Search the archive

“EVERYTHING THAT MOVES WILL BE ROBOTIC”: NVIDIA CEO JENSEN HUANG SAYS ROBOTS AND SELF-DRIVING CARS ARE JUST AROUND THE CORNER

Jensen Huang believes that Nvidia will have a big role to play in a future where we are surrounded by robots.

MICROSOFT BRINGS A DEEPSEEK MODEL TO ITS CLOUD

DeepSeek R1 is now available on Microsoft's Azure AI Foundry service.

MEET ALIBABA'S QWEN 2.5, AN AI MODEL CLAIMING TO BEAT BOTH DEEPSEEK AND OPENAI'S CHATGPT

Alibaba's Qwen 2.5-Max reportedly outperforms GPT-4o, DeepSeek-V3, and Llama-3.1-405B.

SOFTBANK IN TALKS TO INVEST AS MUCH AS $25B IN OPENAI, REPORT SAYS

SoftBank may invest up to $25 billion in OpenAI as part of a broader partnership that could see it spend more than $40 billion on AI initiatives.

Read more about Radar →

APPLE AND SPACEX LINK UP TO SUPPORT STARLINK SATELLITE NETWORK ON IPHONES

Apple has been working with SpaceX and T-Mobile US to add support for the Starlink network in its latest iPhone update. T-Mobile this week enabled a small number of iPhones as part of its beta test of the Starlink service. Users in the program have access to a new toggle switch in their iPhone cellular data settings to manage the satellite feature. T-Mobile is aiming to support the vast majority of modern smartphones when the service fully launches.

DEEPSEEK CHIEF'S JOURNEY FROM MATH GEEK TO GLOBAL DISRUPTOR

China's DeepSeek is led by Liang Wenfeng, a math geek who caught the investing bug. Liang's hedge fund, High-Flyer, manages around $8 billion, making it one of China's largest quantitative funds. He has been described as a hands-on leader who sometimes sleeps in the office with colleagues as they rush to get projects done. Liang keeps a low profile and was surprised to see DeepSeek turn into a frenzy overnight. DeepSeek's sudden popularity caught the company off guard, making Liang work with his team to address the demand before taking a break for the Lunar New Year holiday in China.

BOOM'S XB-1 BECOMES FIRST CIVIL AIRCRAFT TO GO SUPERSONIC

Boom Superconic's XB-1 demonstrator plane went supersonic in the skies over California's Mojave Desert on its 12th test flight. It cleared Mach 1 and stayed supersonic for around four minutes, reaching Mach 1.1. This is the first time a supersonic engine has come from something other than a nation-state. Boom already has customers waiting for its technology when it matures. The company is doing some work with the US Air Force, but it remains focused on making supersonic passenger travel a reality again.

NEW GLOWING MOLECULE, INVENTED BY AI, WOULD HAVE TAKEN 500 MILLION YEARS TO EVOLVE IN NATURE, SCIENTISTS SAY

Scientists used an AI model to simulate 500 million years of molecular evolution, which resulted in the code for a previously unknown glowing protein similar to those found in jellyfish and corals. While the simulated protein only exists in computer code, the model used to create it may help in the development of new medicines. The model, called ESM3, doesn't design proteins within the usual constraints of evolution - it is a problem-solver that fills in gaps of incomplete protein code. ESM3 learns fundamental biology and then creates functional proteins outside the space explored by evolution.

COMPLETE HARDWARE + SOFTWARE SETUP FOR RUNNING DEEPSEEK-R1 LOCALLY

This thread details a $6,000 setup for running the full DeepSeek-R1 model locally. The setup requires two CPUs and 768GB of RAM across 24 RAM channels. It doesn't include a GPU, though a GPU can be added for faster generation speed. The build described in the thread can generate six to eight tokens per second, depending on the specific CPU and RAM speed purchased. A video of the build running inference is available in the thread.

INTRODUCING CODENAME GOOSE

Goose is an open source AI agent built to automate tasks. It can be powered by a range of large language models and features a user-friendly desktop interface and command-line interface. Goose is designed to enhance productivity and workflows by taking instructions and then doing the work for users. Its first use cases are engineering focused, but the community has been exploring other non-engineering use cases.

DEEPSEEK PANIC AT THE APP STORE

DeepSeek's app shot to #1 on the US App Store recently and is getting around 300,000 global daily downloads. It may be the first non-GPT-based assistant to get mainstream usage in the US. Released on January 11, many users see it as a free version of OpenAI's premium models. This article looks at DeepSeek's rise, why its app is currently topping the charts, DeepSeek's business model, the company's impact on the stock market, and much more. DeepSeek's open source software means that anyone can run the technology on their own computer, so it is unlikely that the US government will be able to block it in any reasonable way.

INSIDE CHINA, DEEPSEEK PROVIDES A NATIONAL MIC DROP MOMENT

The news that a Chinese startup could build artificial intelligence models as good as OpenAI's at a fraction of the cost tanked the stock market and sent Silicon Valley into a panic. DeepSeek's achievement was a testament to China's ability to innovate, especially when faced with efforts to limit the country's access to the most advanced technologies. The company's rise may be putting it in an awkward position - China has been cracking down on the tech sector in recent years as it is wary of tech companies' growing influence. Investors are hoping the government will refrain from interfering with DeepSeek.

APPLE CHIPS CAN BE HACKED TO LEAK SECRETS FROM GMAIL, ICLOUD, AND MORE

Two vulnerabilities that open up Macs, iPhones, and iPads to side channel attacks have been discovered in later generations of Apple A- and M-series chip sets.

DISCOVERY CODING

Discovery coding is when developers try to understand a problem by writing code first rather than attempting to design processes or thinking beforehand.

ASK HN: WOULD YOU STILL CHOOSE RUBY ON RAILS FOR A STARTUP IN 2025? (HACKER NEWS THREAD)

'Boring' technologies that have been around for a long time and battle tested are great for building apps.

WHY SORTING IS HARDER THAN IT SEEMS

Sorting in JavaScript requires a suitable comparator - mistakes virtually guarantee that some arrays will sort incorrectly.

SEVEN THINGS I KNOW AFTER 25 YEARS OF DEVELOPMENT

A loose transcript of a talk by Victor Shepelev, principal engineer at Hubstaff and a Ukrainian actively serving the Armed Forces, presented at the EuRuKo conference in September 2024.

AUTOMATTIC AND OTHERS BACK OPENVIBE, AN APP THAT'S UNIFYING THE OPEN SOCIAL WEB

Openvibe is an app that integrates multiple open social networks, including Bluesky, Mastodon, Nostr, and Threads.

DEEPSEEK FAQ

DeepSeek's R1 model, which is similar to OpenAI's o1, caused a meltdown in the tech industry over the weekend. Many of the revelations that contributed to the meltdown were included in DeepSeek's announcement over Christmas and most of the breakthroughs in V3 were actually revealed with the release of the V2 model last January. DeepSeek-V2 introduced two important breakthroughs, DeepSeekMoE and DeepSeekMLA - the implications of these breakthroughs only became apparent with V3, which added a new approach to load balancing and multi-token prediction in training. The cost of training the model amounted to only $5.576 million, but that excludes the costs associated with prior research, experiments, and data. Many of the decisions DeepSeek made in training its model were due to the constraints of the chip ban - if DeepSeek had access to more powerful chips, it probably would have used a different method.

IPHONE SE 4 APPEARS IN NEW PHOTOS AND VIDEO, NOTCH AND ALL

A leaker shared a short video of the new iPhone SE over the weekend. Unlike previous models, the device appears to only have a single rear camera. It is the first iPhone SE model to feature a USB-C port. The device uses Apple's older notched display rather than the Dynamic Island design. Rumors point to a launch around March or April. The video is available in the article.

HUGE STUDY FINDS CONSTELLATION OF HEALTH BENEFITS FOR OZEMPIC BEYOND WEIGHT LOSS

A new study that tracked millions of diabetes patient outcomes over a period of 3.5 years found that those on GLP-1 medications saw significant improvements across a staggering range of health concerns, far beyond anything clearly linked to weight or blood sugar. The patients experienced improvements in addiction and psychotic disorders and the risk of Alzheimer's was decreased. GLP-1RA drugs act on receptors that are expressed in brain areas involved in impulse control, reward, and addiction, which may potentially explain their effectiveness in curbing appetite and addiction disorders. While the findings are positive, the researchers found some issues that seemed to be exacerbated by taking GLP-1s, including an increase in risk for arthritis and pancreatitis.

WITH SUCCESSFUL NEW GLENN FLIGHT, BLUE ORIGIN MAY FINALLY BE TURNING THE CORNER

Blue Origin is finally a real rocket company with the successful launch of New Glenn. The company's next steps will be to get better at building engines and rockets and flying New Glenn regularly. Its current priorities are rate manufacturing and driving urgency around building the 'machine that builds the machine'. Blue Origin is the world's largest single-donor nonprofit, so it has zero incentive to operate like SpaceX.

like. Get the whitepaper

MICROSOFT BUILDS OPEN SOURCE DOCUMENT DATABASE ON POSTGRESQL, SUGGESTS FERRETDB AS FRONT END

Developers looking to build something new on a document database now have more alternatives to choose from. Microsoft has launched a document database platform constructed on a relational PostgreSQL back end. It has suggested that users kick things off by implementing a FerretDB solution on the platform to get started. Microsoft has been increasingly accepting of the open source world. The joint development between Microsoft, PostgreSQL, and FerretDB is a threat to source-available MongoDB and Couchbase.

MY FAILED ATTEMPT TO SHRINK ALL NPM PACKAGES BY 5%

Developer Evan Hahn proposed an idea in 2022 that could decrease the size of all newly-published npm packages by about 5%. Despite it being completely backwards compatible, improving performance, and reducing storage costs, the npm maintainers rejected the proposal. Hahn had thought the idea was great until he received feedback - implementing his idea would be difficult and would make publishing much slower. The changes would also not retroactively apply to existing packages. After receiving feedback and thinking about the tradeoffs, Hahn felt that the idea wasn't worth it and closed the proposal.

META IS REPORTEDLY SCRAMBLING ‘WAR ROOMS' OF ENGINEERS TO FIGURE OUT HOW DEEPSEEK'S AI IS BEATING EVERYONE ELSE AT A FRACTION OF THE PRICE

Meta's Mark Zuckerberg responded to the release of DeepSeek R1 by assembling four war rooms of engineers to figure out how the Chinese hedge fund built the AI. DeepSeek's model may potentially outperform the next version of Meta's Llama AI due to be released in early 2025. Meta is working on deciphering how DeepSeek lowered the cost of training and running its model, finding out what data the Chinese lab used to train its model, and how Meta can restructure its models based on the attributes of the DeepSeek models. Zuckerberg recently announced that Meta would spend as much as $65 billion on AI-related projects in the coming year, just days after OpenAI, SoftBank, Oracle, and others announced a $500 billion White-house backed AI infrastructure project.

NVIDIA SHEDS ALMOST $600 BILLION IN MARKET CAP, BIGGEST ONE-DAY LOSS IN U.S. HISTORY

Nvidia experienced the biggest market cap drop for any company on a single day in US history on Monday when it lost close to $600 billion in market cap. Its stock price plummeted 17% and caused a 3.1% slide in the tech-heavy NASDAQ. The sell-off was sparked by DeepSeek presenting increased competition in the global AI industry. The Chinese AI lab released a free and open source large language model that only took two months and less than $6 million to build using Nvidia H800s. The release has caused investors to rethink spending on compute.

NEW PEBBLE SMARTWATCH IS COMING AS GOOGLE OPENS SOURCE CODE

Google has released the source code for Pebble OS to help the volunteer efforts that have kept Pebble smartwatches alive and running for the past several years.

OPENHAYSTACK

OpenHaystack allows users to create their own tracking tags for Apple's Find My network.

VIRAL AI COMPANY DEEPSEEK RELEASES NEW IMAGE MODEL FAMILY

DeepSeek's new multimodal AI models reportedly outperform OpenAI's DALL-E 3.

GROK 3 SEEMINGLY WENT LIVE FOR SOME USERS

Several users on X briefly gained access to Grok 3 over the weekend - users managed to get it to respond to a few questions before their access was revoked.

SHUNPO

Shunpo is a bash tool that makes directory navigation in terminal faster by providing a system for managing bookmarks and jumping to directories with just a few keystrokes.

DEEPSEEK V3 AND THE ACTUAL COST OF TRAINING FRONTIER AI MODELS

DeepSeek shows that a lot of the modern AI pipeline is not magic and that cutting-edge AI is not restricted to a select cohort of the San Francisco in-group.

DEEPSEEK GETS SILICON VALLEY TALKING

Many in the tech industry have been praising DeepSeek over the recent release of its reasoning model, R1. R1 seemingly matches or beats OpenAI's o1 model on certain benchmarks. It only cost $5.6 million to train, much less than the hundreds of millions of dollars it has taken leading US companies to train their models. The company achieved all this while facing US sanctions that prohibit the sale of advanced chips to Chinese companies. DeepSeek's AI assistant is now the top free app in the Apple App Store.

ORACLE AND MICROSOFT ARE REPORTEDLY IN TALKS TO TAKE OVER TIKTOK

Oracle, Microsoft, and a group of investors are in talks to take over TikTok's global operations. The deal will allow ByteDance to keep a minority stake in TikTok while the app's algorithm, data collection, and software updates will be overseen by Oracle. Oracle's server network already provides the bulk of TikTok's backbone. President Trump earlier issued an executive order giving TikTok and others a 75-day reprieve from any enforcement action related to the US' requirement that ByteDance divest from TikTok.

WHAT DID DEEPSEEK FIGURE OUT ABOUT REASONING WITH DEEPSEEK-R1?

Chinese AI lab DeepSeek recently released a reasoning model trained on a GPU cluster a fraction the size of any of the big western AI labs that is better than the current best reasoning models. Reasoning models are able to 'think' about questions, which results in better answers. DeepSeek used a reinforcement learning approach, as opposed to a fine-tuning approach, to train its reasoning model, which resulted in cost and quality benefits. This approach can only really train the model on coding and mathematics, but this might not matter as superintelligence in coding/mathematics might transfer to other domains - this has yet to be demonstrated in practice.

WHAT HAPPENS WHEN WE CAN'T JUST BUILD BIGGER AI DATACENTERS ANYMORE?

AI's continued growth may ultimately hinge on a new kind of supercomputer that spans entire countries. This will involve stitching together existing datacenters to distribute the workload. The infrastructure required to stitch datacenters together already exists, but it is tuned more for disaster recovery and high-availability than large-scale AI training. Research is already underway to improve the technology, which will help address power challenges by allowing datacenters in different regions to work as one.

products get noticed. Learn more about advertising with TLDR.

INTRODUCING SCOPEDB: MANAGE DATA IN PETABYTES FOR AN OBSERVABILITY PLATFORM

ScopeDB is a database built on top of S3 that unleashes the power of cloud elasticity and workload isolation. It achieves ten times cost-efficiency and performance improvement compared to databases built with shared-nothing architecture. ScopeDB servers use a single stateless binary, so there is no need to manage local disks, data sharding, or complex leader action. A demo playground to test out the features of ScopeDB is available.

LAB

Lab allows developers to experiment with code instantly without the overhead of creating files and folders. Users can just type 'lab' with any extension (eg. lab py) and start coding. It automatically organizes experiments with smart names in a folder and cleans them up after seven days to keep things tidy. A demo video is available.

THE SECRET SAUCE OF CHINESE SOCIAL MEDIA APPS

In the days leading up to the deadline for TikTok to be banned in the US, users flocked to other Chinese apps such as Xiaohongshu and Lemon8. These apps have many of the same features that made TikTok so appealing: an infinite scroll of hyper-targeted recommendations and interactions with friendly strangers. They blend entertainment and networking with sales to monetize their addictive algorithms. This strategy has been very profitable, making China the global leader in social commerce.

HERE ARE APPLE'S TWO MAIN PRIORITIES FOR AI THIS YEAR, PER LEAKED MEMO

Apple has two main internal priorities for its AI efforts this year: revamp Siri and improve its AI models. Siri is undergoing a big transformation into 'LLM Siri', which will be shipping in the spring of 2026, possibly in iOS 19.4. Apple's AI models have been criticized for regularly including errors, so there is a clear need to improve them. The company has temporarily disabled summaries in iOS 18.3 for select app categories presumably until the backing AI model can be brought up to par.