The Daily Index — Friday, August 11, 2023 · TLDR Index

Across today’s newsletters

Today’s Index

TLDR AI

CHATGPT EXPANDS ITS ‘CUSTOM INSTRUCTIONS’ FEATURE TO FREE USERS

OpenAI announced that it’s expanding custom instructions — a way to give users more control over how ChatGPT responds — to all users, including those on the free tier of the service.

techcrunch.com · 1 min

TLDR Web Dev

HOW QUORA SCALED MYSQL TO 100K+ QUERIES PER SECOND

Quora has over 300 million monthly active users with tens of thousands of questions posted per day. To serve this traffic, it makes heavy use of MySQL. Quora has a sharded configuration that stores tens of terabytes and can scale to serve hundreds of thousands of queries per second. This article goes over how Quora handles such a heavy database load using MySQL with a focus on scaling read volume, write volume, and data size.

blog.quastor.org · 9 min

TLDR Crypto

BITSTAMP TO SUSPEND TRADING OF MAJOR ALTCOINS FOR US USERS

Bitstamp is suspending trading of Axie Infinity, Chiliz, Decentraland, Polygon, Near, Sandbox, and Solana for US customers, with all seven tokens being named by the SEC as securities. Users will still be able to hold the seven tokens in their accounts and withdraw them at any time. The SEC’s allegations against the tokens have not yet been proven in a court of law. It currently alleges that 68 cryptocurrencies are securities.

cointelegraph.com · 3 min

TLDR Founders

SURPRISING DATA POINTS ABOUT THE VENTURE CAPITAL MARKET

Tomasz slices-and-dices his thoughts on the Carta report on funding outlook based on data from the last few quarters. It contains some good news and some bad news. Must read for sure.

tomtunguz.com · 2 min

TLDR AI

NEWS OUTLETS DEMAND NEW RULES FOR AI TRAINING DATA

Several media organizations have called for rules protecting copyright in data used to train generative AI models. They have signed an open letter urging lawmakers around the globe to consider regulations that require transparency in training datasets and the consent of rights holders before using data for training. The letter also requests allowing media companies to negotiate with AI model operators and identify AI-generated content, and mandates that AI companies eliminate bias and misinformation in their services.

theverge.com · 14 min

TLDR AI

ENHANCING LLAMA-2 MODEL'S PERFORMANCE THROUGH IN-CONTEXT LEARNING

This paper studied in-context learning with the pretrained Llama-2 language model by retrieving 9 demonstration examples during chat-style prompts. The method, which didn't alter the model's weights, resulted in a 7x win-rate increase compared to the text-davinci-003 model, making it on par with models that underwent alignment fine-tuning.

arxiv.org · 8 min

TLDR AI

IMPROVING 3D OBJECT DETECTION USING DETR

Current 3D object detectors often misinterpret data, causing them to focus on distant points. The authors of this paper have developed a new method, 3D Vertex Relative Position Encoding, that directs the detector's attention to points close to the target objects.

arxiv.org · 14 min

TLDR AI

STRATEGY

Traditional image-text matching models often mistakenly use highly similar but not matching samples, known as false negatives, which can misguide the learning process. This paper introduces the False Negative Elimination strategy, a novel method that carefully selects negatives by assessing their probability of being false negatives.

arxiv.org · 15 min

TLDR AI

PINECONE’S JOURNEY FROM SEED TO 10,000 SIGN-UPS PER DAY

We’ve posted a lot about product-led growth (PLG) lately. This is another great example of a company nailing PLG. The key takeaways include being really thoughtful about who you’re building the product for, focusing on a defined persona, and then making their pathway to try your product as frictionless as possible. The traditional B2B marketing playbook didn’t land for Pinecone’s product.

kylepoyar.substack.com · 9 min