When people hear Leonardo da Vinci, the Mona Lisa usually appears first. That is understandable — it is one of the most recognized paintings in the world, displayed behind bullet-proof glass in the Louvre to crowds that number in the tens of thousands daily. But the painting is a narrow window into a much larger life. Leonardo spent more hours filling notebooks with observations and designs than he did holding a paintbrush. Read more
Computers do not make decisions in the human sense. They follow logical rules with perfect consistency and no understanding of why those rules exist. The interesting question is how that simple foundation — binary logic operating on electrical signals — scales up to the decision-making behavior we observe in real systems. Read more
People often ask why computers use 1s and 0s when we normally count with ten digits. The answer starts with the hardware — but the real reason runs deeper than transistors. Read more
A computer without an operating system would still be a computer. The processor would still execute instructions. Memory would still store values. But using it would require detailed knowledge of the specific hardware — the exact memory addresses where data should be placed, the precise sequence of instructions to communicate with a storage device, the mechanism for switching between different tasks. Early computers operated this way. Human operators loaded programs manually, assigned resources directly, and managed the hardware as closely as the machine itself. Read more
Microsoft announced on January 23, 2023 a new multi-year, multi-billion-dollar investment in OpenAI, reported as approximately $10 billion — the third major investment after $1 billion in 2019 and approximately $2 billion in 2021. The announcement came 54 days after ChatGPT’s November 30, 2022 public launch, during which ChatGPT reached 100 million monthly active users in 2 months — faster than any consumer internet product including TikTok (9 months) and Instagram (2.5 years) to the same milestone. The investment was structured partly in Azure cloud computing credits rather than fully in cash, deepening OpenAI’s existing exclusive reliance on Microsoft Azure for all training and inference compute. The financial structure included a “capped profit” arrangement: Microsoft and other investors receive returns up to a specific multiple of their investment, after which OpenAI’s nonprofit parent entity receives subsequent profits — a structure designed to allow investor returns while preserving OpenAI’s nonprofit mission orientation. Read more
LastPass disclosed on December 22, 2022 that an August 2022 security incident — in which an attacker had compromised a developer’s workstation and exfiltrated source code, technical documentation, and internal secrets — had been used as the first stage of a second, larger breach. In the second incident (September through October 2022), the attacker used credentials and keys stolen from a LastPass DevOps engineer’s home computer (compromised through a vulnerable third-party media player with an unpatched remote code execution vulnerability) to access LastPass’s cloud backup storage on Amazon S3. The attacker exfiltrated backups of LastPass customer vault data along with encryption keys and other infrastructure data. The disclosure came four months after the initial August 2022 breach notification, which had incorrectly characterized the damage as limited to source code and had not mentioned that customer vault data was at risk. The delayed and initially incomplete disclosure was widely criticized by security researchers and customers. Read more
Students and graduates face different kinds of problems, and that difference is larger than it first appears. Understanding it early can make the transition significantly less disorienting. Read more
A student can sit quietly through an entire lesson, nod at the right moments, and still learn very little. This is the central problem with passive education. Attention is not the same as understanding, and understanding is not the same as being able to use something. Read more
OpenAI released ChatGPT on November 30, 2022 as a free research preview powered by GPT-3.5-turbo, an optimized version of the GPT-3.5 model family trained using InstructGPT techniques developed by OpenAI researchers in 2022. The training pipeline combined supervised fine-tuning (human contractors writing example conversations demonstrating helpful assistant behavior), reward model training (human raters comparing pairs of model responses and indicating which was better), and reinforcement learning from human feedback (RLHF) using Proximal Policy Optimization (PPO) to update the language model’s weights toward higher-scoring responses. The resulting model was dramatically better than raw GPT-3 at following instructions, maintaining topic coherence across a conversation, declining inappropriate requests, and producing well-structured output — behaviors that emerged from the RLHF process rather than raw pretraining. The web interface (chat.openai.com) required only a free account and a web browser, making it the first large language model experience accessible to a mainstream audience without any technical setup. OpenAI had previously offered GPT-3 access through an API requiring developer registration and payment, which limited its reach to technically sophisticated users. ChatGPT eliminated that barrier: within five days of launch it had one million registered users, and within two months it reached 100 million monthly active users — the fastest product in history to reach that milestone, surpassing TikTok (nine months) and Instagram (two and a half years) by a wide margin. Read more
Elon Musk completed the acquisition of Twitter in October 2022, ending months of legal, financial, and public drama around one of the world’s most visible social platforms. The deal mattered not only because of the price or the personalities involved, but because Twitter had become a strange kind of public infrastructure: part newswire, part social network, part professional identity system, and part political arena. Read more
GitHub Copilot went into technical preview in June 2021. By October 2022, it has been available to the general public for several months, and the reports from developers who use it daily are specific enough to form a clearer picture of what it actually changes. Read more
NVIDIA unveiled the GeForce RTX 40 series at a September 20, 2022 event, with the RTX 4090 and two versions of RTX 4080 (16GB and 12GB) announced. The RTX 4090 launched October 12 at $1,599; the 16GB RTX 4080 launched November 16 at $1,199. The 12GB “4080” was withdrawn after public backlash over the naming and re-released in January 2023 as the RTX 4070 Ti at $799. All used the Ada Lovelace architecture on TSMC’s 4N process node (a customized version of 5nm). The RTX 4090 contained 76.3 billion transistors on the AD102 die, nearly double the RTX 3090 Ti’s 40.4 billion on Samsung 8N. Read more
Every year, lists of software development trends appear. Most of them include the same items at different stages of hype. In September 2022, I want to try something different: instead of simply listing what is popular, I want to say what I think is real, what is overstated, and what is worth investing in now. Read more
Stability AI released Stable Diffusion 1.4 publicly on August 22, 2022, making the model weights downloadable under the CreativeML Open RAIL-M license (allowing commercial use with behavioral restrictions, unlike prior generative image models locked behind APIs). The model had been developed by CompVis (Ludwig Maximilian University of Munich, led by Robin Rombach) with funding from Stability AI and collaboration from Runway ML, based on the “Latent Diffusion Models” paper (Rombach et al., arXiv December 2021, CVPR 2022 Best Paper). The training dataset was LAION-5B (5.85 billion image-text pairs scraped from the public web), the largest openly published training dataset at the time, curated by the LAION non-profit organization. The SD 1.x models fit in approximately 4 GB (fp32 weights) or 2 GB (fp16), runnable on consumer NVIDIA GPUs with 6–8 GB VRAM or Apple Silicon M1/M2 (via CoreML or MPS backends, slower but functional without a discrete GPU) — a stark contrast to DALL-E 2 (OpenAI, April 2022, API-only at ~$0.02/image) and Midjourney (March 2022, Discord bot, subscription, closed source). Read more
NASA publicly released the first full-color science images from the James Webb Space Telescope on July 12, 2022 — President Biden previewed one image (SMACS 0723, a gravitational lensing deep field) at the White House on July 11. JWST had launched December 25, 2021 on an Ariane 5 rocket, deployed its 18-segment primary mirror and sunshield over two weeks in orbit, and reached its operational position at the Sun-Earth Lagrange Point 2 (approximately 1.5 million kilometers from Earth in the direction away from the Sun) on January 24, 2022. Mirror alignment completed April 28, 2022, and commissioning of the four science instruments concluded in late June 2022 before the July release. The five inaugural images demonstrated capabilities across JWST’s instrument suite: the SMACS 0723 deep field showed thousands of galaxies as they existed 13 billion years ago through gravitational lensing; the Carina Nebula image revealed previously obscured star formation regions in infrared wavelengths that visible-light telescopes couldn’t penetrate; the WASP-96b spectrum showed the atmospheric transmission spectrum of an exoplanet with identifiable water vapor, carbon dioxide, and cloud signatures; the Southern Ring Nebula showed a dying star’s ejected shells in both NIRCam and MIRI imaging; and Stephan’s Quintet showed five interacting galaxies in an unprecedented wide-field infrared mosaic. Read more
If I were building a gaming PC on July 1, 2022, this is the machine I would actually want. I am not simply choosing the most expensive part in every category. The priority is gaming performance, then reliability, cooling and enough headroom to keep the system useful for years. Read more
Apple announced the M2 processor at WWDC on June 6, 2022, first shipping inside a redesigned 13-inch MacBook Air that launched June 17 at $1,199. M2 used TSMC’s second-generation 5nm process and kept the same broad heterogeneous architecture as M1 — four performance cores, four efficiency cores, and up to ten GPU cores — while improving each block. Read more