Transformer – Spreadsheet

Tom Yeh, the creator of AI by Hand, has developed a tool using Google Sheets that allows users to create their own AI by Hand exercises with custom numbers and solutions. The goal of this project is to maximize reach and broaden access to AI education, and Tom is seeking feedback on the tool, which can be accessed at https://by-hand.ai/sp/tfmr.

Read more

Gumroad CEO is no longer hiring junior/mid-level software engineers due to AI

Sahil Lavingia, founder of Gumroad, has announced that his company is no longer hiring junior or mid-level software engineers due to advancements in AI technology. He predicts that AI will be able to write 100% of the code for his company's projects, including Gumroad, by the end of 2026, and is already using AI to generate specs and code for smaller projects like Iffy and Shortest.

Read more

Elon Musk's DOGE is feeding sensitive federal data into AI to target cuts

Elon Musk's DOGE team has been using artificial intelligence to analyze sensitive federal data from the Education Department, including personally identifiable information and internal financial data, in an effort to identify potential spending cuts. The use of AI to target cuts is part of a broader plan to deploy the technology across the federal government.

Read more

Elon Musk's DOGE is feeding sensitive federal data into AI to target cuts

There is no text to summarize, as the provided content appears to be a series of image links and placeholders, with no actual text or information to condense into a couple of sentences. The content seems to be a fragment of a webpage or article with missing or incomplete data.

Read more

Hand-Crafted, Machine-Made: How we make learning games with AI

Brilliant, an educational platform, aims to create a visceral learning experience where students develop intuition and understanding through interaction and experimentation, rather than just reading text explanations. To achieve this, they use AI to generate interactive puzzles and problems, handling the technical implementation, while human course authors focus on the creative direction and ensure the quality and correctness of the learning content.

Read more

Develop AI Agents for System Engineering in Factorio

Continuing advances in AI research are driving interest in deploying AI agents in complex systems, but current evaluation methods fail to assess crucial skills like adaptability and trade-off management. To address this, the paper proposes using automation-oriented sandbox games like Factorio to train and evaluate AI agents' system engineering abilities, enabling them to design and optimize complex projects.

ArXiv

Meta AI's latest research: improved LLM reasoning with Latent Tokens

Large Language Models (LLMs) trained on chain-of-thought data can excel at reasoning and planning, but processing the lengthy inputs is computationally expensive. This work proposes a hybrid representation using latent discrete tokens to abstract away initial reasoning steps, reducing input length and achieving better performance than baseline methods in various benchmarks.

ArXiv

Optimizing LLM Persuasion with Personalization and Fabricated Statistics

Large Language Models (LLMs) are becoming increasingly persuasive, particularly when using a mixed strategy of personalized arguments and fabricated statistics in an interactive debate setting, which had a 51% chance of persuading participants to change their opinion. This raises concerns about the potential for LLMs to enable large-scale disinformation campaigns, as they can outperform static human-written arguments and may be used to spread misinformation at a low cost.

ArXiv

SmolLM2: When Smol Goes Big – Data-Centric Training of a Small Language Model

The development of SmolLM2, a 1.7 billion parameter language model, is documented, which achieves strong performance through a multi-stage training process on a large dataset of around 11 trillion tokens. SmolLM2 outperforms other recent small language models, and to facilitate future research, both the model and the datasets used in its development are being released.

ArXiv

Pre-Trained Large Language Models Use Fourier Features for Addition (2024)

Pre-trained large language models add numbers using Fourier features, which are dimensions in the hidden state that represent numbers via a set of features sparse in the frequency domain. The model's MLP and attention layers work together to compute the answer, with MLP layers handling magnitude and attention layers handling modular addition, and pre-training is crucial for this mechanism to work accurately.

ArXiv

Show HN: An API that takes a URL and returns a file with browser screenshots

ScrapeServ is a simple API server that runs on your machine, takes a URL as input, and returns the website data as a file along with screenshots of the site. The server uses Playwright to launch a Firefox browser context for each job, handles tasks in a queue, and provides features like automatic 302 redirect handling and configurable memory allocation, but is resource-intensive and requires Docker and docker compose to be installed.

Repo

AI by Hand Exercises in Excel

This text lists various AI and deep learning concepts, including basic and advanced topics, that are being implemented or explained using Excel, with accompanying exercises and lectures. The topics range from simple functions like Softmax and LeakyReLU to more complex models like Transformers, Residual Networks, and Autoencoders, with additional topics like GANs and VAEs coming soon.

Repo

Show HN: Former – Open-source Cursor for SQL

Former is a desktop and web SQL editor that utilizes AI assistance to improve the SQL writing experience, offering a more efficient alternative to copying and pasting database context and SQL into external AI tools. It can be used through a cloud-hosted instance with a free tier and trial period, or self-hosted for free by compiling the Electron app and setting up a NextJS server and Supabase instance.

Repo

Show HN: Inferable – Build Reliable AI Agents with Durable Execution

Inferable is a platform for building production-ready AI Agents, allowing users to register tools, define agents that can use these tools to perform multi-step reasoning and take actions, and create durable workflows that compose agent interactions with "workflow as code". The platform provides a range of features, including tool orchestration, agent reasoning, and workflow management, with support for multiple programming languages and a minimal adoption curve.

Repo

ArchGW: Open-source, AI-native (edge and LLM) proxy for prompt traffic

Arch is an intelligent proxy designed for agentic applications, helping to protect, observe, and build tasks by connecting existing APIs, with features such as task routing, prompt guardrails, and observability. It is engineered with purpose-built large language models (LLMs) to handle critical tasks, including detecting jailbreak attempts and managing prompt-based interactions, and can be used to build fast, secure, and personalized AI applications.

Repo