q QUANTUM SUPERCOMPUTER

NVIDIA will power Japan's new quantum supercomputer, ABCI-Q, alongside Fujitsu, integrating 2,000 NVIDIA H100 AI GPUs and CUDA-Q platform for quantum-classical computing applications. The project aims to advance Japan's capabilities in quantum computing and AI. This collaboration is part of a broader technological partnership between NVIDIA and Japan.

BRAVE SEARCH IS ADOPTING AI TO ANSWER YOUR QUERIES

Brave Search has revamped its answer engine to include AI-synthesized responses worldwide, leveraging large language models and trusted data sources. The upgrade enables automatic or manual AI-enhanced answers to user queries. Brave acknowledges potential impacts on web publishers and plans to monitor and address the effects of AI-generated content on site traffic.

MOVING OBJECT SEGMENTATION

Segmenting objects across videos is challenging due to temporal consistency. This work presents the combination of a powerful image segmentation model and optical flow which combined gives compelling performance on this task.

YOUR LANGUAGE MODEL IS A Q-FUNCTION

A fairly technical RL paper that shows the theoretical underpinning of reward models and base models in language.

FINE-TUNE LLAMA 3 WITH ORPO

While this tune performs worse than Hermes, it is likely due to the small dataset used. Otherwise, this is a great walkthrough on how to tune these models to improve desired performance on certain tasks.

FINEWEB: 15T HIGH QUALITY WEB TOKENS (HUGGINGFACE HUB)

The newest Llama 3 models were trained on 15T tokens. This new dataset contains an extensively deduplicated corpus from common crawl and results in high quality models.

MISTRAL COMMON

Mistral has released its tokenizer utilities, chat templates, and more basic tools which it uses internally.

TRAINING-FREE CONTEXT EXTENSION

Researchers have developed a way to expand the context window of embedding models up to 32,000 tokens without additional training using strategies like position interpolation.